47
DO INVESTMENT-CASH FLOW SENSITIVITIES PROVIDE USEFUL MEASURES OF FINANCING CONSTRAINTS?* STEVEN N. KAPLAN AND LUIGI ZINGALES No. This paper investigates the relationship between nancing constraints and investment-cash ow sensitivities by analyzing the rms identi ed by Faz- zari, Hubbard, and Petersen as having unusually high investment-cash ow sen- sitivities. We nd that rms that appear less nancially constrained exhibit signi cantly greater sensitivities than rms that appear more nancially con- strained. We nd this pattern for the entire sample period, subperiods, and indi- vidual years. These results (and simple theoretical arguments) suggest that higher sensitivities cannot be interpreted as evidence that rms are more nan- cially constrained. These ndings call into question the interpretation of most previous research that uses this methodology. “Our nancial position is sound . . . Most of the company’s funds are generated by operations and these funds grew at an average annual rate of 29% [over the past 3 years]. Throughout the company’s history this self- nancing concept has not been a constraint on the com- pany’s growth. With recent growth restrained by depressed economic conditions, the company’s net cash position has grown substantially” [Hewlett-Packard 1982 Annual Report]. A large nance and macroeconomics literature studies the relation between corporate investment and cash ow to test for the presence and importance of nancing constraints. Beginning with “Financing Constraints and Corporate Investment” by Faz- zari, Hubbard, and Petersen [1988], (hereinafter FHP [1988]), these studies divide a sample of rms according to an a priori measure of nancing constraints and compare the investment- *Previous versions of this paper [Kaplan and Zingales 1995] circulated under the title “Do Financing Constraints Explain Why Investment Is Correlated with Cash Flow?” Benjamin Bridgman and Violet Law provided excellent research as- sistance. Comments from Charles Calomiris, John Cochrane, Zsuzsanna Fluck, Robert Gertner, David Gross, R. Glenn Hubbard, Bengt Holmstrom, Anil Kas- hyap, Owen Lamont, Stewart Myers, Walter Novaes, Bruce Petersen, Raghuram Rajan, Andrei Shleifer, Amy Sweeney, Sheridan Titman, Robert Vishny, and espe- cially David Scharfstein and Jeremy Stein (the referees) were very helpful. Semi- nar participants at Boston College, the CEPR Summer Symposium in Financial Markets in Gerzensee, the Federal Reserve Board, Harvard Business School, In- diana University, Massachusetts Institute of Technology, the NBER Summer In- stitute, the University of California at Los Angeles, the University of Chicago, the University of Southern California, the University of Texas, the University of Washington, and the Nobel Symposium on Law and Finance also provided useful comments. We also thank Bruce Petersen for providing a list of sample companies. This research has been supported by the Center For Research in Security Prices and by the Olin Foundation through grants to the Center for the Study of the Economy and the State. Address correspondence to Graduate School of Business, University of Chicago, 1101 East 58th Street, Chicago, IL 60637. q 1997 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology. The Quarterly Journal of Economics, February 1997.

do investment-cash flow sensitivities provide useful measures of financing constraints?

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

DO INVESTMENT-CASH FLOW SENSITIVITIES PROVIDEUSEFUL MEASURES OF FINANCING CONSTRAINTS

STEVEN N KAPLAN AND LUIGI ZINGALES

No This paper investigates the relationship between nancing constraintsand investment-cash ow sensitivities by analyzing the rms identied by Faz-zari Hubbard and Petersen as having unusually high investment-cash ow sen-sitivities We nd that rms that appear less nancially constrained exhibitsignicantly greater sensitivities than rms that appear more nancially con-strained We nd this pattern for the entire sample period subperiods and indi-vidual years These results (and simple theoretical arguments) suggest thathigher sensitivities cannot be interpreted as evidence that rms are more nan-cially constrained These ndings call into question the interpretation of mostprevious research that uses this methodology

ldquoOur nancial position is sound Most of the companyrsquos funds aregenerated by operations and these funds grew at an average annualrate of 29 [over the past 3 years] Throughout the companyrsquos historythis self-nancing concept has not been a constraint on the com-panyrsquos growth With recent growth restrained by depressed economicconditions the companyrsquos net cash position has grown substantiallyrdquo[Hewlett-Packard 1982 Annual Report]

A large nance and macroeconomics literature studies therelation between corporate investment and cash ow to test forthe presence and importance of nancing constraints Beginningwith ldquoFinancing Constraints and Corporate Investmentrdquo by Faz-zari Hubbard and Petersen [1988] (hereinafter FHP [1988])these studies divide a sample of rms according to an a priorimeasure of nancing constraints and compare the investment-

Previous versions of this paper [Kaplan and Zingales 1995] circulated underthe title ldquoDo Financing Constraints Explain Why Investment Is Correlated withCash Flowrdquo Benjamin Bridgman and Violet Law provided excellent research as-sistance Comments from Charles Calomiris John Cochrane Zsuzsanna FluckRobert Gertner David Gross R Glenn Hubbard Bengt Holmstrom Anil Kas-hyap Owen Lamont Stewart Myers Walter Novaes Bruce Petersen RaghuramRajan Andrei Shleifer Amy Sweeney Sheridan Titman Robert Vishny and espe-cially David Scharfstein and Jeremy Stein (the referees) were very helpful Semi-nar participants at Boston College the CEPR Summer Symposium in FinancialMarkets in Gerzensee the Federal Reserve Board Harvard Business School In-diana University Massachusetts Institute of Technology the NBER Summer In-stitute the University of California at Los Angeles the University of Chicagothe University of Southern California the University of Texas the University ofWashington and the Nobel Symposium on Law and Finance also provided usefulcomments We also thank Bruce Petersen for providing a list of sample companiesThis research has been supported by the Center For Research in Security Pricesand by the Olin Foundation through grants to the Center for the Study of theEconomy and the State Address correspondence to Graduate School of BusinessUniversity of Chicago 1101 East 58th Street Chicago IL 60637

q 1997 by the President and Fellows of Harvard College and the Massachusetts Instituteof TechnologyThe Quarterly Journal of Economics February 1997

cash ow sensitivities of the different subsamples The studiesinterpret a greater investment-cash ow sensitivity for rms con-sidered more likely to face a larger wedge between the internaland the external cost of funds as evidence that the rms are in-deed constrained This methodology has been widely applied toidentify rms that are more affected by nancing constraintsand institutions that are more likely to alleviate those con-straints For example Hoshi Kashyap and Scharfstein [1991]nd that investment by Japanese rms that belong to a keiretsu(corporate group) is less sensitive to cash ow than investment byindependent rms They conclude that a group (and concomitantbank) afliation alleviates underinvestment problems caused bycapital market imperfections

Despite the size and policy-importance of this literature thefundamental assumptions underlying it have remained largelyunexplored While subsequent work has replicated the ndingsof FHP [1988] by using different a priori criteria no paper (ofwhich we are aware) has veried directly whether a higherinvestment-cash ow sensitivity is related to nancing problemsand if it is in what way In particular there is no test of thefundamental assumptionmdashimplicit in all these testsmdashthatinvestment-cash ow sensitivities increase monotonically withthe degree of nancing constraints As we show in Section I thisis particularly surprising because there is no strong theoreticalreason to expect a monotonic relationship

This paper investigates the relation between investment-cash ow sensitivities and nancing constraints by undertakingan in-depth analysis of a sample of rms exhibiting an unusuallyhigh sensitivity of investment to cash ow These rms are the49 low dividend rms that FHP [1988] identify as nancially con-strained according to the investment-cash ow criterion

By using detailed and previously unexplored data sourceswe try to determine the availability of and the demand for fundsfor each of the sample rms We examine each rmrsquos annualreport or 10-K for each sample year and we read managementrsquosdiscussion of liquidity that describes the rmrsquos future needs forfunds and the sources it plans to use to meet those needs Weintegrate this information with quantitative data and with publicnews to derive as complete a picture as possible of the availabilityof internal and external funds for each rm as well as each rmrsquosdemand for funds On this basis we rank the extent to which thesample rms are nancially constrained each year We use the

QUARTERLY JOURNAL OF ECONOMICS170

rm-year classications to group the sample rms over seven- oreight-year subperiods and over the entire sample period Finallywe compare investment-cash ow sensitivities across the differ-ent groups of rms for the entire sample period for subperiodsand for individual years

Surprisingly we nd that in only 15 percent of rm-years isthere some question as to a rmrsquos ability to access internal orexternal funds to increase investment In 85 percent of rm-yearsthe rms could have increased their investmentmdashin many casessubstantiallymdashif they had so chosen In fact almost 40 percent ofthe sample rms including Hewlett-Packard (cited above) couldhave increased their investment in every year of the sample pe-riod Our partially qualitative measures of nancial constraintsare strongly corroborated by quantitive data on debt to total capi-tal interest coverage the presence of restrictions on dividendsand nancial slack (the level of cash and unused line of creditrelative to investment)

More strikingly those rms classied as less nanciallyconstrained exhibit a signicantly greater investment-cash owsensitivity than those rms classied as more nancially con-strained We nd this pattern for the entire sample period forsubperiods and for individual years This pattern is also robustto different criteria to divide constrained and unconstrainedrms For example rms with healthy interest coverage in everysample year or in every subperiod year have investment-cashow sensitivities twice as large as the remaining rms in thesample

As we explain in Section I these results should not be verysurprising There is no strong theoretical reason for investment-cash ow sensitivities to increase monotonically with the degreeof nancing constraints Nevertheless we consider several pos-sible reasons why estimated investment-cash ow sensitivitiescould decrease in the degree of nancing constraints even if thetrue relationship is increasing

First cash ow may act as a proxy for investment opportuni-ties not captured by Tobinrsquos Q and do so differentially acrossrms Our results however are robust to the use of an Eulerequation test [Bond and Meghir 1994] which does not rely onTobinrsquos Q and thus is not affected by its mismeasurement

Second differences in sensitivities might be driven by a fewinuential outliers We nd evidence that the high overall sensi-tivity of our sample (FHPrsquos [1988] low dividend payout rms) rela-

INVESTMENT-CASH FLOW SENSITIVITIES 171

tive to FHPrsquos higher dividend payout rms is explained by a rela-tively few company-years characterized by exceptionally highsales growth We also nd however that these outliers do notexplain our cross-section results that the least constrained rmshave the highest sensitivities

Third our nding of nonmonotonic relationship may be spe-cic to a few distressed rms that are forced to use cash ow torepay their debt and may not apply to more ldquonormalrdquo samplesThe nancial conditions of the constrained rms though are notconsistent with this hypothesis

In sum we provide both theoretical reasons and empiricalevidence that a greater sensitivity of investment to cash ow isnot a reliable measure of the differential cost between internaland external nance In so doing we address (and refute) thecriticisms in Fazzari Hubbard and Petersen [1996] [FHP 1996]

We conclude the paper with a discussion of the generalityof our results We argue that our analysis calls into questionthe interpretation of most previous research that uses thismethodology

The paper proceeds as follows Section I presents the theoreti-cal framework Section II describes the sample Section III ex-plains the criteria used to identify the extent to which rms arenancially constrained Section IV reports the investment-cashow regression results Section V discusses the results and con-siders alternative explanations for them Section VI discusses theimplications and generality of our results for the previous litera-ture Section VII concludes

I THEORETICAL FRAMEWORK

A Denition of Financing Constraints

In order to discuss the relationship between investment-cashow sensitivity and the degree of nancing constraints we mustdene what it means to be nancially constrained The most pre-cise (but also broadest) denition classies rms as nanciallyconstrained if they face a wedge between the internal and exter-nal costs of funds By this denition all rms are likely to be clas-sied as constrained A small transaction cost of raising externalfunds would be sufcient to put a rm into this category Thisdenition however provides a useful framework to differentiaterms according to the extent to which they are nancially con-

QUARTERLY JOURNAL OF ECONOMICS172

strained A rm is considered more nancially constrained as thewedge between its internal and external cost of funds increases

Our classication scheme which we detail below is designedto distinguish the relative differences in the degree to which rmsare nancially constrained In general our unconstrained or lessconstrained rms are those rms with relatively large amountsof liquid assets and net worth

In classifying rms we are agnostic on whether the wedgebetween the cost of internal and external funds is caused by hid-den information problems as in Myers and Majluf [1984] andGreenwald Stiglitz and Weiss [1984] or agency problems as inJensen and Meckling [1976] Grossman and Hart [1982] Jensen[1986] Stulz [1990] and Hart and Moore [1995] In fact unlikeBlanchard Lopez-de-Silanes and Shleifer [1994] the purpose ofour analysis is not to identify the source of the capital marketimperfection but rather to understand the effects capital marketimperfections have on investment We next review what economictheory has to say about the impact of nancing constraints oninvestment

B The Impact of Financing Constraints on Investments

FHP [1988] was the rst of many papers to consider higherinvestment-cash ow sensitivities as evidence of greater nancingconstraints Given the magnitude and the importance of thisliterature it is surprising that little attention has been given tothe theoretical foundation of the investment-cash ow sensitivitycriterion1 While it is easy to show that constrained rms shouldbe sensitive to internal cash ow while unconstrained rmsshould not it is not necessarily true that the magnitude of thesensitivity increases in the degree of nancing constraints Thisis the crucial question given that investment is sensitive to cashow for the vast majority of rms analyzed (It is easy to justifythis sensitivity based on the fact that external funds are morecostly than internal funds for all rms as long as some trans-action costs are involved)

The difculty of interpreting cross-sectional differences ininvestment-cash ow sensitivities can be illustrated with asimple one-period model Consider a rm that chooses the levelof investment to maximize prots The return to an investmentI is given by a production function F(I ) where F 9 and F 0 0

1 We thank Jeremy Stein for encouraging us to develop this point

INVESTMENT-CASH FLOW SENSITIVITIES 173

Investment can be nanced either with internal funds (W) or withexternal funds (E) The opportunity cost of internal funds equalsthe cost of capital R which for simplicity we set equal to 1 Be-cause of information agency or risk aversion problems we as-sume that the use of external funds generates a deadweight costwhichmdashin a competitive capital marketmdashis borne by the issuingrm We represent (in reduced form) this additional cost of exter-nal funds with the function C(Ek) where E is the amount of ex-ternal funds raised and k is a measure of a rmrsquos wedge betweenthe internal and the external costs of funds It is natural to as-sume that the total cost of raising external funds increases inthe amount of funds raised and in the extent of the agency orinformation problems (represented by k) All the a priori mea-sures of nancing constraints used in the literature can bethought of as different proxies for k (which is unobservable) or ofW (the availability of internal funds)

Each rm then chooses I to maximize

(1) max ( ) ( ) F I C E k I I W E such that + - - =

To guarantee that the above program is well behaved we alsoassume that C() is convex in E2

The rst-order condition of problem (1) then is given by

(2) F I C I W k1 11( ) ( ) + = -

where C1(0) represents the partial derivative of C with respect toits rst argument and F1() the rst derivative of F with respectto I The effects of the availability of internal nance on invest-ments can be easily obtained by implicit differentiation of (2)

(3)dI

dWC

C F

11

11

=- 11

which is clearly positive (to the extent that C is convex) In otherwords in an imperfect capital market world investments aresensitive to internal funds while in a perfect capital marketworld they are not (because C() 5 0 and thus C11 5 0)

Similarly it is possible to derive the sensitivity of investmentto the wedge between the cost of internal and external nancingBy implicit differentiation of (2) we obtain

2 This is a reasonable but not obvious assumption For example Calomirisand Himmelberg [1995] document that the average transaction cost of issuingsecurities decreases in the amount raised which suggests that C() may be con-cave While these transaction costs may be only a small component of the overallcost C() we note that this basic assumption might not be warranted

QUARTERLY JOURNAL OF ECONOMICS174

(4)dIdk

CC F

12

11

=-

- 11

which is negative if the marginal cost of raising external nanceis increasing in k (ie C12 0)

Most papers in this literature however do not test either ofthese two propositions On the one hand the estimatedinvestment-cash ow sensitivity is generally positive and signi-cant for all rms suggesting that all rms are constrained insome sense and so making the test of the rst implication redun-dant Second most of the proxies for W or k used in the literatureare only able to identify constrained rms not constrained rm-years This makes it impossible to disentangle the effect of -nancing constraints from a rm-specic effect on the level of in-vestment

For these reasons previous papers focus on cross-sectionaldifferences in the investment-cash-ow sensitivity across groupsof rms likely to have a different wedge between internal andexternal funds But this corresponds to looking at differences indIdW as a function of W or k Such an exercise is meaningfulonly if the investment-cash ow sensitivity is monotonically de-creasing with respect to W (or increasing with respect to k) inother words only if d2IdW2 is negative (or d2IdWdk is positive)From equation (3) we obtain

(5)d IdW

F C C FC F

2

2111 11

2112

113

111

11

=--( )

If both C11() and F11() are different from zero we can rewrite equa-tion (5) as

(6) d IdW

FF

CC

F CC F

2

2111

112

111

112

112

112

113

11

= -aelig

egraveccedil

ouml

oslashdivide -( )

Given that the second term is always positive it follows that d2IdW2 is negative if and only if [F111F11

2 2 C111C112] is negative

This condition implies a certain relationship between the curva-ture of the production function and the curvature of the cost func-tion at the optimal level of investment It is easy to see how sucha condition can be violated For example if the cost function isquadratic d2IdW 2 will be positive if the third derivative of theproduction function is positive (as is the case with a simple pro-duction function like I r where 0 r 1) In such a case theinvestment-cash ow sensitivity increases with a rmrsquos internal

INVESTMENT-CASH FLOW SENSITIVITIES 175

liquidity Of course many simple production functions have posi-tive third derivatives Although we will not produce them herethe conditions necessary to ensure that d2IdWdk be positive areat least as demanding

In sum even in a one-period model investment-cash owsensitivities do not necessarily increase with the degree of nan-cing constraints In a multiperiod model precautionary savingsmotives make it even more difcult to assess the theoretical rela-tionship between investment-cash ow sensitivities and the de-gree of nancing constraints For example Gross [1995] buildsand simulates an intertemporal investment model and nds anonmonotonic relationship between investment-cash ow sensi-tivities and the extent of nancing constraints

Finally the relationship between investment-cash ow sen-sitivities and degree of nancing constraints can be further com-plicated by the presence of irrational or overly risk-averse man-agers who choose to rely primarily on internal cash ow to investdespite the availability of low cost funds

II SAMPLE

In this paper we analyze the sample of 49 low-dividend pay-ing rms in FHP [1988] FHP divide all manufacturing rms inthe Value Line database with uninterrupted data from 1970 to1984 into three classes based on dividend payout policy Their 49Class 1 rms (which we analyze) have a dividend payout ratio ofless than 10 percent in at least ten of the fteen years FHP clas-sify 39 rms that have a dividend payout ratio between 10 per-cent and 20 percent as Class 2 rms and all 334 other rms intheir sample as Class 3 rms FHP argue that the Class 1 rmsare more likely a priori to have been nancially constrained Intheir analysis they nd that the Class 1 rms have aninvestment-cash ow sensitivity that is signicantly greater thanthat for rms that pay higher dividends

We choose this sample for three reasons First these rmsexhibit a strong relation between investment and cash ow Sec-ond FHP argue strongly that these rms are nancially con-strained most likely because of information problems BecauseFHP [1988] can legitimately be considered the parent of all pa-pers in this literature there can be no disagreement that we haveadversely selected our sample Finally given the high cost of ourresearch design the number of rms is manageable

We follow this sample for the same fteen years 1970 to

QUARTERLY JOURNAL OF ECONOMICS176

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

cash ow sensitivities of the different subsamples The studiesinterpret a greater investment-cash ow sensitivity for rms con-sidered more likely to face a larger wedge between the internaland the external cost of funds as evidence that the rms are in-deed constrained This methodology has been widely applied toidentify rms that are more affected by nancing constraintsand institutions that are more likely to alleviate those con-straints For example Hoshi Kashyap and Scharfstein [1991]nd that investment by Japanese rms that belong to a keiretsu(corporate group) is less sensitive to cash ow than investment byindependent rms They conclude that a group (and concomitantbank) afliation alleviates underinvestment problems caused bycapital market imperfections

Despite the size and policy-importance of this literature thefundamental assumptions underlying it have remained largelyunexplored While subsequent work has replicated the ndingsof FHP [1988] by using different a priori criteria no paper (ofwhich we are aware) has veried directly whether a higherinvestment-cash ow sensitivity is related to nancing problemsand if it is in what way In particular there is no test of thefundamental assumptionmdashimplicit in all these testsmdashthatinvestment-cash ow sensitivities increase monotonically withthe degree of nancing constraints As we show in Section I thisis particularly surprising because there is no strong theoreticalreason to expect a monotonic relationship

This paper investigates the relation between investment-cash ow sensitivities and nancing constraints by undertakingan in-depth analysis of a sample of rms exhibiting an unusuallyhigh sensitivity of investment to cash ow These rms are the49 low dividend rms that FHP [1988] identify as nancially con-strained according to the investment-cash ow criterion

By using detailed and previously unexplored data sourceswe try to determine the availability of and the demand for fundsfor each of the sample rms We examine each rmrsquos annualreport or 10-K for each sample year and we read managementrsquosdiscussion of liquidity that describes the rmrsquos future needs forfunds and the sources it plans to use to meet those needs Weintegrate this information with quantitative data and with publicnews to derive as complete a picture as possible of the availabilityof internal and external funds for each rm as well as each rmrsquosdemand for funds On this basis we rank the extent to which thesample rms are nancially constrained each year We use the

QUARTERLY JOURNAL OF ECONOMICS170

rm-year classications to group the sample rms over seven- oreight-year subperiods and over the entire sample period Finallywe compare investment-cash ow sensitivities across the differ-ent groups of rms for the entire sample period for subperiodsand for individual years

Surprisingly we nd that in only 15 percent of rm-years isthere some question as to a rmrsquos ability to access internal orexternal funds to increase investment In 85 percent of rm-yearsthe rms could have increased their investmentmdashin many casessubstantiallymdashif they had so chosen In fact almost 40 percent ofthe sample rms including Hewlett-Packard (cited above) couldhave increased their investment in every year of the sample pe-riod Our partially qualitative measures of nancial constraintsare strongly corroborated by quantitive data on debt to total capi-tal interest coverage the presence of restrictions on dividendsand nancial slack (the level of cash and unused line of creditrelative to investment)

More strikingly those rms classied as less nanciallyconstrained exhibit a signicantly greater investment-cash owsensitivity than those rms classied as more nancially con-strained We nd this pattern for the entire sample period forsubperiods and for individual years This pattern is also robustto different criteria to divide constrained and unconstrainedrms For example rms with healthy interest coverage in everysample year or in every subperiod year have investment-cashow sensitivities twice as large as the remaining rms in thesample

As we explain in Section I these results should not be verysurprising There is no strong theoretical reason for investment-cash ow sensitivities to increase monotonically with the degreeof nancing constraints Nevertheless we consider several pos-sible reasons why estimated investment-cash ow sensitivitiescould decrease in the degree of nancing constraints even if thetrue relationship is increasing

First cash ow may act as a proxy for investment opportuni-ties not captured by Tobinrsquos Q and do so differentially acrossrms Our results however are robust to the use of an Eulerequation test [Bond and Meghir 1994] which does not rely onTobinrsquos Q and thus is not affected by its mismeasurement

Second differences in sensitivities might be driven by a fewinuential outliers We nd evidence that the high overall sensi-tivity of our sample (FHPrsquos [1988] low dividend payout rms) rela-

INVESTMENT-CASH FLOW SENSITIVITIES 171

tive to FHPrsquos higher dividend payout rms is explained by a rela-tively few company-years characterized by exceptionally highsales growth We also nd however that these outliers do notexplain our cross-section results that the least constrained rmshave the highest sensitivities

Third our nding of nonmonotonic relationship may be spe-cic to a few distressed rms that are forced to use cash ow torepay their debt and may not apply to more ldquonormalrdquo samplesThe nancial conditions of the constrained rms though are notconsistent with this hypothesis

In sum we provide both theoretical reasons and empiricalevidence that a greater sensitivity of investment to cash ow isnot a reliable measure of the differential cost between internaland external nance In so doing we address (and refute) thecriticisms in Fazzari Hubbard and Petersen [1996] [FHP 1996]

We conclude the paper with a discussion of the generalityof our results We argue that our analysis calls into questionthe interpretation of most previous research that uses thismethodology

The paper proceeds as follows Section I presents the theoreti-cal framework Section II describes the sample Section III ex-plains the criteria used to identify the extent to which rms arenancially constrained Section IV reports the investment-cashow regression results Section V discusses the results and con-siders alternative explanations for them Section VI discusses theimplications and generality of our results for the previous litera-ture Section VII concludes

I THEORETICAL FRAMEWORK

A Denition of Financing Constraints

In order to discuss the relationship between investment-cashow sensitivity and the degree of nancing constraints we mustdene what it means to be nancially constrained The most pre-cise (but also broadest) denition classies rms as nanciallyconstrained if they face a wedge between the internal and exter-nal costs of funds By this denition all rms are likely to be clas-sied as constrained A small transaction cost of raising externalfunds would be sufcient to put a rm into this category Thisdenition however provides a useful framework to differentiaterms according to the extent to which they are nancially con-

QUARTERLY JOURNAL OF ECONOMICS172

strained A rm is considered more nancially constrained as thewedge between its internal and external cost of funds increases

Our classication scheme which we detail below is designedto distinguish the relative differences in the degree to which rmsare nancially constrained In general our unconstrained or lessconstrained rms are those rms with relatively large amountsof liquid assets and net worth

In classifying rms we are agnostic on whether the wedgebetween the cost of internal and external funds is caused by hid-den information problems as in Myers and Majluf [1984] andGreenwald Stiglitz and Weiss [1984] or agency problems as inJensen and Meckling [1976] Grossman and Hart [1982] Jensen[1986] Stulz [1990] and Hart and Moore [1995] In fact unlikeBlanchard Lopez-de-Silanes and Shleifer [1994] the purpose ofour analysis is not to identify the source of the capital marketimperfection but rather to understand the effects capital marketimperfections have on investment We next review what economictheory has to say about the impact of nancing constraints oninvestment

B The Impact of Financing Constraints on Investments

FHP [1988] was the rst of many papers to consider higherinvestment-cash ow sensitivities as evidence of greater nancingconstraints Given the magnitude and the importance of thisliterature it is surprising that little attention has been given tothe theoretical foundation of the investment-cash ow sensitivitycriterion1 While it is easy to show that constrained rms shouldbe sensitive to internal cash ow while unconstrained rmsshould not it is not necessarily true that the magnitude of thesensitivity increases in the degree of nancing constraints Thisis the crucial question given that investment is sensitive to cashow for the vast majority of rms analyzed (It is easy to justifythis sensitivity based on the fact that external funds are morecostly than internal funds for all rms as long as some trans-action costs are involved)

The difculty of interpreting cross-sectional differences ininvestment-cash ow sensitivities can be illustrated with asimple one-period model Consider a rm that chooses the levelof investment to maximize prots The return to an investmentI is given by a production function F(I ) where F 9 and F 0 0

1 We thank Jeremy Stein for encouraging us to develop this point

INVESTMENT-CASH FLOW SENSITIVITIES 173

Investment can be nanced either with internal funds (W) or withexternal funds (E) The opportunity cost of internal funds equalsthe cost of capital R which for simplicity we set equal to 1 Be-cause of information agency or risk aversion problems we as-sume that the use of external funds generates a deadweight costwhichmdashin a competitive capital marketmdashis borne by the issuingrm We represent (in reduced form) this additional cost of exter-nal funds with the function C(Ek) where E is the amount of ex-ternal funds raised and k is a measure of a rmrsquos wedge betweenthe internal and the external costs of funds It is natural to as-sume that the total cost of raising external funds increases inthe amount of funds raised and in the extent of the agency orinformation problems (represented by k) All the a priori mea-sures of nancing constraints used in the literature can bethought of as different proxies for k (which is unobservable) or ofW (the availability of internal funds)

Each rm then chooses I to maximize

(1) max ( ) ( ) F I C E k I I W E such that + - - =

To guarantee that the above program is well behaved we alsoassume that C() is convex in E2

The rst-order condition of problem (1) then is given by

(2) F I C I W k1 11( ) ( ) + = -

where C1(0) represents the partial derivative of C with respect toits rst argument and F1() the rst derivative of F with respectto I The effects of the availability of internal nance on invest-ments can be easily obtained by implicit differentiation of (2)

(3)dI

dWC

C F

11

11

=- 11

which is clearly positive (to the extent that C is convex) In otherwords in an imperfect capital market world investments aresensitive to internal funds while in a perfect capital marketworld they are not (because C() 5 0 and thus C11 5 0)

Similarly it is possible to derive the sensitivity of investmentto the wedge between the cost of internal and external nancingBy implicit differentiation of (2) we obtain

2 This is a reasonable but not obvious assumption For example Calomirisand Himmelberg [1995] document that the average transaction cost of issuingsecurities decreases in the amount raised which suggests that C() may be con-cave While these transaction costs may be only a small component of the overallcost C() we note that this basic assumption might not be warranted

QUARTERLY JOURNAL OF ECONOMICS174

(4)dIdk

CC F

12

11

=-

- 11

which is negative if the marginal cost of raising external nanceis increasing in k (ie C12 0)

Most papers in this literature however do not test either ofthese two propositions On the one hand the estimatedinvestment-cash ow sensitivity is generally positive and signi-cant for all rms suggesting that all rms are constrained insome sense and so making the test of the rst implication redun-dant Second most of the proxies for W or k used in the literatureare only able to identify constrained rms not constrained rm-years This makes it impossible to disentangle the effect of -nancing constraints from a rm-specic effect on the level of in-vestment

For these reasons previous papers focus on cross-sectionaldifferences in the investment-cash-ow sensitivity across groupsof rms likely to have a different wedge between internal andexternal funds But this corresponds to looking at differences indIdW as a function of W or k Such an exercise is meaningfulonly if the investment-cash ow sensitivity is monotonically de-creasing with respect to W (or increasing with respect to k) inother words only if d2IdW2 is negative (or d2IdWdk is positive)From equation (3) we obtain

(5)d IdW

F C C FC F

2

2111 11

2112

113

111

11

=--( )

If both C11() and F11() are different from zero we can rewrite equa-tion (5) as

(6) d IdW

FF

CC

F CC F

2

2111

112

111

112

112

112

113

11

= -aelig

egraveccedil

ouml

oslashdivide -( )

Given that the second term is always positive it follows that d2IdW2 is negative if and only if [F111F11

2 2 C111C112] is negative

This condition implies a certain relationship between the curva-ture of the production function and the curvature of the cost func-tion at the optimal level of investment It is easy to see how sucha condition can be violated For example if the cost function isquadratic d2IdW 2 will be positive if the third derivative of theproduction function is positive (as is the case with a simple pro-duction function like I r where 0 r 1) In such a case theinvestment-cash ow sensitivity increases with a rmrsquos internal

INVESTMENT-CASH FLOW SENSITIVITIES 175

liquidity Of course many simple production functions have posi-tive third derivatives Although we will not produce them herethe conditions necessary to ensure that d2IdWdk be positive areat least as demanding

In sum even in a one-period model investment-cash owsensitivities do not necessarily increase with the degree of nan-cing constraints In a multiperiod model precautionary savingsmotives make it even more difcult to assess the theoretical rela-tionship between investment-cash ow sensitivities and the de-gree of nancing constraints For example Gross [1995] buildsand simulates an intertemporal investment model and nds anonmonotonic relationship between investment-cash ow sensi-tivities and the extent of nancing constraints

Finally the relationship between investment-cash ow sen-sitivities and degree of nancing constraints can be further com-plicated by the presence of irrational or overly risk-averse man-agers who choose to rely primarily on internal cash ow to investdespite the availability of low cost funds

II SAMPLE

In this paper we analyze the sample of 49 low-dividend pay-ing rms in FHP [1988] FHP divide all manufacturing rms inthe Value Line database with uninterrupted data from 1970 to1984 into three classes based on dividend payout policy Their 49Class 1 rms (which we analyze) have a dividend payout ratio ofless than 10 percent in at least ten of the fteen years FHP clas-sify 39 rms that have a dividend payout ratio between 10 per-cent and 20 percent as Class 2 rms and all 334 other rms intheir sample as Class 3 rms FHP argue that the Class 1 rmsare more likely a priori to have been nancially constrained Intheir analysis they nd that the Class 1 rms have aninvestment-cash ow sensitivity that is signicantly greater thanthat for rms that pay higher dividends

We choose this sample for three reasons First these rmsexhibit a strong relation between investment and cash ow Sec-ond FHP argue strongly that these rms are nancially con-strained most likely because of information problems BecauseFHP [1988] can legitimately be considered the parent of all pa-pers in this literature there can be no disagreement that we haveadversely selected our sample Finally given the high cost of ourresearch design the number of rms is manageable

We follow this sample for the same fteen years 1970 to

QUARTERLY JOURNAL OF ECONOMICS176

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

rm-year classications to group the sample rms over seven- oreight-year subperiods and over the entire sample period Finallywe compare investment-cash ow sensitivities across the differ-ent groups of rms for the entire sample period for subperiodsand for individual years

Surprisingly we nd that in only 15 percent of rm-years isthere some question as to a rmrsquos ability to access internal orexternal funds to increase investment In 85 percent of rm-yearsthe rms could have increased their investmentmdashin many casessubstantiallymdashif they had so chosen In fact almost 40 percent ofthe sample rms including Hewlett-Packard (cited above) couldhave increased their investment in every year of the sample pe-riod Our partially qualitative measures of nancial constraintsare strongly corroborated by quantitive data on debt to total capi-tal interest coverage the presence of restrictions on dividendsand nancial slack (the level of cash and unused line of creditrelative to investment)

More strikingly those rms classied as less nanciallyconstrained exhibit a signicantly greater investment-cash owsensitivity than those rms classied as more nancially con-strained We nd this pattern for the entire sample period forsubperiods and for individual years This pattern is also robustto different criteria to divide constrained and unconstrainedrms For example rms with healthy interest coverage in everysample year or in every subperiod year have investment-cashow sensitivities twice as large as the remaining rms in thesample

As we explain in Section I these results should not be verysurprising There is no strong theoretical reason for investment-cash ow sensitivities to increase monotonically with the degreeof nancing constraints Nevertheless we consider several pos-sible reasons why estimated investment-cash ow sensitivitiescould decrease in the degree of nancing constraints even if thetrue relationship is increasing

First cash ow may act as a proxy for investment opportuni-ties not captured by Tobinrsquos Q and do so differentially acrossrms Our results however are robust to the use of an Eulerequation test [Bond and Meghir 1994] which does not rely onTobinrsquos Q and thus is not affected by its mismeasurement

Second differences in sensitivities might be driven by a fewinuential outliers We nd evidence that the high overall sensi-tivity of our sample (FHPrsquos [1988] low dividend payout rms) rela-

INVESTMENT-CASH FLOW SENSITIVITIES 171

tive to FHPrsquos higher dividend payout rms is explained by a rela-tively few company-years characterized by exceptionally highsales growth We also nd however that these outliers do notexplain our cross-section results that the least constrained rmshave the highest sensitivities

Third our nding of nonmonotonic relationship may be spe-cic to a few distressed rms that are forced to use cash ow torepay their debt and may not apply to more ldquonormalrdquo samplesThe nancial conditions of the constrained rms though are notconsistent with this hypothesis

In sum we provide both theoretical reasons and empiricalevidence that a greater sensitivity of investment to cash ow isnot a reliable measure of the differential cost between internaland external nance In so doing we address (and refute) thecriticisms in Fazzari Hubbard and Petersen [1996] [FHP 1996]

We conclude the paper with a discussion of the generalityof our results We argue that our analysis calls into questionthe interpretation of most previous research that uses thismethodology

The paper proceeds as follows Section I presents the theoreti-cal framework Section II describes the sample Section III ex-plains the criteria used to identify the extent to which rms arenancially constrained Section IV reports the investment-cashow regression results Section V discusses the results and con-siders alternative explanations for them Section VI discusses theimplications and generality of our results for the previous litera-ture Section VII concludes

I THEORETICAL FRAMEWORK

A Denition of Financing Constraints

In order to discuss the relationship between investment-cashow sensitivity and the degree of nancing constraints we mustdene what it means to be nancially constrained The most pre-cise (but also broadest) denition classies rms as nanciallyconstrained if they face a wedge between the internal and exter-nal costs of funds By this denition all rms are likely to be clas-sied as constrained A small transaction cost of raising externalfunds would be sufcient to put a rm into this category Thisdenition however provides a useful framework to differentiaterms according to the extent to which they are nancially con-

QUARTERLY JOURNAL OF ECONOMICS172

strained A rm is considered more nancially constrained as thewedge between its internal and external cost of funds increases

Our classication scheme which we detail below is designedto distinguish the relative differences in the degree to which rmsare nancially constrained In general our unconstrained or lessconstrained rms are those rms with relatively large amountsof liquid assets and net worth

In classifying rms we are agnostic on whether the wedgebetween the cost of internal and external funds is caused by hid-den information problems as in Myers and Majluf [1984] andGreenwald Stiglitz and Weiss [1984] or agency problems as inJensen and Meckling [1976] Grossman and Hart [1982] Jensen[1986] Stulz [1990] and Hart and Moore [1995] In fact unlikeBlanchard Lopez-de-Silanes and Shleifer [1994] the purpose ofour analysis is not to identify the source of the capital marketimperfection but rather to understand the effects capital marketimperfections have on investment We next review what economictheory has to say about the impact of nancing constraints oninvestment

B The Impact of Financing Constraints on Investments

FHP [1988] was the rst of many papers to consider higherinvestment-cash ow sensitivities as evidence of greater nancingconstraints Given the magnitude and the importance of thisliterature it is surprising that little attention has been given tothe theoretical foundation of the investment-cash ow sensitivitycriterion1 While it is easy to show that constrained rms shouldbe sensitive to internal cash ow while unconstrained rmsshould not it is not necessarily true that the magnitude of thesensitivity increases in the degree of nancing constraints Thisis the crucial question given that investment is sensitive to cashow for the vast majority of rms analyzed (It is easy to justifythis sensitivity based on the fact that external funds are morecostly than internal funds for all rms as long as some trans-action costs are involved)

The difculty of interpreting cross-sectional differences ininvestment-cash ow sensitivities can be illustrated with asimple one-period model Consider a rm that chooses the levelof investment to maximize prots The return to an investmentI is given by a production function F(I ) where F 9 and F 0 0

1 We thank Jeremy Stein for encouraging us to develop this point

INVESTMENT-CASH FLOW SENSITIVITIES 173

Investment can be nanced either with internal funds (W) or withexternal funds (E) The opportunity cost of internal funds equalsthe cost of capital R which for simplicity we set equal to 1 Be-cause of information agency or risk aversion problems we as-sume that the use of external funds generates a deadweight costwhichmdashin a competitive capital marketmdashis borne by the issuingrm We represent (in reduced form) this additional cost of exter-nal funds with the function C(Ek) where E is the amount of ex-ternal funds raised and k is a measure of a rmrsquos wedge betweenthe internal and the external costs of funds It is natural to as-sume that the total cost of raising external funds increases inthe amount of funds raised and in the extent of the agency orinformation problems (represented by k) All the a priori mea-sures of nancing constraints used in the literature can bethought of as different proxies for k (which is unobservable) or ofW (the availability of internal funds)

Each rm then chooses I to maximize

(1) max ( ) ( ) F I C E k I I W E such that + - - =

To guarantee that the above program is well behaved we alsoassume that C() is convex in E2

The rst-order condition of problem (1) then is given by

(2) F I C I W k1 11( ) ( ) + = -

where C1(0) represents the partial derivative of C with respect toits rst argument and F1() the rst derivative of F with respectto I The effects of the availability of internal nance on invest-ments can be easily obtained by implicit differentiation of (2)

(3)dI

dWC

C F

11

11

=- 11

which is clearly positive (to the extent that C is convex) In otherwords in an imperfect capital market world investments aresensitive to internal funds while in a perfect capital marketworld they are not (because C() 5 0 and thus C11 5 0)

Similarly it is possible to derive the sensitivity of investmentto the wedge between the cost of internal and external nancingBy implicit differentiation of (2) we obtain

2 This is a reasonable but not obvious assumption For example Calomirisand Himmelberg [1995] document that the average transaction cost of issuingsecurities decreases in the amount raised which suggests that C() may be con-cave While these transaction costs may be only a small component of the overallcost C() we note that this basic assumption might not be warranted

QUARTERLY JOURNAL OF ECONOMICS174

(4)dIdk

CC F

12

11

=-

- 11

which is negative if the marginal cost of raising external nanceis increasing in k (ie C12 0)

Most papers in this literature however do not test either ofthese two propositions On the one hand the estimatedinvestment-cash ow sensitivity is generally positive and signi-cant for all rms suggesting that all rms are constrained insome sense and so making the test of the rst implication redun-dant Second most of the proxies for W or k used in the literatureare only able to identify constrained rms not constrained rm-years This makes it impossible to disentangle the effect of -nancing constraints from a rm-specic effect on the level of in-vestment

For these reasons previous papers focus on cross-sectionaldifferences in the investment-cash-ow sensitivity across groupsof rms likely to have a different wedge between internal andexternal funds But this corresponds to looking at differences indIdW as a function of W or k Such an exercise is meaningfulonly if the investment-cash ow sensitivity is monotonically de-creasing with respect to W (or increasing with respect to k) inother words only if d2IdW2 is negative (or d2IdWdk is positive)From equation (3) we obtain

(5)d IdW

F C C FC F

2

2111 11

2112

113

111

11

=--( )

If both C11() and F11() are different from zero we can rewrite equa-tion (5) as

(6) d IdW

FF

CC

F CC F

2

2111

112

111

112

112

112

113

11

= -aelig

egraveccedil

ouml

oslashdivide -( )

Given that the second term is always positive it follows that d2IdW2 is negative if and only if [F111F11

2 2 C111C112] is negative

This condition implies a certain relationship between the curva-ture of the production function and the curvature of the cost func-tion at the optimal level of investment It is easy to see how sucha condition can be violated For example if the cost function isquadratic d2IdW 2 will be positive if the third derivative of theproduction function is positive (as is the case with a simple pro-duction function like I r where 0 r 1) In such a case theinvestment-cash ow sensitivity increases with a rmrsquos internal

INVESTMENT-CASH FLOW SENSITIVITIES 175

liquidity Of course many simple production functions have posi-tive third derivatives Although we will not produce them herethe conditions necessary to ensure that d2IdWdk be positive areat least as demanding

In sum even in a one-period model investment-cash owsensitivities do not necessarily increase with the degree of nan-cing constraints In a multiperiod model precautionary savingsmotives make it even more difcult to assess the theoretical rela-tionship between investment-cash ow sensitivities and the de-gree of nancing constraints For example Gross [1995] buildsand simulates an intertemporal investment model and nds anonmonotonic relationship between investment-cash ow sensi-tivities and the extent of nancing constraints

Finally the relationship between investment-cash ow sen-sitivities and degree of nancing constraints can be further com-plicated by the presence of irrational or overly risk-averse man-agers who choose to rely primarily on internal cash ow to investdespite the availability of low cost funds

II SAMPLE

In this paper we analyze the sample of 49 low-dividend pay-ing rms in FHP [1988] FHP divide all manufacturing rms inthe Value Line database with uninterrupted data from 1970 to1984 into three classes based on dividend payout policy Their 49Class 1 rms (which we analyze) have a dividend payout ratio ofless than 10 percent in at least ten of the fteen years FHP clas-sify 39 rms that have a dividend payout ratio between 10 per-cent and 20 percent as Class 2 rms and all 334 other rms intheir sample as Class 3 rms FHP argue that the Class 1 rmsare more likely a priori to have been nancially constrained Intheir analysis they nd that the Class 1 rms have aninvestment-cash ow sensitivity that is signicantly greater thanthat for rms that pay higher dividends

We choose this sample for three reasons First these rmsexhibit a strong relation between investment and cash ow Sec-ond FHP argue strongly that these rms are nancially con-strained most likely because of information problems BecauseFHP [1988] can legitimately be considered the parent of all pa-pers in this literature there can be no disagreement that we haveadversely selected our sample Finally given the high cost of ourresearch design the number of rms is manageable

We follow this sample for the same fteen years 1970 to

QUARTERLY JOURNAL OF ECONOMICS176

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

tive to FHPrsquos higher dividend payout rms is explained by a rela-tively few company-years characterized by exceptionally highsales growth We also nd however that these outliers do notexplain our cross-section results that the least constrained rmshave the highest sensitivities

Third our nding of nonmonotonic relationship may be spe-cic to a few distressed rms that are forced to use cash ow torepay their debt and may not apply to more ldquonormalrdquo samplesThe nancial conditions of the constrained rms though are notconsistent with this hypothesis

In sum we provide both theoretical reasons and empiricalevidence that a greater sensitivity of investment to cash ow isnot a reliable measure of the differential cost between internaland external nance In so doing we address (and refute) thecriticisms in Fazzari Hubbard and Petersen [1996] [FHP 1996]

We conclude the paper with a discussion of the generalityof our results We argue that our analysis calls into questionthe interpretation of most previous research that uses thismethodology

The paper proceeds as follows Section I presents the theoreti-cal framework Section II describes the sample Section III ex-plains the criteria used to identify the extent to which rms arenancially constrained Section IV reports the investment-cashow regression results Section V discusses the results and con-siders alternative explanations for them Section VI discusses theimplications and generality of our results for the previous litera-ture Section VII concludes

I THEORETICAL FRAMEWORK

A Denition of Financing Constraints

In order to discuss the relationship between investment-cashow sensitivity and the degree of nancing constraints we mustdene what it means to be nancially constrained The most pre-cise (but also broadest) denition classies rms as nanciallyconstrained if they face a wedge between the internal and exter-nal costs of funds By this denition all rms are likely to be clas-sied as constrained A small transaction cost of raising externalfunds would be sufcient to put a rm into this category Thisdenition however provides a useful framework to differentiaterms according to the extent to which they are nancially con-

QUARTERLY JOURNAL OF ECONOMICS172

strained A rm is considered more nancially constrained as thewedge between its internal and external cost of funds increases

Our classication scheme which we detail below is designedto distinguish the relative differences in the degree to which rmsare nancially constrained In general our unconstrained or lessconstrained rms are those rms with relatively large amountsof liquid assets and net worth

In classifying rms we are agnostic on whether the wedgebetween the cost of internal and external funds is caused by hid-den information problems as in Myers and Majluf [1984] andGreenwald Stiglitz and Weiss [1984] or agency problems as inJensen and Meckling [1976] Grossman and Hart [1982] Jensen[1986] Stulz [1990] and Hart and Moore [1995] In fact unlikeBlanchard Lopez-de-Silanes and Shleifer [1994] the purpose ofour analysis is not to identify the source of the capital marketimperfection but rather to understand the effects capital marketimperfections have on investment We next review what economictheory has to say about the impact of nancing constraints oninvestment

B The Impact of Financing Constraints on Investments

FHP [1988] was the rst of many papers to consider higherinvestment-cash ow sensitivities as evidence of greater nancingconstraints Given the magnitude and the importance of thisliterature it is surprising that little attention has been given tothe theoretical foundation of the investment-cash ow sensitivitycriterion1 While it is easy to show that constrained rms shouldbe sensitive to internal cash ow while unconstrained rmsshould not it is not necessarily true that the magnitude of thesensitivity increases in the degree of nancing constraints Thisis the crucial question given that investment is sensitive to cashow for the vast majority of rms analyzed (It is easy to justifythis sensitivity based on the fact that external funds are morecostly than internal funds for all rms as long as some trans-action costs are involved)

The difculty of interpreting cross-sectional differences ininvestment-cash ow sensitivities can be illustrated with asimple one-period model Consider a rm that chooses the levelof investment to maximize prots The return to an investmentI is given by a production function F(I ) where F 9 and F 0 0

1 We thank Jeremy Stein for encouraging us to develop this point

INVESTMENT-CASH FLOW SENSITIVITIES 173

Investment can be nanced either with internal funds (W) or withexternal funds (E) The opportunity cost of internal funds equalsthe cost of capital R which for simplicity we set equal to 1 Be-cause of information agency or risk aversion problems we as-sume that the use of external funds generates a deadweight costwhichmdashin a competitive capital marketmdashis borne by the issuingrm We represent (in reduced form) this additional cost of exter-nal funds with the function C(Ek) where E is the amount of ex-ternal funds raised and k is a measure of a rmrsquos wedge betweenthe internal and the external costs of funds It is natural to as-sume that the total cost of raising external funds increases inthe amount of funds raised and in the extent of the agency orinformation problems (represented by k) All the a priori mea-sures of nancing constraints used in the literature can bethought of as different proxies for k (which is unobservable) or ofW (the availability of internal funds)

Each rm then chooses I to maximize

(1) max ( ) ( ) F I C E k I I W E such that + - - =

To guarantee that the above program is well behaved we alsoassume that C() is convex in E2

The rst-order condition of problem (1) then is given by

(2) F I C I W k1 11( ) ( ) + = -

where C1(0) represents the partial derivative of C with respect toits rst argument and F1() the rst derivative of F with respectto I The effects of the availability of internal nance on invest-ments can be easily obtained by implicit differentiation of (2)

(3)dI

dWC

C F

11

11

=- 11

which is clearly positive (to the extent that C is convex) In otherwords in an imperfect capital market world investments aresensitive to internal funds while in a perfect capital marketworld they are not (because C() 5 0 and thus C11 5 0)

Similarly it is possible to derive the sensitivity of investmentto the wedge between the cost of internal and external nancingBy implicit differentiation of (2) we obtain

2 This is a reasonable but not obvious assumption For example Calomirisand Himmelberg [1995] document that the average transaction cost of issuingsecurities decreases in the amount raised which suggests that C() may be con-cave While these transaction costs may be only a small component of the overallcost C() we note that this basic assumption might not be warranted

QUARTERLY JOURNAL OF ECONOMICS174

(4)dIdk

CC F

12

11

=-

- 11

which is negative if the marginal cost of raising external nanceis increasing in k (ie C12 0)

Most papers in this literature however do not test either ofthese two propositions On the one hand the estimatedinvestment-cash ow sensitivity is generally positive and signi-cant for all rms suggesting that all rms are constrained insome sense and so making the test of the rst implication redun-dant Second most of the proxies for W or k used in the literatureare only able to identify constrained rms not constrained rm-years This makes it impossible to disentangle the effect of -nancing constraints from a rm-specic effect on the level of in-vestment

For these reasons previous papers focus on cross-sectionaldifferences in the investment-cash-ow sensitivity across groupsof rms likely to have a different wedge between internal andexternal funds But this corresponds to looking at differences indIdW as a function of W or k Such an exercise is meaningfulonly if the investment-cash ow sensitivity is monotonically de-creasing with respect to W (or increasing with respect to k) inother words only if d2IdW2 is negative (or d2IdWdk is positive)From equation (3) we obtain

(5)d IdW

F C C FC F

2

2111 11

2112

113

111

11

=--( )

If both C11() and F11() are different from zero we can rewrite equa-tion (5) as

(6) d IdW

FF

CC

F CC F

2

2111

112

111

112

112

112

113

11

= -aelig

egraveccedil

ouml

oslashdivide -( )

Given that the second term is always positive it follows that d2IdW2 is negative if and only if [F111F11

2 2 C111C112] is negative

This condition implies a certain relationship between the curva-ture of the production function and the curvature of the cost func-tion at the optimal level of investment It is easy to see how sucha condition can be violated For example if the cost function isquadratic d2IdW 2 will be positive if the third derivative of theproduction function is positive (as is the case with a simple pro-duction function like I r where 0 r 1) In such a case theinvestment-cash ow sensitivity increases with a rmrsquos internal

INVESTMENT-CASH FLOW SENSITIVITIES 175

liquidity Of course many simple production functions have posi-tive third derivatives Although we will not produce them herethe conditions necessary to ensure that d2IdWdk be positive areat least as demanding

In sum even in a one-period model investment-cash owsensitivities do not necessarily increase with the degree of nan-cing constraints In a multiperiod model precautionary savingsmotives make it even more difcult to assess the theoretical rela-tionship between investment-cash ow sensitivities and the de-gree of nancing constraints For example Gross [1995] buildsand simulates an intertemporal investment model and nds anonmonotonic relationship between investment-cash ow sensi-tivities and the extent of nancing constraints

Finally the relationship between investment-cash ow sen-sitivities and degree of nancing constraints can be further com-plicated by the presence of irrational or overly risk-averse man-agers who choose to rely primarily on internal cash ow to investdespite the availability of low cost funds

II SAMPLE

In this paper we analyze the sample of 49 low-dividend pay-ing rms in FHP [1988] FHP divide all manufacturing rms inthe Value Line database with uninterrupted data from 1970 to1984 into three classes based on dividend payout policy Their 49Class 1 rms (which we analyze) have a dividend payout ratio ofless than 10 percent in at least ten of the fteen years FHP clas-sify 39 rms that have a dividend payout ratio between 10 per-cent and 20 percent as Class 2 rms and all 334 other rms intheir sample as Class 3 rms FHP argue that the Class 1 rmsare more likely a priori to have been nancially constrained Intheir analysis they nd that the Class 1 rms have aninvestment-cash ow sensitivity that is signicantly greater thanthat for rms that pay higher dividends

We choose this sample for three reasons First these rmsexhibit a strong relation between investment and cash ow Sec-ond FHP argue strongly that these rms are nancially con-strained most likely because of information problems BecauseFHP [1988] can legitimately be considered the parent of all pa-pers in this literature there can be no disagreement that we haveadversely selected our sample Finally given the high cost of ourresearch design the number of rms is manageable

We follow this sample for the same fteen years 1970 to

QUARTERLY JOURNAL OF ECONOMICS176

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

strained A rm is considered more nancially constrained as thewedge between its internal and external cost of funds increases

Our classication scheme which we detail below is designedto distinguish the relative differences in the degree to which rmsare nancially constrained In general our unconstrained or lessconstrained rms are those rms with relatively large amountsof liquid assets and net worth

In classifying rms we are agnostic on whether the wedgebetween the cost of internal and external funds is caused by hid-den information problems as in Myers and Majluf [1984] andGreenwald Stiglitz and Weiss [1984] or agency problems as inJensen and Meckling [1976] Grossman and Hart [1982] Jensen[1986] Stulz [1990] and Hart and Moore [1995] In fact unlikeBlanchard Lopez-de-Silanes and Shleifer [1994] the purpose ofour analysis is not to identify the source of the capital marketimperfection but rather to understand the effects capital marketimperfections have on investment We next review what economictheory has to say about the impact of nancing constraints oninvestment

B The Impact of Financing Constraints on Investments

FHP [1988] was the rst of many papers to consider higherinvestment-cash ow sensitivities as evidence of greater nancingconstraints Given the magnitude and the importance of thisliterature it is surprising that little attention has been given tothe theoretical foundation of the investment-cash ow sensitivitycriterion1 While it is easy to show that constrained rms shouldbe sensitive to internal cash ow while unconstrained rmsshould not it is not necessarily true that the magnitude of thesensitivity increases in the degree of nancing constraints Thisis the crucial question given that investment is sensitive to cashow for the vast majority of rms analyzed (It is easy to justifythis sensitivity based on the fact that external funds are morecostly than internal funds for all rms as long as some trans-action costs are involved)

The difculty of interpreting cross-sectional differences ininvestment-cash ow sensitivities can be illustrated with asimple one-period model Consider a rm that chooses the levelof investment to maximize prots The return to an investmentI is given by a production function F(I ) where F 9 and F 0 0

1 We thank Jeremy Stein for encouraging us to develop this point

INVESTMENT-CASH FLOW SENSITIVITIES 173

Investment can be nanced either with internal funds (W) or withexternal funds (E) The opportunity cost of internal funds equalsthe cost of capital R which for simplicity we set equal to 1 Be-cause of information agency or risk aversion problems we as-sume that the use of external funds generates a deadweight costwhichmdashin a competitive capital marketmdashis borne by the issuingrm We represent (in reduced form) this additional cost of exter-nal funds with the function C(Ek) where E is the amount of ex-ternal funds raised and k is a measure of a rmrsquos wedge betweenthe internal and the external costs of funds It is natural to as-sume that the total cost of raising external funds increases inthe amount of funds raised and in the extent of the agency orinformation problems (represented by k) All the a priori mea-sures of nancing constraints used in the literature can bethought of as different proxies for k (which is unobservable) or ofW (the availability of internal funds)

Each rm then chooses I to maximize

(1) max ( ) ( ) F I C E k I I W E such that + - - =

To guarantee that the above program is well behaved we alsoassume that C() is convex in E2

The rst-order condition of problem (1) then is given by

(2) F I C I W k1 11( ) ( ) + = -

where C1(0) represents the partial derivative of C with respect toits rst argument and F1() the rst derivative of F with respectto I The effects of the availability of internal nance on invest-ments can be easily obtained by implicit differentiation of (2)

(3)dI

dWC

C F

11

11

=- 11

which is clearly positive (to the extent that C is convex) In otherwords in an imperfect capital market world investments aresensitive to internal funds while in a perfect capital marketworld they are not (because C() 5 0 and thus C11 5 0)

Similarly it is possible to derive the sensitivity of investmentto the wedge between the cost of internal and external nancingBy implicit differentiation of (2) we obtain

2 This is a reasonable but not obvious assumption For example Calomirisand Himmelberg [1995] document that the average transaction cost of issuingsecurities decreases in the amount raised which suggests that C() may be con-cave While these transaction costs may be only a small component of the overallcost C() we note that this basic assumption might not be warranted

QUARTERLY JOURNAL OF ECONOMICS174

(4)dIdk

CC F

12

11

=-

- 11

which is negative if the marginal cost of raising external nanceis increasing in k (ie C12 0)

Most papers in this literature however do not test either ofthese two propositions On the one hand the estimatedinvestment-cash ow sensitivity is generally positive and signi-cant for all rms suggesting that all rms are constrained insome sense and so making the test of the rst implication redun-dant Second most of the proxies for W or k used in the literatureare only able to identify constrained rms not constrained rm-years This makes it impossible to disentangle the effect of -nancing constraints from a rm-specic effect on the level of in-vestment

For these reasons previous papers focus on cross-sectionaldifferences in the investment-cash-ow sensitivity across groupsof rms likely to have a different wedge between internal andexternal funds But this corresponds to looking at differences indIdW as a function of W or k Such an exercise is meaningfulonly if the investment-cash ow sensitivity is monotonically de-creasing with respect to W (or increasing with respect to k) inother words only if d2IdW2 is negative (or d2IdWdk is positive)From equation (3) we obtain

(5)d IdW

F C C FC F

2

2111 11

2112

113

111

11

=--( )

If both C11() and F11() are different from zero we can rewrite equa-tion (5) as

(6) d IdW

FF

CC

F CC F

2

2111

112

111

112

112

112

113

11

= -aelig

egraveccedil

ouml

oslashdivide -( )

Given that the second term is always positive it follows that d2IdW2 is negative if and only if [F111F11

2 2 C111C112] is negative

This condition implies a certain relationship between the curva-ture of the production function and the curvature of the cost func-tion at the optimal level of investment It is easy to see how sucha condition can be violated For example if the cost function isquadratic d2IdW 2 will be positive if the third derivative of theproduction function is positive (as is the case with a simple pro-duction function like I r where 0 r 1) In such a case theinvestment-cash ow sensitivity increases with a rmrsquos internal

INVESTMENT-CASH FLOW SENSITIVITIES 175

liquidity Of course many simple production functions have posi-tive third derivatives Although we will not produce them herethe conditions necessary to ensure that d2IdWdk be positive areat least as demanding

In sum even in a one-period model investment-cash owsensitivities do not necessarily increase with the degree of nan-cing constraints In a multiperiod model precautionary savingsmotives make it even more difcult to assess the theoretical rela-tionship between investment-cash ow sensitivities and the de-gree of nancing constraints For example Gross [1995] buildsand simulates an intertemporal investment model and nds anonmonotonic relationship between investment-cash ow sensi-tivities and the extent of nancing constraints

Finally the relationship between investment-cash ow sen-sitivities and degree of nancing constraints can be further com-plicated by the presence of irrational or overly risk-averse man-agers who choose to rely primarily on internal cash ow to investdespite the availability of low cost funds

II SAMPLE

In this paper we analyze the sample of 49 low-dividend pay-ing rms in FHP [1988] FHP divide all manufacturing rms inthe Value Line database with uninterrupted data from 1970 to1984 into three classes based on dividend payout policy Their 49Class 1 rms (which we analyze) have a dividend payout ratio ofless than 10 percent in at least ten of the fteen years FHP clas-sify 39 rms that have a dividend payout ratio between 10 per-cent and 20 percent as Class 2 rms and all 334 other rms intheir sample as Class 3 rms FHP argue that the Class 1 rmsare more likely a priori to have been nancially constrained Intheir analysis they nd that the Class 1 rms have aninvestment-cash ow sensitivity that is signicantly greater thanthat for rms that pay higher dividends

We choose this sample for three reasons First these rmsexhibit a strong relation between investment and cash ow Sec-ond FHP argue strongly that these rms are nancially con-strained most likely because of information problems BecauseFHP [1988] can legitimately be considered the parent of all pa-pers in this literature there can be no disagreement that we haveadversely selected our sample Finally given the high cost of ourresearch design the number of rms is manageable

We follow this sample for the same fteen years 1970 to

QUARTERLY JOURNAL OF ECONOMICS176

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

Investment can be nanced either with internal funds (W) or withexternal funds (E) The opportunity cost of internal funds equalsthe cost of capital R which for simplicity we set equal to 1 Be-cause of information agency or risk aversion problems we as-sume that the use of external funds generates a deadweight costwhichmdashin a competitive capital marketmdashis borne by the issuingrm We represent (in reduced form) this additional cost of exter-nal funds with the function C(Ek) where E is the amount of ex-ternal funds raised and k is a measure of a rmrsquos wedge betweenthe internal and the external costs of funds It is natural to as-sume that the total cost of raising external funds increases inthe amount of funds raised and in the extent of the agency orinformation problems (represented by k) All the a priori mea-sures of nancing constraints used in the literature can bethought of as different proxies for k (which is unobservable) or ofW (the availability of internal funds)

Each rm then chooses I to maximize

(1) max ( ) ( ) F I C E k I I W E such that + - - =

To guarantee that the above program is well behaved we alsoassume that C() is convex in E2

The rst-order condition of problem (1) then is given by

(2) F I C I W k1 11( ) ( ) + = -

where C1(0) represents the partial derivative of C with respect toits rst argument and F1() the rst derivative of F with respectto I The effects of the availability of internal nance on invest-ments can be easily obtained by implicit differentiation of (2)

(3)dI

dWC

C F

11

11

=- 11

which is clearly positive (to the extent that C is convex) In otherwords in an imperfect capital market world investments aresensitive to internal funds while in a perfect capital marketworld they are not (because C() 5 0 and thus C11 5 0)

Similarly it is possible to derive the sensitivity of investmentto the wedge between the cost of internal and external nancingBy implicit differentiation of (2) we obtain

2 This is a reasonable but not obvious assumption For example Calomirisand Himmelberg [1995] document that the average transaction cost of issuingsecurities decreases in the amount raised which suggests that C() may be con-cave While these transaction costs may be only a small component of the overallcost C() we note that this basic assumption might not be warranted

QUARTERLY JOURNAL OF ECONOMICS174

(4)dIdk

CC F

12

11

=-

- 11

which is negative if the marginal cost of raising external nanceis increasing in k (ie C12 0)

Most papers in this literature however do not test either ofthese two propositions On the one hand the estimatedinvestment-cash ow sensitivity is generally positive and signi-cant for all rms suggesting that all rms are constrained insome sense and so making the test of the rst implication redun-dant Second most of the proxies for W or k used in the literatureare only able to identify constrained rms not constrained rm-years This makes it impossible to disentangle the effect of -nancing constraints from a rm-specic effect on the level of in-vestment

For these reasons previous papers focus on cross-sectionaldifferences in the investment-cash-ow sensitivity across groupsof rms likely to have a different wedge between internal andexternal funds But this corresponds to looking at differences indIdW as a function of W or k Such an exercise is meaningfulonly if the investment-cash ow sensitivity is monotonically de-creasing with respect to W (or increasing with respect to k) inother words only if d2IdW2 is negative (or d2IdWdk is positive)From equation (3) we obtain

(5)d IdW

F C C FC F

2

2111 11

2112

113

111

11

=--( )

If both C11() and F11() are different from zero we can rewrite equa-tion (5) as

(6) d IdW

FF

CC

F CC F

2

2111

112

111

112

112

112

113

11

= -aelig

egraveccedil

ouml

oslashdivide -( )

Given that the second term is always positive it follows that d2IdW2 is negative if and only if [F111F11

2 2 C111C112] is negative

This condition implies a certain relationship between the curva-ture of the production function and the curvature of the cost func-tion at the optimal level of investment It is easy to see how sucha condition can be violated For example if the cost function isquadratic d2IdW 2 will be positive if the third derivative of theproduction function is positive (as is the case with a simple pro-duction function like I r where 0 r 1) In such a case theinvestment-cash ow sensitivity increases with a rmrsquos internal

INVESTMENT-CASH FLOW SENSITIVITIES 175

liquidity Of course many simple production functions have posi-tive third derivatives Although we will not produce them herethe conditions necessary to ensure that d2IdWdk be positive areat least as demanding

In sum even in a one-period model investment-cash owsensitivities do not necessarily increase with the degree of nan-cing constraints In a multiperiod model precautionary savingsmotives make it even more difcult to assess the theoretical rela-tionship between investment-cash ow sensitivities and the de-gree of nancing constraints For example Gross [1995] buildsand simulates an intertemporal investment model and nds anonmonotonic relationship between investment-cash ow sensi-tivities and the extent of nancing constraints

Finally the relationship between investment-cash ow sen-sitivities and degree of nancing constraints can be further com-plicated by the presence of irrational or overly risk-averse man-agers who choose to rely primarily on internal cash ow to investdespite the availability of low cost funds

II SAMPLE

In this paper we analyze the sample of 49 low-dividend pay-ing rms in FHP [1988] FHP divide all manufacturing rms inthe Value Line database with uninterrupted data from 1970 to1984 into three classes based on dividend payout policy Their 49Class 1 rms (which we analyze) have a dividend payout ratio ofless than 10 percent in at least ten of the fteen years FHP clas-sify 39 rms that have a dividend payout ratio between 10 per-cent and 20 percent as Class 2 rms and all 334 other rms intheir sample as Class 3 rms FHP argue that the Class 1 rmsare more likely a priori to have been nancially constrained Intheir analysis they nd that the Class 1 rms have aninvestment-cash ow sensitivity that is signicantly greater thanthat for rms that pay higher dividends

We choose this sample for three reasons First these rmsexhibit a strong relation between investment and cash ow Sec-ond FHP argue strongly that these rms are nancially con-strained most likely because of information problems BecauseFHP [1988] can legitimately be considered the parent of all pa-pers in this literature there can be no disagreement that we haveadversely selected our sample Finally given the high cost of ourresearch design the number of rms is manageable

We follow this sample for the same fteen years 1970 to

QUARTERLY JOURNAL OF ECONOMICS176

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

(4)dIdk

CC F

12

11

=-

- 11

which is negative if the marginal cost of raising external nanceis increasing in k (ie C12 0)

Most papers in this literature however do not test either ofthese two propositions On the one hand the estimatedinvestment-cash ow sensitivity is generally positive and signi-cant for all rms suggesting that all rms are constrained insome sense and so making the test of the rst implication redun-dant Second most of the proxies for W or k used in the literatureare only able to identify constrained rms not constrained rm-years This makes it impossible to disentangle the effect of -nancing constraints from a rm-specic effect on the level of in-vestment

For these reasons previous papers focus on cross-sectionaldifferences in the investment-cash-ow sensitivity across groupsof rms likely to have a different wedge between internal andexternal funds But this corresponds to looking at differences indIdW as a function of W or k Such an exercise is meaningfulonly if the investment-cash ow sensitivity is monotonically de-creasing with respect to W (or increasing with respect to k) inother words only if d2IdW2 is negative (or d2IdWdk is positive)From equation (3) we obtain

(5)d IdW

F C C FC F

2

2111 11

2112

113

111

11

=--( )

If both C11() and F11() are different from zero we can rewrite equa-tion (5) as

(6) d IdW

FF

CC

F CC F

2

2111

112

111

112

112

112

113

11

= -aelig

egraveccedil

ouml

oslashdivide -( )

Given that the second term is always positive it follows that d2IdW2 is negative if and only if [F111F11

2 2 C111C112] is negative

This condition implies a certain relationship between the curva-ture of the production function and the curvature of the cost func-tion at the optimal level of investment It is easy to see how sucha condition can be violated For example if the cost function isquadratic d2IdW 2 will be positive if the third derivative of theproduction function is positive (as is the case with a simple pro-duction function like I r where 0 r 1) In such a case theinvestment-cash ow sensitivity increases with a rmrsquos internal

INVESTMENT-CASH FLOW SENSITIVITIES 175

liquidity Of course many simple production functions have posi-tive third derivatives Although we will not produce them herethe conditions necessary to ensure that d2IdWdk be positive areat least as demanding

In sum even in a one-period model investment-cash owsensitivities do not necessarily increase with the degree of nan-cing constraints In a multiperiod model precautionary savingsmotives make it even more difcult to assess the theoretical rela-tionship between investment-cash ow sensitivities and the de-gree of nancing constraints For example Gross [1995] buildsand simulates an intertemporal investment model and nds anonmonotonic relationship between investment-cash ow sensi-tivities and the extent of nancing constraints

Finally the relationship between investment-cash ow sen-sitivities and degree of nancing constraints can be further com-plicated by the presence of irrational or overly risk-averse man-agers who choose to rely primarily on internal cash ow to investdespite the availability of low cost funds

II SAMPLE

In this paper we analyze the sample of 49 low-dividend pay-ing rms in FHP [1988] FHP divide all manufacturing rms inthe Value Line database with uninterrupted data from 1970 to1984 into three classes based on dividend payout policy Their 49Class 1 rms (which we analyze) have a dividend payout ratio ofless than 10 percent in at least ten of the fteen years FHP clas-sify 39 rms that have a dividend payout ratio between 10 per-cent and 20 percent as Class 2 rms and all 334 other rms intheir sample as Class 3 rms FHP argue that the Class 1 rmsare more likely a priori to have been nancially constrained Intheir analysis they nd that the Class 1 rms have aninvestment-cash ow sensitivity that is signicantly greater thanthat for rms that pay higher dividends

We choose this sample for three reasons First these rmsexhibit a strong relation between investment and cash ow Sec-ond FHP argue strongly that these rms are nancially con-strained most likely because of information problems BecauseFHP [1988] can legitimately be considered the parent of all pa-pers in this literature there can be no disagreement that we haveadversely selected our sample Finally given the high cost of ourresearch design the number of rms is manageable

We follow this sample for the same fteen years 1970 to

QUARTERLY JOURNAL OF ECONOMICS176

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

liquidity Of course many simple production functions have posi-tive third derivatives Although we will not produce them herethe conditions necessary to ensure that d2IdWdk be positive areat least as demanding

In sum even in a one-period model investment-cash owsensitivities do not necessarily increase with the degree of nan-cing constraints In a multiperiod model precautionary savingsmotives make it even more difcult to assess the theoretical rela-tionship between investment-cash ow sensitivities and the de-gree of nancing constraints For example Gross [1995] buildsand simulates an intertemporal investment model and nds anonmonotonic relationship between investment-cash ow sensi-tivities and the extent of nancing constraints

Finally the relationship between investment-cash ow sen-sitivities and degree of nancing constraints can be further com-plicated by the presence of irrational or overly risk-averse man-agers who choose to rely primarily on internal cash ow to investdespite the availability of low cost funds

II SAMPLE

In this paper we analyze the sample of 49 low-dividend pay-ing rms in FHP [1988] FHP divide all manufacturing rms inthe Value Line database with uninterrupted data from 1970 to1984 into three classes based on dividend payout policy Their 49Class 1 rms (which we analyze) have a dividend payout ratio ofless than 10 percent in at least ten of the fteen years FHP clas-sify 39 rms that have a dividend payout ratio between 10 per-cent and 20 percent as Class 2 rms and all 334 other rms intheir sample as Class 3 rms FHP argue that the Class 1 rmsare more likely a priori to have been nancially constrained Intheir analysis they nd that the Class 1 rms have aninvestment-cash ow sensitivity that is signicantly greater thanthat for rms that pay higher dividends

We choose this sample for three reasons First these rmsexhibit a strong relation between investment and cash ow Sec-ond FHP argue strongly that these rms are nancially con-strained most likely because of information problems BecauseFHP [1988] can legitimately be considered the parent of all pa-pers in this literature there can be no disagreement that we haveadversely selected our sample Finally given the high cost of ourresearch design the number of rms is manageable

We follow this sample for the same fteen years 1970 to

QUARTERLY JOURNAL OF ECONOMICS176

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

1984 studied by FHP [1988] For each rm we collected datafrom several sources First we collected letters to shareholdersmanagement discussions of operations and liquidity (when avail-able) nancial statements and the notes to those statementsfrom the annual report or 10-K for each rm-year We obtainedWall Street Journal Index entries over the fteen-year sample pe-riod3 We obtained standard accounting variables from COMPU-STAT except those for Coleco which we obtained from Colecorsquosannual reports Because FHP obtained their data from ValueLine not COMPUSTAT our data are not precisely the same astheirs

We measure investment or capital expenditures usingCOMPUSTAT item 128 We measure cash ow as the sum ofearnings before extraordinary items (item 18) and depreciation(item 14) We deate investment and cash ow by capital whichwe measure as net property plant and equipment (item 8) at thebeginning of the scal year This measure of capital differsslightly from the replacement cost measure employed by FHP

We measure average Tobinrsquos Q as the market value of assetsdivided by the book value of assets (item 6) where the marketvalue of assets equals the book value of assets plus the marketvalue of common equity less the sum of the book value of commonequity (item 60) and balance sheet deferred taxes (item 74) Asdo most papers in this literature we calculate Q at the beginningof a rmrsquos scal year4 (Our results are similar when we use end-of-period Q)

In Table I we compare the basic regression results for oursample with those reported in Table 4 of FHP [1988] These re-gressions regress investment on cash ow and Q and control forxed rm and year effects Our results are qualitatively similarto those reported by FHP although they differ slightly in somedetails5 For each of the three time periods our coefcients oncash ow are lower than those reported by FHP Those differ-

3 Fiscal years ending before June 15 are assigned to the previous calendaryear scal years ending after June 15 are assigned to the current calendar year

4 Our measure differs from FHPrsquos in two ways First FHP compute Q basedon replacement costs while we simply use a market-to-book ratio The results inPerfect and Wiles [1994] indicate that the improvements obtained from the moreinvolved computation of Q are fairly limited particularly when regressions areestimated with rm xed effects Second FHP use the average market value ofequity in the fourth quarter while we use the actual market value of equity atscal year end

5 We use 719 observations not 735 because rms switched scal years(three rm-years) rms did not le nancial statements with the SEC (six rm-years) and rms did not have an available stock price (seven rm-years) FHP[1988] do not report how many observations they include

INVESTMENT-CASH FLOW SENSITIVITIES 177

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

TAB

LE

IC

OM

PA

RIS

ON

OF

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

WIT

HFA

ZZ

AR

IH

UB

BA

RD

AN

DP

ET

ER

SE

NR

ES

UL

TS

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

Faz

zari

H

ubba

rd

and

Pet

erse

n[1

988]

(h

erei

naft

erF

HP

[198

8])

from

1970

to19

84co

mpa

red

wit

hes

tim

ates

inF

HP

KZ

refe

rsto

our

esti

mat

esI

nves

tmen

tis

capi

tale

xpen

ditu

res

(CO

M-

PU

STA

Tit

em12

8)C

ash

ow

equ

als

the

sum

ofea

rnin

gsbe

fore

extr

aord

inar

yit

ems

(CO

MP

US

TA

Tit

em18

)and

depr

ecia

tion

(CO

MP

US

-TA

Tit

em14

)In

vest

men

tan

dca

sh

owar

ede

ate

dby

begi

nnin

gof

year

capi

tal

(Kt2

1)

wh

ich

we

de

neas

net

prop

erty

pl

ant

and

equi

pmen

t(C

OM

PU

STA

Tit

em8)

Qeq

ual

sth

em

arke

tva

lue

ofas

sets

divi

ded

byth

ebo

okva

lue

ofas

sets

(CO

MP

US

TAT

item

6)M

arke

tva

lue

ofas

sets

equa

lsth

ebo

okva

lue

ofas

sets

plu

sth

em

arke

tva

lue

ofco

mm

onst

ock

less

the

sum

ofth

ebo

okva

lue

ofco

mm

onst

ock

(CO

MP

US

TA

Tit

em6)

and

bala

nce

shee

tde

ferr

edta

xes

(CO

MP

US

TAT

item

74)

All

regr

essi

ons

incl

ude

rm

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

KZ

KZ

FH

PK

ZK

ZF

HP

KZ

KZ

FH

P19

70ndash8

419

70ndash8

419

70ndash8

419

70ndash7

919

70ndash7

919

70ndash7

919

70ndash7

519

70ndash7

519

70ndash7

5

CF

tK

t21

039

50

500

046

10

477

057

80

540

055

80

634

067

0[0

026

][0

023

][0

027

][0

035

][0

030

][0

036

][0

040

][0

034

][0

044

]Q

t21

003

90

0008

003

00

0002

002

12

000

10[0

005

][0

000

4][0

006

][0

000

4][0

006

][0

000

4]A

djR

20

584

054

80

460

649

062

70

470

764

075

30

55N

obs

719

719

NA

47

647

6N

A

280

280

NA

QUARTERLY JOURNAL OF ECONOMICS178

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

ences however appear to be only marginally signicant if at allAt the same time our coefcients for Q are signicantly greaterthan those reported by FHP

We attribute the differences between our results and FHPrsquosto the different denitions of Q When we exclude Q from ourregressions we obtain coefcients on cash ow that exceed thosein FHP except for the 1970ndash1975 period where our coefcient isinsignicantly smaller Because the FHP measure is constructedwith an average stock price in the previous year rather than the(more appropriate) stock price at the beginning of the year wesuspect that our measure of Q provides better information aboutinvestment opportunities The FHP measure will not distinguishbetween a rm whose stock price declines from 20 to 10 and arm whose stock price increases from 10 to 20 at the end of theprevious year6

III CLASSIFICATION SCHEME

A Description

The SEC requires companies listed on a stock exchange thathave more than 500 shareholders and $5 million in assets to lean annual report or 10-K that contains the basic nancial state-ments and their notes as well as all material information regard-ing a companyrsquos business and nancial condition The annualreports are generally introduced by a letter to shareholders fromthe chief executive ofcer (CEO) This letter usually describes themajor events of the previous scal year and the major projectsplanned for the future

In 1977 the SEC strengthened these reporting requirementsby adopting Regulation S-K which requires rms to discuss ex-plicitly their liquidity capital resources and results of opera-tions This section is usually titled managementrsquos discussion ofoperations Item 303 of Regulation S-K states

(1) Liquidity Identify any known trends or any known demandscommitments events or uncertainties that will result in theregistrantrsquos liquidity increasing or decreasing in any material way

6 FHP [1996] question our measure of Q as a possible source of error becausewe use book value rather than replacement value of assets This concern is un-founded for two reasons First our measure of Q explains more variation in in-vestment than the measure used by FHP suggesting that their measure is noisierthan ours Second as we show below we obtain similar results using an Eulerequation approach that does not rely on Q

INVESTMENT-CASH FLOW SENSITIVITIES 179

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

If a material deciency is identied indicate the course of actionthat the registrant has taken or proposes to take to remedy thedeciency Also identify and separately describe internal and exter-nal sources of liquidity and briey discuss any material unusedsources of liquid assets

(2) Capital Resources (i) Describe the registrantrsquos material commit-ments for capital expenditures as of the end of the latest scal pe-riod and indicate the general purpose of such commitments and theanticipated source of funds needed to fulll such commitments (ii) Describe any known material trends favorable or unfavorablein the registrantrsquos capital resources Indicate any expected materialchanges in the mix and the relative cost of such resources

Instructions 5 The term ldquoliquidityrdquo refers to the ability of anenterprise to generate adequate amounts of cash to meet the enter-prisersquos needs for cash Liquidity shall generally be discussed onboth a long-term and short-term basis7

In short Regulation S-K explicitly requires rms to disclosewhether or not they are having difculty in nancing their in-vestments Consistent with the timing of the new SEC regula-tions post-1977 annual report information for our sample rmstends to be more detailed than the information for earlier yearsTo the extent that our classication scheme has errors theyshould be smaller for years after 1977

We use the qualitative information in the annual reports to-gether with quantitative information in the companiesrsquo nancialstatements and notes to classify each rm-year into one of vegroups

The rst group contains rms that we deem denitely notnancially constrained in that year We refer to these rm-yearsas not nancially constrained (NFC) We place a rm-year in theNFC group if the rm initiated or increased cash dividends re-purchased stock or explicitly indicated in its annual report thatthe rm had more liquidity than it would need for investment inthe foreseeable future8 We also were more likely to label a rm-year NFC if the rm had a large cash position (relative to invest-ment) or if the rmrsquos lenders did not restrict the rm from making

7 See SEC 63031 in Murray Decker and Dittmar [1993]8 For example Plantronicsrsquo 1971 annual report states ldquoWe ended the year

in an exceptionally strong nancial condition for a company of our size Duringthe year we paid off all long-term debt and our cash and cash-equivalent assetshave throughout the year exceeded all current liabilitiesrdquo

QUARTERLY JOURNAL OF ECONOMICS180

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

large dividend payments (relative to investment) NFC rm-years therefore tend to include nancially healthy companieswith low debt and high cash In NFC rm-years therefore wend no evidence that the rms could not have invested appreciablymore if their managers had so chosen In NFC rm-years rmsalso have large amounts of internal funds and collateralizable re-sources relative to the amount of funds required

The second group includes rm-years that we label likely notto be nancially constrained (LNFC) In LNFC rm-years therms are healthy nancially and do not give any indication ofbeing liquidity constrained These rms also tend to have sizablecash reserves unused lines of credit and healthy interest cover-age We distinguish LNFC rm-years from NFC rm-years by themagnitude of the liquidity measures and by the absence of anexplicit statement of excess liquidity Again in LNFC rm-yearswe nd no evidence that these rms could not have invested moreif their managers had so chosen For example despite the quotein our introduction we classify Hewlett-Packard as LNFC in verm-years in the 1970s

The third group includes rm-years we found difcult to clas-sify either as nancially constrained or as unconstrained We callthese rm-years possibly nancially constrained (PFC) In PFCrm-years rms do not report any clear signs of nancing con-straints but they do not look particularly liquid either Fre-quently these rms face an adverse product market environmentbut are not explicitly strapped for cash This category also in-cludes rm-years that provide contradictory indications of theirnancial situation For example this might include a companythat increases its dividend but laments its lack of nancial re-sources in the letter to shareholders

The fourth group contains all rm-years in which rms arelikely to be nancially constrained (LFC) This group includesrms that mention having difculties in obtaining nancing Forexample we include rm-years in which rms postpone an equityor convertible debt offering due to adverse market conditions orclaim they need equity capital but are waiting for improved mar-ket conditions Generally these rms are prevented from payingdividends and have little cash available Firms that cut dividendsalso are more likely to fall in this category unless other adversefactors assign them to the fth group

The last group includes all rm-years in which rms are un-doubtedly nancially constrained (FC) In these rm-years these

INVESTMENT-CASH FLOW SENSITIVITIES 181

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

companies are in violation of debt covenants have been cut outof their usual source of credit are renegotiating debt paymentsor declare that they are forced to reduce investments because ofliquidity problems

Our classication scheme is subject to the criticism thatmanagers do not always report truthfully and therefore somerm-years will be misclassied We do not view management mis-reporting as a serious problem for several reasons First manag-ers are held liable not only for disclosing false information butalso for not disclosing material information This is particularlytrue after 1977 when Regulation S-K is in effect9 Second we readannual reports over a fteen-year period While a rm may beable to misreport in any given rm-year it seems unlikely that arm can misreport every year Third we do not rely exclusivelyon the management discussions but also read the nancial state-ments carefully Finally any management reluctance to reportnegative information should bias our results against nding -nancially constrained companies and differences across groupsTo the extent that we nd some companies to be nancially con-strained we can be certain that they are indeed constrained

Overall our classication scheme captures relative differ-ences in sample rmsrsquo availability of internal and external fundsin a given year The nancial statements and management dis-cussions strongly indicate that NFC (and LNFC) rms could haveinvested more (often substantially more) in that year had they sochosen In the language of our model these are rms for whomW is very high even after they invest Therefore these rmsshould face a C(Ek) that is close to 0 if not equal to 0 This isunlikely to be true for the PFC rms and denitely not true forthe LFC and FC rms Our classication scheme therefore cap-tures relative differences in sample rmsrsquo wedge between exter-nal and internal nance

B Classication Results

Table II summarizes our classication of rm-years We clas-sify 545 percent of rm-years as not (NFC) and 309 percent ofrm-years as likely not nancially constrained (LNFC) for a totalof 853 percent of rm-years in which we nd no evidence of nan-cing constraints that restrict investment We classify 73 percent

9 For example the SEC took action against Caterpillar Inc for not re-porting that a large increase in Caterpillarrsquos 1989 net income was caused by ahyperination in Brazil

QUARTERLY JOURNAL OF ECONOMICS182

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

INVESTMENT-CASH FLOW SENSITIVITIES 183

TA

BL

EII

SU

MM

AR

YO

FA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

S

Dis

trib

utio

nof

nan

cin

gco

nstr

aint

sby

year

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]

from

1970

to19

84

Fir

mn

anci

ngco

nstr

ain

tst

atu

sfo

rea

chye

arar

eno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyno

t

nanc

ially

cons

trai

ned

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)

NF

C1

LN

FC

PF

C1

LF

C1

FC

NF

CL

NF

CP

FC

LF

CF

CN

otor

Pos

sibl

ylik

ely

orN

otL

ikel

yno

tP

ossi

bly

Lik

ely

De

nite

lyli

kely

not

den

itel

y

nan

cial

ly

nan

cial

lyn

anci

ally

nan

cial

lyn

anci

ally

na

ncia

llyn

anci

ally

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

cons

trai

ned

con

stra

ined

1970

340

44

7

149

2

14

378

7

213

19

7138

334

017

010

70

072

327

719

7243

835

412

58

30

079

220

819

7339

645

86

34

24

285

414

619

7436

728

612

216

36

165

334

719

7530

642

914

38

24

173

526

519

7651

038

82

04

14

189

810

219

7759

228

64

10

08

287

812

219

7867

326

52

02

02

093

86

219

7961

226

510

22

00

087

812

219

8073

520

44

12

00

093

96

119

8171

420

46

10

02

091

88

219

8269

424

52

02

02

093

96

119

8369

424

52

04

10

093

96

119

8469

422

40

06

12

091

88

2

Tota

l54

530

97

34

82

685

314

7

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

of rm-years as possibly constrained 48 percent as likely con-strained and 26 percent as denitely constrained for a total ofonly 147 percent rm-years in which there is some possibility ofnancing constraints The fraction of rms that are at least pos-sibly constrained varies over time with more rms being poten-tially constrained in the early part of the sample (when theserms were smaller) and particularly around the 1974ndash1975 re-cession This time pattern is consistent with the results in FHP[1988] and in Table I that investment-cash ow sensitivities de-cline over the sample period (In the Appendix we report the year-by-year classications for all 49 rms)

We consider the accuracy of our classication scheme by re-porting quantitative measures of operational and nancial healthacross our ve classications in Table III In panel A mediancash ow net cash ow (cash ow less investment) and TobinrsquosQ decline monotonically across the ve categories For examplethe median level of net cash ow for NFC rms is 11 percent ofcapital (net property plant and equipment) while the medianlevel of net cash ow for FC rms is almost 2 20 percent Thissuggests that NFC rms could have increased their investmentwithout tapping external sources of capital

Panel A also suggests that our classication scheme is suc-cessful in capturing the degree of nancing constraints Equation(3) predicts that investment will decline as nancing constraintsincrease Consistent with this the median level of investment issignicantly lower for LFC and FC rm-years than for the otherthree groups (We test this more formally in subsection IVCwhere we control for investment opportunities) Furthermore themean level of investment in acquisitions (as a fraction of capital)is substantially higher for rms in the rst two groups (NFC andLNFC) than for rms in the other three groups10 Acquisitions arecompletely absent in FC rm-years

Panel B reports summary statistics on rm nancial statusInterest coveragemdashearnings before interest taxes depreciationand amortization or EBITDA (COMPUSTAT item 13) to interestexpense (item 15)mdashdeclines monotonically across our classica-tions11 Debt to total capital also decreases monotonically debt is

10 We calculate acquisitions as the value of businesses or companies ac-quired in a given rm-year as a fraction of beginning-of-year capital We valuepurchase acquisitions using information in the statement of changes We valuepooling acquisitions using the notes to nancial statements

11 We set interest coverage to 100 if coverage exceeds 100 or interest ex-pense is negative We set interest coverage to 0 if EBITDA is negative

QUARTERLY JOURNAL OF ECONOMICS184

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

TABLE IIISUMMARY STATISTICS FOR FIRM CHARACTERISTICS BY YEARLY FINANCING

CONSTRAINT STATUS

Distribution of nancial variables by annual nancing constraint status for49 low-dividend rms from FHP [1988] from 1970 to 1984 Firm nancing con-straint status for each year is not nancially constrained (NFC) likely not nan-cially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) Each entry re-ports the median mean tenth percentile ninetieth percentile and number ofobservations Investment (It) cash ow Q and capital (Kt 2 1) are dened in TableI Acquisitions (Acqs) equals the value of purchase and pooling acquisitions In-terest coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value short-term andlong-term debt Total capital is the sum of debt the book value of preferred stockand the book value of common equity Free divs is the amount of retained earn-ings that are not restricted from being paid out as dividends Cash is cash andmarketable securities Unused linet is the amount of unused line of credit at theend of year t Slack is the sum of cash and unused line Change in debt is thechange in sum of the book value of short-term and long-term debt Equity issueis the sum of the equity issued to the public and to acquisition targets

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

A Investment cash ow growth

It Kt 2 1 0368 0324 0359 0273 0243 03480461 0413 0450 0350 0313 04360159 0159 0122 0073 0068 01270831 0831 0824 0909 0544 0810

393 221 52 34 19 719

Cash FlowtKt 2 1 0506 0350 0313 0243 0020 04210614 0435 0366 0191 2 0047 05050209 0104 2 0125 2 0126 2 0436 01221075 0871 1084 0528 0366 1007

393 221 52 34 19 719

(Cash Flowt 2 It )Kt 2 1 0110 0026 2 0026 2 0071 2 0198 00510155 0022 2 0085 2 0159 2 0360 0069

2 0180 2 0316 2 0474 2 0642 2 0785 2 02850503 0323 0420 0141 2 0076 0442

393 221 52 34 19 719

Qt 1313 1171 1159 1096 1082 12311647 1542 1312 1527 1402 15800809 0755 0793 0734 0795 07852781 2799 1934 2659 1789 2749

393 221 52 34 19 719

Fraction rms with 0244 0244 0154 0176 0000 0228acquisitions in 393 221 52 34 19 719yeart

INVESTMENT-CASH FLOW SENSITIVITIES 185

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

AcqstKt 2 1 0000 0000 0000 0000 0000 00000122 0159 0063 0023 0000 01210000 0000 0000 0000 0000 00000287 0300 0044 0029 0000 0252

388 217 52 34 19 710

Sales growtht 0211 0150 0123 0136 0008 01800226 0165 0097 0113 0049 01880021 2 0071 2 0136 2 0145 2 0275 2 00510484 0385 0319 0338 0305 0452

393 221 52 34 19 719

Inventory growtht 0199 0117 0144 0063 2 0064 01540215 0160 0135 0049 2 0013 0179

2 0073 2 0175 2 0056 2 0499 2 0487 2 01350545 0475 0376 0562 0543 0512

393 221 52 34 19 719

B Financial policy

Interest coveraget 7971 5886 4203 2836 1093 640618026 11777 4745 3455 1650 140232746 1608 0000 0666 0000 1707

46722 23605 9598 6960 3827 33325393 221 52 34 19 719

Debtt to total capitalt 0296 0351 0431 0541 0565 03490293 0352 0454 0573 0621 03440051 0117 0258 0316 0361 00750526 0585 0689 0791 0912 0585

393 221 52 34 19 719

DividendstKt 0000 0000 0000 0000 0000 00000015 0006 0006 0002 0001 00110000 0000 0000 0000 0000 00000046 0023 0028 0028 0007 0037

393 221 52 34 19 719

Fraction of years 0061 0276 0462 0686 0789 0206dividends restricted 393 221 52 34 19 719

Free divst Kt 2 1 0208 0013 0000 0000 0000 01010334 0139 0043 0019 0000 02290004 0000 0000 0000 0000 00000740 0430 0078 0089 0000 0634

247 129 34 29 15 454

QUARTERLY JOURNAL OF ECONOMICS186

TABLE III(CONTINUED)

NFC LNFC PFC LFCNot Likely Possibly Likely FC Alln not n n n Fin rm-

constr constr constr constr constr years

CashtKt 2 1 0331 0150 0150 0077 0085 01680726 0253 0263 0156 0139 03640050 0034 0041 0029 0016 00331276 0596 0721 0389 0292 0784

393 221 52 34 19 719

Unused linet 0 0723 0652 0654 0529 0579 0683393 221 52 34 19 719

Unused linet Kt 2 1 0270 0178 0136 0043 0072 02030523 0313 0291 0151 0159 04150000 0000 0000 0000 0000 00001097 0733 0900 0449 0900 0979

393 221 52 34 19 719

Slackt Kt 2 1 0725 0420 0344 0211 0229 05571249 0566 0449 0374 0320 09190217 0118 0059 0044 0001 01262039 1129 0923 0721 1065 1679

393 221 52 34 19 719

Ch debtt Kt 2 1 0048 0048 0153 0272 0017 00620168 0157 0405 0473 0012 0191

2 0304 2 0354 2 0470 2 0414 2 0546 2 03540718 0760 0983 1581 0974 0797

393 221 52 34 19 719

Equity issuet Kt 2 1 0000 0000 0000 0000 0000 00000224 0149 0042 0020 0046 01770000 000 000 0000 0000 00000634 0419 0044 0000 0256 0455

373 193 38 31 16 651

the sum of the book value of short-term and long-term debt (items9 and 34) while total capital is the sum of debt the book value ofpreferred stock and the book value of common equity It is worthpointing out that NFC rm-years have a large median interestcoverage of almost eight times while the LNFC rm-years havea median coverage of almost six In contrast the median coveragein LFC rm-years is less than three times and in FC rm-yearsbarely exceeds one

INVESTMENT-CASH FLOW SENSITIVITIES 187

The notes to the nancial statements typically state whethera rmrsquos debt covenants if any restrict a rm from paying divi-dends We interpret a rm as being more nancially constrainedthe greater the restrictions placed on dividend payments by cove-nants Table III reports that the fraction of rm-years in whichdebt covenants forbid the payment of dividends increases mono-tonically across our classications NFC rm-years are restricted61 percent of the time while LFC and FC rms are restrictedmore than 68 percent of the time

In the majority of rm-years the notes to nancial state-ments also report exactly how much of retained earnings are freefor dividend payments under the strictest debt covenants12 PanelB of Table III indicates that this amount falls monotonicallyacross our ve groups In NFC rm-years the median amount ofearnings free for dividends equals 208 percent of beginning-of-year capital and almost 58 percent of the yearrsquos investment Inother words the median NFC rm could have paid out a dividendequal to 58 percent of its capital expenditures without the permis-sion of existing lenders

Finally cash (COMPUSTAT item 1) unused line of creditand slack (the sum of cash and unused line of credit) all declinemonotonically across our classications Slack provides a mea-sure of the amount of funds or liquidity immediately availableto a rm at year-end Slack may overstate true liquidity slightlybecause some rms were required to maintain compensating bal-ances That qualication notwithstanding the median slack inNFC rm-years is 725 percent of beginning-of-year capital and191 percent of the yearrsquos investment In LNFC rm-years theanalogous amounts are 42 percent and 119 percent

As an additional check we estimate ordered logit models ofthe probability that a rm falls in one of the ve categories withNFC being the lowest state and FC the highest The results arepresented in Table IV The likelihood of being classied as nan-cially constrained is signicantly greater in rms with higherdebt to total capital higher Q and for whom dividend paymentsare forbidden The likelihood is signicantly lower in rms withhigh cash ow high cash high dividends paid high retainedearnings free for dividends and with any unused line of credit at

12 This information is not reported in years that a rm has no debt as wellas some of the earlier rm-years

QUARTERLY JOURNAL OF ECONOMICS188

TABLE IVORDERED LOGITS FOR PREDICTABILITY OF FINANCING CONSTRAINT STATUS

Ordered logits for the determination of annual nancing constraint status for49 low-dividend rms are from FHP [1988] from 1970 to 1984 Financing con-straint for each year is ordered from not nancially constrained (NFC) likely notnancially constrained (LNFC) possibly nancially constrained (PFC) likely -nancially constrained (LFC) to nancially constrained (FC) Variable denitionsare in Tables I and III Standard errors are in brackets

Dependent variable is nancing constraint status

Cash owt Kt 2 1 2 0886 2 1164 2 0688 2 0839[0230] [0256] [0222] [0235]

Qt 0276 0370[0080] [0087]

Debttotal capitalt 2071 2251 1825 1938[0470] [0480] [0464] [0471]

DividendstKt 2 1 2 23039 2 21787 2 22551 2 20409[5949] [6134] [5905] [6043]

Dividends restricted (Y 5 1 N 5 0) 1496 1365 1472 1294[0213] [0224] [0213] [0222]

Unrestricted ret earningsKt 2 1 2 1897 2 1936 2 1896 2 1956[0497] [0513] [0499] [0513]

CashtKt 2 1 2 1704 2 1590 2 1675 2 1567[0311] [0323] [0311] [0320]

Unused line of credit 0 2 0711 2 0547 2 0758 2 0511[0176] [0207] [0175] [0206]

_cut1 2 0252 0608 2 0693 0119[0312] [0480] [0285] [0462]

_cut2 1973 2928 1510 2413[0328] [0499] [0298] [0478]

_cut3 2987 3988 2501 3433[0353] [0518] [0320] [0494]

_cut4 4307 5353 3790 4736[0413] [0562] [0378] [0532]

Year dummies No Yes No YesLog likelihood 2 6450 2 6270 2 6506 2 6357Pseudo-R2 0201 0223 0194 0213

all All the coefcients are statistically signicant at the 1 percentlevel and all the coefcients except perhaps the one on Q havethe expected sign Q has a positive impact on the probability ofbeing nancially constrained This is true despite the univariateresult in Table III that Q decreases with rm nancial healthThe likely explanation for this result is Qrsquos partial correlationwith cash ow In the absence of cash ow the coefcient onQ becomes negative One way to interpret this result is that con-

INVESTMENT-CASH FLOW SENSITIVITIES 189

ditional on having a low cash ow we classify a rm as morelikely to be constrained if it has more investment opportunities(high Q)

Overall we feel that the monotonic patterns of most of theoperating and nancial variables in Table III and the results inTable IV provide a strong quantitative validation of our classi-cation scheme

C Overall Financial Status

In order to analyze investment-cash ow sensitivities overfteen years we aggregate each rmrsquos annual nancial statusinto an overall measure of nancial status We refer to this assample nancial status (rather than rm-year nancial status)We distinguish rms that were never nancially constrainedfrom those that were We do this to account for the likelihood thatrms which become constrained will behave as if they areconstrained

Our classication provides a great deal of variation Nine-teen rms are never constrained they are classied as NFC orLNFC in all fteen sample years These rms never showed anysign of being nancially constrained over the entire period Eightrms are possibly constrained These rms were possibly con-strained in at least one year and not constrained (NFC andLNFC) in all the rest Finally 22 rms are likely constrainedThese rms were classied as LFC or FC in at least one sampleyear

We also aggregate annual nancial status into overall statusover two subperiods 1970 to 1977 and 1978 to 1984 We classifyrms according to whether they were likely constrained possiblyconstrained or not constrained within each subperiod We do thisfor four reasons First the classication over the entire sampleperiod will classify a rm as nancially constrained even if thatrm was constrained in only one of fteen years By measuringnancial status over subperiods we increase the precision of ourclassications Second the research design in FHP biases thesample toward companies that were small in 1970 but were es-tablished enough by 1984 to be included in the Value Line dataset Therefore a rm in the earlier part of the sample is conceiv-ably different from the same rm later on Third as noted earlierthe information contained in the management discussions andfootnotes of annual reports improves after 1977 Therefore webelieve that the precision of our classications increases in the

QUARTERLY JOURNAL OF ECONOMICS190

second subperiod Finally as noted earlier FHP [1988] reportthat the sensitivity of investment to cash ow is particularlystrong in the rst half of the sample when these rms weresmaller and more likely to have been nancially constrained

IV REGRESSION RESULTS

Armed with a direct measure of a rmrsquos nancially con-strained status we can now test whether the worsening of nan-cing constraints is associated with a monotonic increase ininvestment-cash ow sensitivity (as would occur if d 2IdW2 werenegative)

A Financing Constraints and Investment-Cash FlowSensitivities

We rst examine the relationship between nancing con-straints and investment-cash ow sensitivities by following theFHP methodology and estimating separate regressions by rmsample nancial status We use the regression specications pre-sented in Table I over the entire sample period Table V presentsour basic results Firms classied as never constrained (NFC orLNFC in every sample year) exhibit the highest investment-cashow sensitivity (0702) exceeding that for the entire sample(0395) for rms that were likely constrained (0340) and forrms that were possibly constrained (0180) The coefcient forthe never constrained rms is economically and statisticallygreater than the coefcients for the other rms

As we noted earlier we are not entirely comfortable with theclassication of possibly constrained rms If managers tend tounderreport negative information about their rmrsquos nancial con-dition then it would be appropriate to group the eight possiblyconstrained rms with the likely constrained rms Not surpris-ingly this grouping lowers the coefcient on constrained rms to0250 and widens the gap between constrained rms and uncon-strained rms It is worth pointing out that the 0250 sensitivityis insignicantly different from that of the high-dividend FHPClass 3 rms and actually smaller than that of the FHP Class2 rms

Alternatively it is possible that we have been excessivelyconservative and have classied rms as possibly constrainedwhen they were in fact unconstrained Accordingly we also esti-mate a regression in which we group the possibly constrained

INVESTMENT-CASH FLOW SENSITIVITIES 191

QUARTERLY JOURNAL OF ECONOMICS192

TA

BL

EV

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

RE

NT

IRE

SA

MP

LE

PE

RIO

D

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nst

rain

edst

atus

wh

ere

19

rms

are

neve

rn

anci

ally

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyn

ot

nan

cial

lyco

nstr

ain

ed(L

NF

C)

poss

ibly

na

nci

ally

cons

trai

ned

(PF

C)

like

ly

nanc

ially

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lylik

ely

neve

rpo

ssib

lypo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nst

rain

edco

nstr

ain

edco

nstr

aine

dco

nst

rain

edN

549

N5

19N

58

N5

22N

527

N5

30

CF

tKt2

10

395

070

20

180

034

00

439

025

0[0

026

][0

041

][0

060

][0

042

][0

035

][0

032

]Q

t21

003

90

009

001

60

070

003

30

059

[00

05]

[00

06]

[00

49]

[00

18]

[00

06]

[00

17]

Adj

R2

058

40

793

024

00

410

065

50

358

Nob

s71

927

911

332

739

244

0

rms with the never constrained rms While this lowers theinvestment-cash ow sensitivity substantially (to 0439) it doesnot alter the basic result that unconstrained rms exhibit agreater investment-cash ow sensitivity

In Tables VI and VII we repeat our basic analysis but breakthe sample into two subperiods 1970 to 1977 and 1978 to 1984Table VI treats a rm in the 1970ndash1977 subperiod as differentfrom the same rm in the 1978ndash1984 subperiod The regressionspresented in Table VI therefore include 98 rm-subperiods (withrm-subperiod xed effects) Again the coefcients sharply rejectthe hypothesis that nancially constrained rms have greaterinvestment-cash ow sensitivities In Table VI rms that are notconstrained in a subperiod have an investment-cash ow sensi-tivity of 0680 This is signicantly greater than the sensitivity of0436 for all rm subperiods and greater than the sensitivity ofrms that are possibly constrained (at 0259) or likely con-strained (at 0274)

Table VI also presents regression results for the fteen rm-subperiods for which we classify the rm as NFC in every year inthe subperiod13 Ten of the fteen subperiods fall in the 1978ndash1984 period during which even FHP argue the sample rms wereless likely to be constrained Strikingly the investment-cash owsensitivity for these fteen subperiods of 0779 exceeds any of thecoefcients for any group of rms we present in Tables VI and VII

Based on our classication scheme and the quantitative sup-port for that scheme in Tables III and IV we nd it impossible toargue that these rms were unable to invest more during any ofthese fteen subperiods We also nd it difcult to argue thatthese rms faced a particularly high cost of external nanceHewlett-Packard for example is included among these fteensubperiods in 1978ndash1984 (although not in 1970ndash1977) AndHewlett-Packard has an investment-cash ow sensitivity of 097over the 1978ndash1984 subperiod 091 over the 1970ndash1977 subpe-riod and 115 over the entire sample period It is worth stressingthat the fteen rms that are NFC in every subperiod year havenancial characteristics that are similar to those of FHPrsquos Class3 rms that pay high dividends and have a low investment-cashow sensitivity (023) For example the NFC rms and FHPrsquosClass 3 rms have interest coverage ratios that are economically

13 We do not create this classication over the entire sample period becausewe classify only two rms as NFC in all fteen years

INVESTMENT-CASH FLOW SENSITIVITIES 193

QUARTERLY JOURNAL OF ECONOMICS194

TA

BL

EV

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

IN

TW

OS

UB

PE

RIO

DS

TR

EA

TIN

GF

IRM

-SU

BP

ER

IOD

SA

SD

IFF

ER

EN

TF

IRM

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

mn

anci

alco

nstr

ain

tst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

enr

m-s

ubpe

riod

sar

ene

ver

nan

cial

lyco

nstr

ain

ed(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

aine

d(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

lyn

anci

ally

con

stra

ined

(LF

Cor

FC

inso

me

year

)an

d15

rm

-su

bper

iods

are

NF

Cev

ery

year

O

vera

llsu

bper

iod

stat

usis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

lyn

anci

ally

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ion

sin

clud

er

mx

edef

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

dx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyli

kely

NF

CA

llr

ms

con

stra

ined

con

stra

ined

con

stra

ined

cons

trai

ned

con

stra

ined

ever

yye

arN

598

N5

57N

514

N5

27N

571

N5

41N

515

CF

tKt2

10

436

068

00

259

027

40

523

026

20

779

[00

28]

[00

41]

[00

67]

[00

50]

[00

34]

[00

37]

[00

84]

Qt2

10

033

001

00

081

004

80

025

005

40

002

[00

05]

[00

06]

[00

59]

[00

23]

[00

06]

[00

21]

[00

09]

Adj

R2

060

40

721

040

20

391

066

10

402

083

7N

obs

719

416

9820

551

430

311

0

INVESTMENT-CASH FLOW SENSITIVITIES 195T

AB

LE

VII

RE

GR

ES

SIO

NO

FIN

VE

ST

ME

NT

ON

CA

SH

FL

OW

AN

DQ

BY

FIN

AN

CIA

LL

YC

ON

ST

RA

INE

DS

TA

TU

SO

VE

R19

70ndash1

977

AN

D19

78ndash1

984

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Sam

ple

isdi

vide

din

totw

osu

bper

iods

197

0ndash19

77an

d19

78ndash1

984

Fir

m

nan

cing

cons

trai

ntst

atus

isde

term

ined

wit

hin

each

subp

erio

dF

ifty

-sev

en

rm-s

ubp

erio

dsar

ene

ver

nan

cial

lyco

nstr

aine

d(N

FC

orL

NF

Cev

ery

year

)14

rm

-su

bper

iods

are

poss

ibly

nan

cial

lyco

nstr

ain

ed(P

FC

inso

me

year

)27

rm

-sub

peri

ods

are

like

ly

nan

cial

lyco

nst

rain

ed(L

FC

orF

Cin

som

eye

ar)

and

15r

m-

subp

erio

dsar

eN

FC

ever

yye

ar

Ove

rall

subp

erio

dst

atu

sis

base

don

rm

na

nci

ngco

nst

rain

tst

atu

sfo

rea

chye

arof

not

na

nci

ally

cons

trai

ned

(NF

C)

like

lyno

tn

anci

ally

con

stra

ined

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

lyn

anci

ally

cons

trai

ned

(LF

C)

and

nan

cial

lyco

nstr

aine

d(F

C)

Reg

ress

ions

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

Sta

ndar

der

rors

are

inbr

acke

ts

1970

ndash197

7

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

523

N5

7N

519

N5

80N

519

N5

5

CF

tKt2

10

505

074

60

247

036

40

553

030

60

783

[00

37]

[00

51]

[00

86]

[00

69]

[00

45]

[00

49]

[01

42]

Qt2

10

035

000

60

027

002

50

023

002

90

002

[00

17]

[00

07]

[00

82]

[00

23]

[00

07]

[00

22]

[00

15]

Adj

R2

069

60

827

038

10

454

075

50

446

083

2N

obs

378

179

5014

922

919

940

1978

ndash198

4

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

nev

erpo

ssib

lyli

kely

neve

rpo

ssib

lypo

ssib

lyl

ikel

yN

FC

rm

sco

nst

rain

edco

nst

rain

edco

nst

rain

edco

nstr

ain

edco

nstr

ain

edev

ery

year

N5

49N

534

N5

7N

58

N5

41N

515

N5

10

CF

tKt2

10

326

057

10

272

014

10

470

016

00

800

[00

44]

[00

69]

[01

52]

[00

61]

[00

58]

[00

53]

[01

26]

Qt

005

42

001

90

154

041

30

007

027

22

005

4[0

026

][0

028

][0

088

][0

084

][0

027

][0

073

][0

047

]R

20

392

046

70

422

045

90

439

040

20

703

Nob

s34

123

748

5628

510

470

and statistically indistinguishable It seems difcult to under-stand how one set of rms can be constrained while the otheris not

Table VII presents results for each of the two subperiods sepa-rately Again we nd no evidence that nancing constraints ex-plain the sensitivity of investment to cash ow In both subperi-ods the rms that we classify as NFC every year and as neverconstrained have a signicantly higher investment-cash ow sen-sitivity than the other groups of rms Furthermore in the 1978ndash1984 period where we are more certain of our classications thesensitivity declines monotonically with the extent to which weclassify rms as constrained

B Quantitative Denitions of Financially Constrained Status

Given the results in the previous section some readers maybe concerned that we have misclassied rms by using qualita-tive data (For example see FHP [1996]) To address this concernwe report the results of grouping rms based on quantitativeob-jective data

In regressions (1)ndash(3) of Table VIII we present estimates ofthe investment-cash ow sensitivities for (i) the 25 percent ofsample rms whose interest coverage never drops below 25 and(ii) rms whose dividends are never restricted over the entiresample period Given the two severe recessions over the sampleperiod these criteria should identify rms that were relativelynancially healthy The investment-cash ow sensitivity for thethirteen rms whose coverage never drops below 25 is signi-cantly greater at 0673 than the sensitivity of 0395 for the entiresample The investment-cash ow sensitivity for the seventeenrms whose dividends are never restricted at 0435 also exceedsthe sensitivity for the entire sample although not signicantly

In regressions (4)ndash(6) we split the sample into subperiods aswe did in Table VI We present estimates of the investment-cashow sensitivities for (i) the 25 percent of rm-subperiods whoseinterest coverage never drops below 45 in the subperiod and (ii)rms whose dividends are never restricted over the subperiodThe patterns are qualitatively similar and quantitativelystronger than those for the entire sample period The investment-cash ow sensitivity for the 21 rms whose coverage never dropsbelow 45 in a subperiod is a remarkably high 0801 We shouldpoint out that the median interest coverage for rms rated BBBby Standard amp Poorrsquos in 1979ndash1981 was 382 the median for

QUARTERLY JOURNAL OF ECONOMICS196

INVESTMENT-CASH FLOW SENSITIVITIES 197

TAB

LE

VII

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YO

TH

ER

ME

AS

UR

ES

OF

FIN

AN

CIA

LLY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

ES

AM

PL

EP

ER

IOD

AN

DE

NT

IRE

SU

BP

ER

IOD

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

esI

and

III

Reg

ress

ions

(1)ndash

(3)

are

esti

mat

edfo

rto

tal

sam

ple

and

by(i

)w

het

her

rm

sev

erha

din

tere

stco

vera

gebe

low

25

and

(ii)

rm

sar

eno

tex

plic

itly

rest

rict

edfr

ompa

ying

divi

den

dsov

erth

een

tire

sam

ple

peri

odI

nter

est

cove

rage

isth

era

tio

ofE

BIT

DA

toin

tere

stex

pen

seR

egre

ssio

ns

(4)ndash

(6)a

rees

tim

ated

usin

gr

mn

anci

alst

atu

sov

ersa

mpl

esu

bper

iods

1970

ndash197

7an

d19

78ndash1

984

Reg

res-

sion

s(1

)ndash(3

)in

clud

er

mx

edef

fect

san

dye

aref

fect

sR

egre

ssio

ns

(4)ndash

(6)

incl

ude

rm

xe

def

fect

sfo

rea

chsu

bper

iod

resu

ltin

gin

upto

98r

m-p

erio

d

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(4)

Fir

ms

that

(3)

All

rm

s(5

)(6

)ne

ver

have

Fir

ms

that

subp

erio

dsF

irm

sth

atF

irm

sth

atn

ever

(1)

cove

rage

neve

rha

ve19

70ndash1

984

neve

rh

ave

hav

ere

stri

cted

All

belo

w2

5fr

omre

stri

cted

and

cove

rage

belo

wdi

vide

nds

inr

ms

1970

ndash198

4di

vide

nds

1978

ndash198

44

5in

subp

erio

dsu

bper

iod

N5

49N

513

N5

17N

598

N5

25N

556

CF

tKt2

10

395

067

30

435

043

60

801

049

9[0

026

][0

056

][0

042

][0

028

][0

062

][0

038

]Q

t21

003

90

011

003

50

033

20

003

002

7[0

005

][0

008

][0

007

][0

005

][0

008

][0

006

]A

djR

20

584

075

60

674

060

40

772

071

5N

obs

719

191

247

719

189

402

rms rated A was 656 In other words these rms are not likelyto have faced particularly high costs of external nance in abso-lute terms in the subperiods More importantly in relative termsit is virtually certain that they faced lower costs of external -nance than the other rms in our sample and yet show a higherinvestment-cash ow sensitivity

C Predetermined Classication of Financially ConstrainedStatus

One important potential criticism of our results is that ouruse of nancial status over the entire period (or subperiod) mayldquohardwirerdquo our results Firms that only increase investmentwhen they have the cash ow to do so will exhibit a highinvestment-cash ow sensitivity and will be less likely to becomeconstrained subsequently In contrast rms that increase invest-ment when they do not have cash ow will exhibit a low sensitiv-ity and will be more likely to become constrained later (if theynance some of the investment with debt) It is possible there-fore that the investment-cash ow sensitivities we have esti-mated reect the way investment was nanced and that thisdrives our overall measure of nancial status rather than viceversa

Although it is fairly standard in this literature to sort rmsaccording to within-sample characteristics (for example FHP[1988] Hoshi Kashyap and Scharfstein [1991] and Lamont[1996]) this approach has received an increasing number of criti-cisms (see Schiantarelli [1995]) To address this concern we usea denition of nancial status that reects only past (not future)information Specically we measure rm nancial status basedon the previous yearrsquos nancial status This should isolate theeffect of nancial status rather than possibly reecting the wayin which investment was nanced In other words we believethat this specication is the most appropriate one to use to testfor the effect of nancing constraints on investment-cash owsensitivities

In regression (1) of Table IX we present differential estimatesof the investment-cash ow sensitivities by interacting cash owwith predetermined annual measures of nancing-constraint sta-tus14 We use four nancing constraint dummies LNFC equals

14 Our results are qualitatively identical when we also include dummy vari-ables for the intercept term

QUARTERLY JOURNAL OF ECONOMICS198

INVESTMENT-CASH FLOW SENSITIVITIES 199T

AB

LE

IXR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YA

NN

UA

LF

INA

NC

ING

CO

NS

TR

AIN

TS

TA

TU

SR

ES

TR

ICT

ED

DIV

IDE

ND

ST

AT

US

AN

DL

OW

SL

AC

KS

TA

TU

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

Q

an

dca

sh

owin

tera

cted

wit

hn

anci

ally

cons

trai

ned

stat

us

rest

rict

eddi

vide

ndst

atus

an

dlo

wca

shan

du

nuse

dli

neof

cred

itst

atus

for

49lo

w-d

ivid

end

rm

sar

efr

omF

HP

[198

8]fr

om19

70to

1984

Var

iabl

esar

ede

ne

din

Tabl

esI

and

III

Fir

mn

anci

ngco

nstr

aint

for

each

year

isno

tn

anci

ally

cons

trai

ned

(NF

C)

like

lyn

otn

anci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nanc

ially

cons

trai

ned

(PF

C)

like

ly

nan

cial

lyco

nstr

ain

ed(L

FC

)or

nan

cial

lyco

nstr

aine

d(F

C)

The

noni

nter

acte

dca

sh

owva

riab

lere

pres

ents

year

sin

wh

ich

rm

sar

eN

FC

R

egre

ssio

ns

(1)

and

(2)

use

na

ncia

lco

nstr

aint

stat

usat

the

begi

nni

ng

ofth

es

cal

year

(bas

edon

stat

usat

the

end

ofth

epr

evio

uss

caly

ear)

Reg

ress

ion

(3)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquosco

vena

nts

rest

rict

itfr

ompa

ying

divi

dend

sin

the

prev

iou

ss

cal

year

Reg

ress

ion

(4)

inte

ract

sca

sho

ww

ith

adu

mm

yva

riab

leth

ateq

uals

one

ifa

rm

rsquossl

ack

inth

epr

evio

uss

cal

year

isin

the

low

est

quar

tile

ofr

m-y

ears

(les

sth

an0

28of

net

prop

erty

pl

ant

and

equi

pmen

t)

Slac

kis

the

sum

ofca

shan

dun

used

line

ofcr

edit

R

egre

ssio

nsin

clu

der

m

xed

effe

cts

and

year

effe

cts

Sta

ndar

der

rors

are

inbr

acke

ts

(2)

(1)

Inve

stm

ent

by(3

)(4

)In

tera

ctan

nual

annu

aln

anci

alIn

tera

ctan

nua

lIn

tera

ctan

nual

low

slac

kn

anci

alco

nstr

aint

stat

usco

nst

rain

tst

atus

rest

rict

eddi

vide

ndst

atus

stat

us

CF

tKt2

10

407

Con

stan

t0

202

CF

tKt2

10

358

CF

tKt2

10

359

[00

43]

[00

27]

[00

29]

[00

27]

CF

tKt2

10

013

LN

FC

20

060

CF

tKt2

12

010

6C

FtK

t21

20

061

3L

NF

C[0

035

][0

026

]3

rest

rict

ed[0

052

]3

low

slac

k[0

040

]di

vide

nds

CF

tKt2

12

023

5P

FC

20

112

3P

FC

[00

55]

[00

45]

CF

tKt2

12

038

2L

FC

20

167

3L

FC

[00

86]

[00

54]

CF

tKt2

12

039

4F

C2

025

13

FC

[01

62]

[00

69]

Qt2

10

041

Qt2

10

101

Qt2

10

048

Qt2

10

051

[00

11]

[00

11]

[00

11]

[00

11]

Adj

R2

050

40

342

047

60

475

Nob

s67

467

467

467

4

one if the rm is likely not nancially constrained in the previousscal year PFC if the rm is possibly nancially constrainedthat year LFC if the rm is likely nancially constrained thatyear and FC if the rm is denitely nancially constrained thatyear The base or constant term measures investment in NFCrm-years We stress that this classication scheme uses only in-formation available at the beginning of the scal year The re-sults are qualitatively identical to those in the previous sectionsThe investment-cash ow sensitivities are signicantly lower forFC LFC and PFC rm-years than for LNFC and NFC rm-years The results are qualitatively identical if FC and LFC rm-years and LNFC and NFC rm-years are classied together

Although we prefer the above method because it fully usesthe annual information on each rmrsquos nancial status we alsotried an alternative approach that is somewhat more consistentwith the previous literature (These results are not reported in atable) For each year from 1970 to 1977 we divided the sampleinto two groups depending on whether the rm was classied asunconstrained (NFC and LFNC) or constrained (LFC and FC) inthat year We then used the following seven years to estimateseparate sensitivity coefcients for the two groups In all eightpaired regressions the estimated investment-cash ow sensitiv-ity of the unconstrained rms is higher than that of the con-strained rms In ve of the eight regressions this difference isstatistically signicant

These ndings conrm our previous empirical results andsupport our theoretical claim that investment-cash ow sensitiv-ity is not necessarily increasing in the degree of nancing con-straints The one remaining question perhaps is our measure ofnancing constraints Fortunately the simple model we presentedin Section I provides a way to test the reliability of our indicatorsEquation (3) makes the unequivocal theoretical prediction thatceteris paribus investment should decrease in the degree of -nancing constraints By looking at the relation of our annual -nancing constraint measures to investment we can assess thevalidity of those measures We do this by introducing our annualnancing constraint indicators in a standard Q model of invest-ment (and controlling for xed rm and year effects) This test isnot possible in the earlier regressions because overall nancingconstraint status is collinear with rm xed effects

The results are reported in regression (2) of Table IX Con-trolling for Q investment levels decline monotonically in the de-

QUARTERLY JOURNAL OF ECONOMICS200

gree of nancing constraints For example investment after LFCrm-years is 2 017 lower than after NFC rm-years The resultsare strongly consistent with the predictions derived in equation(3) They also suggest that our lagged measure of nancing con-straints successfully captures the degree of nancing constraint15

In regressions (3) and (4) we repeat the analysis in regression(1) but instead use quantitative measures of nancial statusagain based on the previous yearrsquos results In regression (3) weinteract cash ow with a dummy variable that equals one whendebt covenants restrict the rm from paying dividends Again wend that the investment-cash ow sensitivity is signicantlylower not higher for rms restricted from paying dividends Inregression (4) we interact cash ow with a dummy variable thatequals one if in the previous rm-year our slack variablemdashthesum of cash and unused lines of credit as a fraction of capitalmdashis in the lowest quartile of rm-years The low slack cutoff is 28percent of beginning-of-year capital (net property plant andequipment) Our results are not sensitive to this cutoff It seemsreasonable to assume that rms with less slack are more nan-cially constrained than rms with more slack Again we nd thatthe investment-cash ow sensitivity is lower not higher for rmswith low slack

Overall then we obtain qualitatively identical results usingboth qualitative and quantitative measures of nancing con-straints that are predetermined

D Sensitivity to Cash Stock

Although most of the literature focuses on the sensitivity ofinvestment to cash ow some authors (eg Kashyap Lamontand Stein [1994]) focus on the sensitivity of investment to thecash stock (cash and marketable securities) that a rm has avail-able at the beginning of the year For completeness in Table Xwe reestimate the annual nancing constraint regressions inTable IX using this alternative measure of liquidity (The resultswe report are qualitatively identical when we instrument cashholdings with its lagged value)

In the regression in column (1) we measure liquidity as cash

15 These results are also interesting for the debate on the relationship be-tween investment and Q in nancially constrained rms Chirinko [1995] arguesthat the effects of nancing constraints will be fully reected in a rmrsquos marketvalue and thus on its Q To the contrary our results suggest that Q is not suf-cient to explain the investment of nancially constrained rms

INVESTMENT-CASH FLOW SENSITIVITIES 201

TABLE XREGRESSION OF INVESTMENT ON CASH FLOW CASH STOCK AND Q BY ANNUAL

FINANCING CONSTRAINT STATUS

Regression of investment on cash ow cash stock Q and cash ow and cashstock interacted with nancially constrained status for 49 low-dividend rms fromFHP [1988] from 1970 to 1984 Variables are dened in Tables I and III Firmnancing constraint status for each year is not nancially constrained (NFC)likely not nancially constrained (LNFC) possibly nancially constrained (PFC)likely nancially constrained (LFC) or nancially constrained (FC) The nonin-teracted cash ow variable represents years in which rms are NFC Regressionsinclude rm xed effects and year effects Standard errors are in brackets

(2) (3)(1) Cash stock Sum of cash stock

Cash stock only and cash ow and cash ow

Casht 2 1 Kt 2 1 0164 Casht 2 1 Kt 2 1 0101 [Casht 2 1 1 CFt]Kt 2 1 0163[0015] [0015] [0011]

Casht 2 1 Kt 2 1 0056 Casht 2 1 Kt 2 1 0014 [Casht 2 1 1 CFt]Kt 2 00793 LNFC [0057] 3 LNFC [0060] 3 LNFC [0024]

Casht 2 1 Kt 2 1 2 0154 Casht 2 1 Kt 2 1 0269 [Casht 2 1 1 CFt]Kt 2 2 00373 PFC [0125] 3 PFC [0129] 3 PFC [0041]

Casht 2 1 Kt 2 1 2 0463 Casht 2 1 Kt 2 1 0249 [Casht 2 1 1 CFt]Kt 2 2 01743 LFC [0219] 3 LFC [0257] 3 LFC [0064]

Casht 2 1 Kt 2 1 2 0523 Casht 2 1 Kt 2 1 0321 [Casht 2 1 1 CFt]Kt 2 2 01963 FC [0340] 3 FC [0355] 3 FC [0121]

CFtKt 2 1 0342[0033]

CFtKt 2 1 00763 LNFC [0041]

CFtKt 2 1 2 02223 PFC [0062]

CFtKt 2 1 2 03843 LFC [0108]

CFtKt 2 1 2 04053 FC [0179]

Qt 2 1 0085 Qt 2 1 0040 Qt 2 1 0040[0011] [0010] [0010]

Adj R2 0306 0441 0393N obs 674 674 674

stock deated by net property plant and equipment both at thebeginning of the year The regression estimates the sensitivity ofinvestment to cash stock as a function of a rmrsquos nancial statusOur ndings are qualitatively identical to those in Table IX theleast constrained rms show the highest sensitivity

In column (2) we include both measures of liquidity cash

QUARTERLY JOURNAL OF ECONOMICS202

ow and cash stock The sensitivity of investment to cash owdecreases with the degree of nancing constraintsmdashthe samepattern encountered throughout the paper In contrast the sensi-tivity of investment to cash stock now increases with the degreeof nancing constraints These latter results however are notstatistically signicant none of the coefcients are statisticallydifferent from each other

These results may raise the question of which sensitivity isthe relevant one The theory however does not distinguish be-tween cash ow and cash stock the effect of an extra dollar offunds should be the same independent of whether it enters therm this period (as cash ow) or whether it was present in therm at the beginning of the period (as cash stock) For this rea-son we estimate a regression in column (3) of Table X that mea-sures liquidity as the sum of cash ow and cash stock Our mainnding is conrmed the least constrained rms show a signi-cantly higher sensitivity of investment to internal funds

We also estimated (but do not report in a table) the regres-sions in Tables V and VII with cash stock and cash ow In allregressions our basic nding holds investment-cash ow sensi-tivities decrease signicantly with the degree of nancing con-straints The results for investment-cash stock sensitivities aremixed Over the entire sample period investment-cash stock sen-sitivities increase signicantly with the degree of nancing con-straints However this pattern does not hold for either the1970ndash1977 or the 1978ndash1984 subperiod

E Alternative Specications

We considered but do not report a number of alternativespecications of our basic regressions (1) We removed Q as anindependent variable leaving cash ow as the only independentvariable (2) We added the ratio of sales to capital as an indepen-dent variable with Q and cash ow to capital (3) We included twolags of cash ow and Q as independent variables (4) To reducethe inuence of outliers we (i) winsorized investment cash owand Q (ii) deated investment and cash ow by total assetsrather than by capital (iii) eliminated observations with negativecash ow and (iv) measured cash ow using EBITDA (5) We ranregressions for each rm individually (6) We checked whetherthe results hold if we exclude any particular rm from the sam-ple Our results are qualitatively and statistically identical under

INVESTMENT-CASH FLOW SENSITIVITIES 203

all of these alternatives These specications and results there-fore address the concerns raised by FHP [1996] that our empiri-cal results could be the artifact of a censored regression16

We also tested the robustness of our results with respectto different denitions of investment Besides the standarddenition (COMPUSTAT item 128) we used the following (1)COMPUSTAT item 30 which includes increases in propertyplant and equipment from acquisitions that use purchase ac-counting (2) change in net property plant and equipment (3)change in net property plant and equipment adding back depre-ciation and (4) the sum of capital expenditures and research anddevelopment All four adjustments yield results that are qualita-tively and statistically identical to our basic results Finally weestimated inventory regressions similar to those estimated byCarpenter Fazzari and Petersen [1995] Again we nd no evi-dence that the sensitivity of inventory investment to cash owincreases with nancing constraints

V DISCUSSION OF THE RESULTS

The results indicate that a high sensitivity of investment tocash ow is not associated with nancially constrained rms inour sample This contrasts with the results in FHP [1988] andmany subsequent papers This section argues in greater detailthat our ndings are not specic to our sample but insteadlikely capture general features of the relationship between corpo-rate investment and cash ow Section VI discusses the implica-tions of these ndings for the previous literature

A Cash Flow as a Proxy for Investment Opportunities

One possible criticism is that our sorting criteria are corre-lated with the mismeasurement of Q and that this effect over-comes the effect of nancing constraints (which go in the oppositedirection) This criticism was rst made in Poterbarsquos [1988] dis-cussion of FHP [1988] Poterba points out that if cash ow pro-vides more information about future investment opportunities forcertain groups of rms (like nondividend paying rms) suchrms on average would have a greater investment-cash ow sen-

16 In fact we believe it is telling that FHP [1996] criticize our results hypo-thetically rather than by showing that the criticisms hold in the data

QUARTERLY JOURNAL OF ECONOMICS204

sitivity independent of their nancial status FHP [1996] presenta similar criticism of our results

In the literature following FHP [1988] this measurementproblem has been addressed by using the so-called Euler equa-tion approach (see Whited [1992] Bond and Meghir [1994] andHubbard Kashyap and Whited [1995]) This approach directlytests the rst-order conditions of an intertemporal maximizationproblem that does not require a measurement of Q and thereforeis (supposedly) unaffected by Qrsquos mismeasurement

To test the robustness of our ndings we followed the Eulerequation approach developed in Bond and Meghir [1994] who ex-plicitly model the wedge between internal and external nanceTheir empirical implementation involves regressing investmenton lagged investment and its square sales cash ow and debtsquared and testing whether the coefcient on cash ow is differ-ent across rms with different dividend policies When we imple-ment this approach we obtain results qualitatively identical tothose from our basic specication Our least constrained rms ex-hibit the highest coefcients

In sum the Euler equation approach provides no evidencethat our ndings are driven by mismeasurement of Q (The alter-native interpretationmdashthat the Euler equation approach fails tocontrol for differences in investment opportunitiesmdashwould callinto question all the results in the literature derived using thatmethodology)

B The Impact of Outliers

The papers in this literature typically deate all the vari-ables by the value of capital (net property plant and equipment)at the beginning of the scal year This method provides consis-tent estimates if all variables are recorded at short intervals or ifthere is no growth In practice however neither of the two as-sumptions is satised Variables are recorded at annual inter-vals and companies grow substantially over the sample perioda median of 18 percent per year for our sample If both invest-ment and cash ow grow at a rate similar to the growth rate ofsales then part of the comovement of investment and cash owmay be due to a scale factor This effect would bias the estimatesof the investment-cash ow sensitivity toward one particularlyin rms with higher annual growth rates

To account for this possibility we estimate regressions that

INVESTMENT-CASH FLOW SENSITIVITIES 205

eliminate or downweight observations with high growth ratesThe rst four columns of Table XI report the results of regres-sions that exclude rm-years with more than 30 percent salesgrowth (the upper quartile) When we eliminate these observa-tions the median rate of sales growth for the constrained pos-sibly constrained and not constrained rms is essentially equal(between 11 percent and 12 percent) The investment-cash owsensitivities decline substantially Nevertheless the patternacross the three groups of rms remains qualitatively the sameand the difference in sensitivities is still statistically signicantThe second four columns of Table XI report qualitatively similarresults when we eliminate rm-years in which net propertyplant and equipment more than doubled17 Finally we obtainqualitatively and statistically similar results (in unreported re-gressions) when we apply a robust estimation technique thatdownweights outliers18

In sum our cross-sectional results are not driven by outliersThe same cannot be said for the overall results in FHP [1988]Eliminating or downweighting high growth rm-years reducesthe estimated investment-cash ow sensitivity of the entire lowdividend payout sample to between 020 and 025 This is effec-tively identical to the estimate of 023 obtained by FHP for theirunconstrained high payout rms Given that these rms are lesslikely to experience such extreme growth rates these results in-dicate that FHPrsquos overall ndings (across payout classes) are atleast partially driven by extreme observations

Unfortunately this problem is not likely to be restricted toFHP [1988] Any splitting criterion that sorts rms into subsam-ples with differential outliers in growth ratesmdashfor examplesplits on size and dividend payout ratiosmdashmay be biased towardnding a difference in coefcients on cash ow This bias maypartially account for the large body of evidence nding a higherinvestment-cash ow sensitivity in fast growing companies thattend to be classied as nancially constrained

17 Following a suggestion of David Scharfstein we investigated all the ob-servations where property plant and equipment more than doubled in a singleyear In most of these cases the increase in investment appears to have beendriven by a sudden surge in both the demand for the rmrsquos product and rmprots

18 This method implemented by STATA performs an initial screening toeliminate gross outliers prior to calculating starting values and then performs assuggested by Li [1985] Huber iterations followed by biweight iterations The re-sults are available upon request

QUARTERLY JOURNAL OF ECONOMICS206

INVESTMENT-CASH FLOW SENSITIVITIES 207

TA

BL

EX

IR

EG

RE

SS

ION

OF

INV

ES

TM

EN

TO

NC

AS

HF

LO

WA

ND

QB

YF

INA

NC

IAL

LY

CO

NS

TR

AIN

ED

ST

AT

US

OV

ER

EN

TIR

EP

ER

IOD

WIT

HO

UT

HIG

HS

AL

ES

GR

OW

TH

OR

HIG

HIN

VE

ST

ME

NT

GR

OW

TH

OB

SE

RV

AT

ION

S

Reg

ress

ion

ofin

vest

men

ton

cash

ow

and

Qfo

r49

low

-div

iden

dr

ms

from

FH

P[1

988]

from

1970

to19

84V

aria

bles

are

den

edin

Tabl

eI

Reg

ress

ions

are

esti

mat

edfo

rto

tals

ampl

ean

dby

nan

cial

lyco

nstr

aine

dst

atus

wh

ere

19r

ms

are

nev

er

nanc

ially

cons

trai

ned

over

the

enti

repe

riod

(NF

Cor

LN

FC

inev

ery

year

)8

rm

sar

epo

ssib

lyn

anci

ally

cons

trai

ned

atso

me

tim

e(P

FC

inso

me

year

)an

d22

rm

sar

eli

kely

nan

cial

lyco

nstr

ain

edat

som

eti

me

inth

epe

riod

(LF

Cor

FC

)O

vera

llst

atus

isba

sed

onr

mn

anci

ng

cons

trai

ntst

atu

sfo

rea

chye

arof

not

nan

cial

lyco

nstr

aine

d(N

FC

)li

kely

not

na

nci

ally

cons

trai

ned

(LN

FC

)po

ssib

ly

nan

cial

lyco

nstr

aine

d(P

FC

)li

kely

na

nci

ally

cons

trai

ned

(LF

C)

and

na

nci

ally

cons

trai

ned

(FC

)A

llre

gres

sion

sin

clud

e

rmx

edef

fect

san

dye

aref

fect

sS

tand

ard

erro

rsar

ein

brac

kets

No

rm

-yea

rsw

ith

inve

stm

ent

exce

edin

gN

or

m-y

ears

wit

hm

ore

than

30

sale

sgr

owth

init

ial

capi

tal(

Kt2

1)

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

Fir

ms

All

neve

rpo

ssib

lyli

kely

All

neve

rpo

ssib

lyli

kely

rm

sco

nstr

ain

edco

nstr

ain

edco

nstr

aine

d

rms

cons

trai

ned

cons

trai

ned

cons

trai

ned

N5

49N

519

N5

8N

522

N5

49N

519

N5

8N

522

CF

tKt2

10

246

053

10

104

023

30

203

036

60

149

021

1[0

050

][0

124

][0

045

][0

058

][0

031

][0

042

][0

046

][0

032

]Q

t21

005

10

033

004

80

049

004

60

023

20

001

006

7[0

012

][0

014

][0

053

][0

024

][0

009

][0

010

][0

027

][0

013

]A

djR

20

328

050

20

155

027

00

449

059

70

252

042

7N

obs

535

201

7925

567

926

310

930

7

C Financially Constrained Equals Financially Distressed

It is plausible that nancially distressed rms will exhibitlow investment-cash ow sensitivities For example an insolventrm might be forced by its creditors to use additional cash owto repay debt rather than for capital expenditures This necessar-ily will reduce the sensitivity of investment to cash ow If therms we classify as constrained and possibly constrained are infact nancially distressed this would reduce the generality andimpact of our results

Tables III and XII however refute this argument Table IIIpresents rm characteristics by rm-year nancial status TableXII presents rm characteristics by overall sample nancial sta-tus First the bottom of Table III indicates that rms increasetheir debt rather than repay it in the years we classify them aspossibly likely or denitely constrained Second although onemight argue that the denitely constrained rm-years are dis-tressed (median interest coverage of 109) Table III shows thatthis is not likely to be the case for the likely constrained rm-years (median interest coverage of 284) and denitely not thecase for the possibly constrained rm-years (median interest cov-erage of 420) Third Table XII shows that over the entire sampleperiod rms we classify as possibly constrained are approxi-mately as healthy as rms we classify as never constrained Fi-nally it is unreasonable to describe the likely constrained rmsas distressed over the entire sample period (median interest cov-erage of 484) despite the fact that they are less healthy overallthan the other two groups In fact FHP [1988] intended to elimi-nate distressed rms because they explicitly excluded rms withoverall negative real sales growth from their sample

VI IMPLICATIONS FOR PREVIOUS WORK

The discussion above suggests that our ndings are notcaused by econometric problems or an inappropriate classica-tion scheme In our sample there is a negative rather than posi-tive correlation between investment-cash ow sensitivities andthe degree of nancing constraints This shows that a nonmono-tonic relationship (or even an inverse relationship) is not onlytheoretically possible but is also empirically relevant Only fu-ture work will be able to ascertain how pervasive this nonmono-tonicity is However our paper shows that monotonicity cannotbe taken for granted

QUARTERLY JOURNAL OF ECONOMICS208

TABLE XIIMEDIAN FIRM CHARACTERISTICS BY FINANCIALLY CONSTRAINED STATUS

IN ENTIRE SAMPLE PERIOD

Median rm characteristics by overall nancial status for 49 low-dividendrms from FHP [1988] from 1970 to 1984 Overall status is based on rm nanc-ing constraint status for each year of not nancially constrained (NFC) likelynot nancially constrained (LNFC) possibly nancially constrained (PFC) likelynancially constrained (LFC) and nancially constrained (FC) For the entireperiod 19 rms are never nancially constrained over the entire period (NFC orLNFC in every year) 8 rms are possibly nancially constrained at some time(PFC in some year) and 22 rms are likely nancially constrained at some timein the period (LFC or FC) Each entry reports the median and number of observa-tions Investment (It) cash ow Q and capital (Kt 2 1) are dened in Table I Inter-est coverage is the ratio of earnings before interest taxes and depreciation(EBITDA) to interest expense Debt is the sum of the book value of short-termand long-term debt Total capital is the sum of debt the book value of preferredstock and the book value of common equity Free divs is the amount of retainedearnings that are not restricted from being paid out as dividends Cash is cashand marketable securities Unused linet is the amount of unused line of credit atthe end of year t Slack is the sum of cash and unused line

Never Possibly Likely Allconstrained constrained constrained rm-years

N 5 279 N 5 113 N 5 327 N 5 719

A Investment cash ow growth

It Kt 2 1 0348 0403 0337 0348Cash FlowtKt 2 1 0451 0517 0364 0421(Cash Flowt 2 It)Kt 2 1 0081 0142 0001 0051Qt 1262 1438 1200 1231Sales growtht 0194 0176 0172 0180

B Financial policy

Interest coveraget 8070 9928 4842 6406Debtt to total capitalt 0289 0249 0415 0349Fraction of yearsdividends restricted 0115 0070 0327 0206Free divstKt 2 1 0186 0315 0023 0101CashtKt 2 1 0215 0239 0109 0168Unused linet 0 0631 0649 0730 0683Unused linet Kt 2 1 0153 0208 0256 0203Slackt Kt 2 1 0626 0630 0481 0557Ch debtt Kt 2 1 0048 0000 0094 0062Years with equity issue 0234 0167 0189 0203

One might argue that we have only raised a possibility andthat our ndings do not generalize beyond the specic FHP[1988] sample In fact FHP [1996]mdashciting the large body of evi-dence which nds that an increased sensitivity is associated witha priori measures of nancing constraintsmdashdismiss our resultsas little more than an empirical counterexample

INVESTMENT-CASH FLOW SENSITIVITIES 209

The existing literature however cannot be brought in as evi-dence against our results for two reasons First it is likely that apublication selection bias exists in this literature Because thenull hypothesis before FHP [1988] was that nancing constraintsdid not matter only papers showing otherwise were likely to bewritten and published (See De Long and Lang [1992])

More importantly (and ignoring the possible publicationbias) the existing evidence can be used to support the monoton-icity assumption only if the theoretical priors used in those stud-ies unequivocally identify those rms as more likely to beconstrained If this were the case then the fact that most studiesnd a higher sensitivity for rms more likely to be constrainedmay be interpreted in favor of a monotonic relation between sen-sitivities and nancing constraints However if the priors are am-biguous (and monotonicity cannot be taken for granted) then theargument is invalid and the interpretation of many of the resultsin this literature becomes questionable high sensitivities per secannot be taken as evidence of nancing constraints

In our view most of the sorting criteria used in this literatureare indeed theoretically ambiguous Consider for example oneof the better known papers in this literature Hoshi Kashyapand Scharfstein [1991] which divides Japanese rms on the basisof whether they belong to a keiretsu and therefore to a largeextent of whether they have a main-bank relationship Althoughit is easy to argue that such a relationship will have an effect ona rmrsquos nancing and investment policy it is much less clear ona priori grounds what the sign of this effect should be Some the-ories (eg Myers and Majluf [1984]) imply a positive role for amain-bank relationship in reducing informational asymmetriesand thus in alleviating nancing constraints Hoshi Kashyapand Scharfstein nd that Japanese rms with an exclusive bankrelationship have a lower investment-cash ow sensitivity Bystressing these theories Hoshi Kashyap and Scharfstein inter-pret their ndings as evidence that a main-bank relationshipmakes rms less constrained

In contrast other theories (eg Sharpe [1990] and Rajan[1991]) imply that banks can exploit an exclusive main-bank rela-tionship and charge client rms a higher cost of capital (ie makethem more nancially constrained) The nding in HoshiKashyap and Scharfstein [1993] that the healthiest Japaneserms (from their original sample) subsequently broke their exclu-sive bank relationships is consistent with this interpretation

QUARTERLY JOURNAL OF ECONOMICS210

Houston and James [1995] nd that U S rms with an exclusivebank relationship have a higher investment-cash ow sensitivityBy stressing these other theories Houston and James interpretthis as evidence that a main-bank relationship makes these rmsmore not less constrained

The theoretical ambiguity is not unique to Hoshi Kashyapand Scharfstein [1991] and Houston and James [1995] but isshared by most of the splitting criteria used in this literatureFirms with a lower-than-average leverage are sometimes inter-preted a priori as relatively unconstrained rms (eg Whited[1992]) because they retain a large debt capacity and can obtainexternal funds very easily In other papers rms with lower-than-average leverage are considered to be relatively constrained(eg Calomiris and Himmelberg [1995]) because they are as-sumed to maintain low leverage because the costs of being nan-cially constrained or distressed would be extremely high

Similarly rms with unusually high cash holdings are some-times classied a priori as relatively unconstrained [KashyapLamont and Stein 1994] because they can invest that cash Andsometimes rms with unusually high cash holdings are classiedas relatively constrained [Calomiris Himmelberg and Wachtel1995] because they are assumed to need to accumulate that cashas precautionary savings to avoid the high costs of being nan-cially constrained or distressed in the future

In sum the theoretical priors are ambiguous As a result ourndings are not a minor counterexample in a large literature thatnds otherwise Rather this is the rst paper to test the very as-sumption upon which the literature is based19

VII CONCLUSION

Our analysis indicates that the investment-cash ow sensi-tivity criterion as a measure of nancing constraints is not well-grounded in theory and is not supported by empirical evidence inthe case we investigate While we believe that the nonmonoton-icity problem we have documented is pervasive and affects manyof the results in this literature future research will be needed toconrm this hypothesis

19 Our methodology is not subject to the same criticisms for two reasonsFirst we classify rm nancing constraint status using direct observation ratherthan theoretical priors Second we conrm the quality of our nancing constraintindicators using a test for which the theory is unequivocal

INVESTMENT-CASH FLOW SENSITIVITIES 211

QUARTERLY JOURNAL OF ECONOMICS212

If the nonmonotonicity result is general then it will be im-portant to understand its source One explanation (implicitly as-sumed in our theoretical model) involves understanding theshape of the cost function of raising external nance where exter-nal nance is costly because of information or agency problemsAlternatively it is possible that the nonmonotonic behavior of theinvestment-cash ow sensitivity is driven by a mischaracteriza-tion of the reasons why rms are reluctant to raise external -nance The most nancially successful and least constrainedrms in our sample appear to rely primarily on internal cash owto invest despite the availability of additional low cost funds andtherefore exhibit a high investment-cash ow sensitivity Thekey questionmdashthat we do not answermdashis why we observe thisbehavior It seems important that future work attempt to distin-guish among these explanations because of their disparate policyimplications for institutional and incentive design

The nal implication of our paper is a methodological oneOur research design and results point out what we think is aweakness in existing research as well as an opportunity for fu-ture research A great deal can be learned through more directobservation Appendix follows references

GRADUATE SCHOOL OF BUSINESS UNIVERSITY OF CHICAGO AND NATIONAL BUREAU

OF ECONOMIC RESEARCH

REFERENCES

Blanchard Olivier Florencio Lopez-de-Silanes and Andrei Shleifer ldquoWhat DoFirms Do with Cash Windfallsrdquo Journal of Financial Economics XXXVI(1994) 337ndash60

Bond Stephen and Costas Meghir ldquoDynamic Investment Models and the FirmrsquosFinancial Policyrdquo Review of Economic Studies LXI (1994) 197ndash222

Calomiris Charles and Charles Himmelberg ldquoInvestment Banking Costs as aMeasure of Access to External Financerdquo manuscript University of Illinois1995

Calomiris Charles Charles Himmelberg and Paul Wachtel ldquoCommercial PaperCorporate Finance and the Business Cyclerdquo Carnegie Rochester Series onPublic Policy (1996) forthcoming

Carpenter Robert Steven Fazzari and Bruce Petersen ldquoThree Financing Con-straint Hypotheses and Inventory Investment New Tests with Time and Sec-toral Heterogeneityrdquo manuscript Washington University 1995

Chirinko Robert ldquoFinance Constraints Liquidity and Investment SpendingCross-Country Evidencerdquo manuscript Emory University 1995

De Long Bradford and Kevin Lang ldquoAre All Economic Hypotheses Falserdquo Jour-nal of Political Economy C (1992) 1257ndash72

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investmentrdquo Brookings Papers on Economic Activity (1988)141ndash95

Fazzari Steven R Glenn Hubbard and Bruce Petersen ldquoFinancing Constraintsand Corporate Investment Response to Kaplan and Zingalesrdquo NBER Work-ing Paper No 5462 1996

INVESTMENT-CASH FLOW SENSITIVITIES 213

Greenwald Bruce Joseph Stiglitz and Andrew Weiss ldquoInformation Imperfec-tions and Macroeconomic Fluctuationsrdquo American Economic Review LXXIV(1984) 194ndash99

Gross David ldquoThe Investment and Financing Decisions of Liquidity-ConstrainedFirmsrdquo manuscript University of Chicago 1995

Grossman Sanford and Oliver Hart ldquoCorporate Financial Structure and Mana-gerial Incentivesrdquo in J J McCall ed The Economics of Information and Un-certainty (Chicago IL University of Chicago Press 1982)

Hart Oliver ldquoTheories of Optimal Capital Structure A Managerial DiscretionPerspectiverdquo in M Blair Ed The Deal Decade (Washington DC The Brook-ings Institution 1993)

Hart Oliver and John Moore ldquoDebt and Seniority An Analysis of the Role ofHard Claims in Constraining Managementrdquo American Economic ReviewLXXXV (1995) 567ndash85

Hayashi Fumio ldquoTobinrsquos Marginal q and Average q A Neoclassical Interpreta-tionrdquo Econometrica L (1982) 213ndash24

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoCorporate Structure Liquid-ity and Investment Evidence from Japanese Panel Datardquo Quarterly Journalof Economics CVI (1991) 33ndash60

Hoshi Takeo Anil Kashyap and David Scharfstein ldquoThe Choice between Publicand Private Debt An Analysis of Post-Deregulation Corporate Financing inJapanrdquo NBER Working Paper No 4421 1993

Houston Joel and Christopher James ldquoBanking Relationships Financial Con-straints and Investments Are Bank Dependent Borrowers More FinanciallyConstrainedrdquo Working Paper University of Florida 1995

Hubbard R Glenn Anil Kashyap and Toni Whited ldquoInternal Finance and FirmInvestmentrdquo Journal of Money Credit and Banking XXVII (1995) 683ndash701

Jensen Michael ldquoAgency Costs of Free Cash Flow Corporate Finance and Take-oversrdquo American Economic Review LXXVI (1986) 323ndash29

Jensen Michael and William Meckling ldquoThe Theory of the Firm ManagerialBehavior Agency Costs and Ownership Structurerdquo Journal of Financial Eco-nomics III (1976) 305ndash60

Kaplan Steven and Luigi Zingales ldquoDo Financing Constraints Explain Why In-vestment Is Correlated with Cash Flowrdquo NBER Working Paper No 52671995

Kashyap Anil Owen Lamont and Jeremy Stein ldquoCredit Conditions and theCyclical Behavior of Inventoriesrdquo Quarterly Journal of Economics CIX(1994) 565ndash93

Lamont Owen ldquoCash Flow and Investment Evidence from Internal Capital Mar-ketsrdquo manuscript University of Chicago 1996

Li G ldquoRobust Regressionrdquo in Exploring Data Tables Trends and Shapes D CHoaglin F Mosteller and J W Takey eds (New York John Wiley amp Sons1985) pp 281ndash340

Murray Ronald William Decker and Nelson Dittmar The Coopers amp LybrandSEC Manual (New Jersey Prentice Hall 1993)

Myers Stewart and Nicholas Majluf ldquoCorporate Financing and Investment Deci-sions When Firms Have Information That Investors Do Not Haverdquo Journalof Financial Economics XIII (1984) 187ndash221

Perfect Steven and Kenneth Wiles ldquoAlternative Constructions of Tobinrsquos q AnEmpirical Comparisonrdquo Journal of Empirical Finance I (1994) 313ndash41

Poterba James ldquoComments on Fazzari Hubbard and Petersenrdquo Brookings Pa-pers on Economic Activity (1988) 200ndash04

Rajan Raghu ldquoInsiders and Outsiders The Choice between Informed and ArmrsquosLength Debtrdquo Journal of Finance XLVII (1992) 1367ndash1400

Schiantarelli Fabio ldquoFinancing Constraints and Investments A Critical ReviewrdquoWorking Paper Boston College May 1995

Sharpe Steven ldquoAsymmetric Information Bank Lending and Implicit ContractsA Stylized Model of Customer Relationshipsrdquo Journal of Finance XLV(1990) 1069ndash87

Stulz Rene ldquoManagerial Discretion and Capital Structurerdquo Journal of FinancialEconomics XXVI (1990) 3ndash28

Whited Toni ldquoDebt Liquidity Constraints and Corporate Investment Evidencefrom Panel Datardquo Journal of Finance XLVII (1992) 1425ndash60

APPENDIX FINANCIAL STATUS BY FIRM-YEAR BY SUBPERIOD AND BY ENTIRE PERIOD

Distribution of nancing constraints by year for 49 low-dividend rms from Fazzari Hubbardand Petersen [1988] from 1970 to 1984 Firm nancing constraint status for each year is not nan-cially constrained (NFC) likely not nancially constrained (LNFC) possibly nancially constrained(PFC) likely nancially constrained (LFC) or nancially constrained (FC) For subperiods and entireperiod rms are NFC if rms are not nancially constrained (NFC) every year rms are NC if rmsare not or likely not nancially constrained (NFC or LNFC) every year PFC if rms are possiblynancially constrained (PFC) in some year and FC if rms are likely or denitely nancially con-strained (LFC or FC) in some year

Company 1970 1971 1972 1973 1974 1975 1976 1977

Barry NFC NFC NFC NFC NFC NFC NFC NFCDWG LNFC LNFC LNFC LNFC LNFC LNFC LNFC LNFCDigital

Equipment LNFC LNFC LNFC LNFC NFC NFC NFC NFCFluke LNFC NFC LNFC LNFC LNFC LNFC LNFC LNFCForest Labs NFC NFC LNFC NFC NFC NFC NFC NFCGCA NFC NFC NFC NFC NFC LNFC NFC NFCHelene Curtis LNFC NFC NFC NFC LNFC NFC NFC NFCHewlett Packard LNFC LNFC LNFC LNFC LNFC NFC NFC NFCIroquois Brands LNFC NFC NFC NFC NFC LNFC NFC NFCJames River NFC NFC NFC LNFC NFC NFCNational Semi NFC NFC NFC NFC NFC NFC NFC NFCScientic Atlanta LNFC LNFC LNFC LNFC LNFC NFC LNFC NFCSouthdown NFC NFC NFC NFC NFC LNFC LNFC LNFCThermo Electron NFC NFC NFC LNFC LNFC LNFC NFC NFCTrico LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCWang Labs NFC NFC LNFC NFC NFC LNFC NFC NFCWinnebago LNFC NFC NFC LNFC NFC LNFC LNFC NFCCameron Iron

Works NFC NFC NFC NFC NFC NFC NFC NFCData General NFC NFC NFC NFC NFC NFC NFC NFC

Coachmen LNFC LNFC LNFC LNFC LNFC NFC NFC NFCCoherent LNFC LNFC LNFC LNFC LNFC PFC LNFC PFCCommodore Intl PFC PFC LNFC LNFCNucor PFC PFC NFC NFC NFC NFC NFC LNFCPlantronics NFC NFC NFC NFC NFC NFC NFC NFCTeradyne NFC NFC NFC NFC PFC NFC LNFC LNFCVernitron PFC PFC NFC NFC NFC LNFC NFC NFCWhitehall NFC NFC NFC NFC NFC LNFC LNFC LNFC

Adams Russell LNFC LNFC PFC LNFC LNFC LNFC LNFC NFCAnalog Devices PFC PFC PFC LNFC LFC PFC LNFC NFCApplied

Magnetics LNFC LNFC LNFC NFC LFC PFC PFC FCAydin FC LFC LFC PFC LNFC NFC NFC LNFCChampion Home NFC LNFC NFC PFC FC PFC FC FCColeco LNFC LNFC NFC PFC PFC PFC LNFC FCCompugraphic LNFC LNFC LNFC LNFC LNFC LNFC NFC NFCControl Data LNFC LFC LNFC LNFC LFC LNFC LNFC NFCCordis LNFC PFC LNFC LNFC LFC FC FC LNFCGalveston

Houston LNFC PFC LFC LNFC LNFC LNFC NFC NFCGerber Scientic PFC NFC LNFC LNFC FC LFC LNFC NFCHesston NFC NFC NFC NFC NFC NFC LFC FCIntl Rectier LNFC PFC PFC LNFC NFC LNFC NFC NFCKaty Inds PFC LNFC NFC NFC PFC FC LFC PFCMohawk Data

Sciences NFC LNFC PFC FC FC LFC LNFC LNFCRaychem PFC LNFC LNFC LNFC LFC LFC LNFC LNFCRecognition

Equipment LNFC LNFC NFC FC LFC LNFC LNFC LNFCRockcor NFC LFC LFC LNFC LNFC LNFC LNFC NFCRogers PFC PFC PFC LNFC PFC LFC NFC LNFCSCI Systems LFC LFC PFC LFC LFC LNFC LNFC LNFCTyson Foods LNFC PFC LNFC LNFC LFC LNFC NFC NFCUS Surgical FC LFC LFC LFC PFC PFC NFC NFC

APPENDIX CONTINUED

Overall Overall Overall1970ndash 1978ndash 1970ndash

1978 1979 1980 1981 1982 1983 1984 1977 1984 1984

NFC NFC NFC LNFC LNFC LNFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NC

NFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC NFC NFC NFC NC NC NCLNFC LNFC NFC NFC LNFC NFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC NFC NC NC NCNFC NFC NFC NFC LNFC LNFC LNFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC LNFC LNFC LNFC NFC LNFC NFC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC LNFC LNFC NFC NFC NC NC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC NFC NFC NFC NFC NFC NFC NC NFC NCNFC LNFC NFC NFC NFC NFC NFC NC NC NC

NFC NFC NFC NFC NFC NFC NFC NFC NFC NFCNFC NFC NFC NFC NFC NFC NFC NFC NFC NFC

NFC PFC LNFC LNFC NFC NFC NFC NC PFC PFCLNFC LNFC PFC LNFC LNFC NFC NFC PFC PFC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC NFC NFC PFC NC PFCLNFC LNFC NFC NFC NFC PFC LNFC NFC PFC PFCLNFC NFC NFC NFC NFC NFC NFC PFC NC PFCNFC NFC NFC NFC NFC NFC NFC PFC NFC PFCNFC PFC LNFC NFC NFC NFC NFC NC PFC PFC

NFC NFC NFC NFC NFC LNFC LFC PFC FC FCNFC NFC LNFC NFC NFC NFC NFC FC NC FC

LNFC LNFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLFC LFC LFC NFC NFC NFC NFC FC FC FCFC PFC NFC NFC NFC LFC LFC FC FC FCNFC NFC PFC FC NFC NFC NFC NC FC FCNFC NFC NFC NFC NFC NFC LNFC FC NC FCLNFC NFC LNFC LNFC LNFC LNFC LNFC FC NC FC

NFC PFC LNFC NFC LNFC LNFC LFC FC FC FCNFC PFC NFC LNFC NFC NFC NFC FC PFC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FCNFC NFC NFC LNFC LFC LNFC NFC PFC FC FCPFC LNFC NFC NFC NFC NFC LNFC FC PFC FC

NFC NFC NFC PFC PFC LFC FC FC FC FCNFC NFC NFC NFC LNFC LNFC LNFC FC NC FC

LNFC LNFC NFC PFC FC LNFC LNFC FC FC FCNFC NFC NFC NFC NFC LNFC NFC FC NC FCNFC LNFC LNFC PFC LNFC LNFC NFC FC PFC FCLNFC NFC NFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FCLNFC LNFC LNFC NFC NFC NFC NFC FC NC FC

INVESTMENT-CASH FLOW SENSITIVITIES 215