In vestment, Dividend, and External Finance Behavior … · significant dependence of the dividend policy of electric utility firms on ... On the other hand, the rather marked reliance

This PDF is a selection from an out-of-print volume from the National Bureauof Economic Research

Volume Title: Determinants of Investment Behavior

Volume Author/Editor: Robert Ferber, editor

Volume Publisher: NBER

Volume ISBN: 0-87014-309-3

Volume URL: http://www.nber.org/books/ferb67-1

Publication Date: 1967

Chapter Title: Investment, Dividend, and External Finance Behavior ofFirms

Chapter Author: Phoebus J. Dhrymes, Mordecai Kurz

Chapter URL: http://www.nber.org/chapters/c1243

Chapter pages in book: (p. 427 - 485)

In vestment, Dividend, and External

Finance Behavior of Firms

PHOEBUS J. DHRYMESUNIVERSITY OF PENNSYLVANIA

AND

MORDECAI KURZSTANFORD UNIVERSITY AND HEBREW UNIVERSITY, ISRAEL

Preliminaries and SummaryThe relations among the investment, dividend, and external financebehavior of firms are often alluded to but seldom studied systematically.Quite clearly, given the institutional milieu of the modern corporation,there exists at least a presumption that these three aspects of the firm'sdecision-making process exhibit some interaction. Yet, in the currentliterature the view is frequently advanced that investment decisions aretaken on solely "real" (nonfinancial) considerations, that dividendpolicies are characterized by a considerable degree of inertia, and thatthe financing of investment by internal or external funds is a mere detail.

To our knowledge, an explicit link among these three decisions hasnot been proposed and econometrically implemented. It is the purposeof this paper to study the question of establishing such a link and toelucidate the extent of the interdependence of these decisions. The

Nom: We wish to acknowledge our gratitude to L. R. Klein and Edwin Kuhfor several helpful suggestions and comments on an earlier version of this paper.We should also like to acknowledge the able research assistance of George Schink,Pietro Balestra, and Hajime Oniki. Financial support was in part provided throughNSF grant GS 571 at the University of Pennsylvania, Office of Naval ResearchGrant 225 (50) at Stanford University, and through computer time made availableto us by the Computer Center of the University of Pennsylvania.

428 Producer Durables

approach employed makes use of a series of cross sections. Specifically,our sample consists of 181 industrial and commercial firms for whicha continuous record exists over the period 1947—60. The "structural"form of this interrelated system will be estimated successively over theperiod 1951—60 in order to test the stability of structure, especially asit is related to business fluctuations.

Our main findings are the following: (1) A strong .interdependenceis evident between the investment and dividend decisions; the externalfinance activities of firms seem to be affected by the other two aspectsof the finn's operation but do not seem much to affect them, exceptpossibly during upswings or peaks. (2) There is compelling evidence tosuggest that in estimating the structure one ought to use full informationmethods. (3) Elements of both the accelerator as well as the rate ofprofit theories of investment seem necessary in order to explain theempirical behavior of investment.

A Brief Review of the LiteratureThe three aspects of the firm's behavior on which this study is focusedhave been studied in the literature with varying degrees of intensity.Thus investment behavior has perhaps been studied most intensively.

The integration of investment theory with the neoclassical theory ofthe firms can be traced to Roos [20], Tmbergen {22, 23], and Klein[21].' An extensive survey of the work of the last two authors can be.found in Meyer and Kuh [17]. Subsequent contributions have beennumerous but are best omitted here since rather recently Eisner andStrotz [7] have provided a very complete review of such works.

It is perhaps accurate to say that the main results of such studieslie in providing tests concerning the empirical relevance of the accelera-tor, capacity-accelerator, or profits (or rate of profit) theories of invest-ment. The issue is not yet satisfactorily resolved, but it appears thatneither the capacity accelerator nor the profits theory is alone sufficient,but rather a combination of elements of both is probably necessary toprovide a satisfactory account of the empirical behavior of investment.

In this connection it seems appropriate to cite, in some detail, a veryimportant recent study by Kuh [13], which is, in some respects, similarto the one we propose to pursue here. Kuh investigates the investment,dividend, and external finance aspects of the firm's behavior in thefollowing context. His basic sample consists of sixty industrial firms forwhich a continuous satisfactory record exists over the years 1935—55.

1 Numbers in brackets refer to Bibliography at end of paper.

Finance Behavior of Firms 429

This sample. was arrived at from a larger one by a process of selectionwhich eliminated firms that were merged into others over the sampleperiod, as well as those that were "too large"—owning gross assets over$120 million in 1953.

The work is divided in two distinct parts, a theoretical and an empir-ical one. In the theoretical part the interdependence of the three deci-sions is quite clearly recognized and an integrated model is presentedcombining the capacity-accelerator model of investment with theLintner [16] model of dividend behavior. The empirical part, unfor-tunately, does not carry out this integration. There, investment is givenby the usual capacity-accelerator model, so that it depends on thecapital stock and sales, as well as on the observed sales-capital ratio,the latter being an approximation to the desired output-capital ratioof the accelerator theory. In this connection models are also tried inwhich the sales variable is replaced by a profit variable. It is found,however, that the former models are more in accord with the data thanthe profit ones. Dividends are explained by the well-known Lintnerhypothesis, on which we have commented at leiigth elsewhere [3]. Theyare thus made to depend on profits and past dividends, the modelbeing essentially an adaptation of the usual flexible accelerator modelof investment; here the role of the capital coefficient is played by thedesired dividend-payout ratio.

Finally, external finance behavior is derived residually through thebudgetary requirement that investment expenditures must equal retainedearnings plus depreciation allowances plus external finance.

Thus, in this context there is a certain direction of causality; invest-ment is independent of dividends and external finance; dividends dependonly on profits (and lagged dividends) which may depend on invest-ment, although the dependence is not explicit. Finally, external financeis more or less residually derived, and hence would depend on invest-ment and dividends. Although Kuh does not state his model in theseterms, the above is, we think, a very plausible interpretation of thespirit in which his econometric investigation is carried out.

His is a penetrating examination of the problems relating to testinghomogeneity and stability of structures by an ingenious combination ofcross-section and time-series data; the problem of simultaneity andinterdependence to which our paper is addressed is, however, entirelyoverlooked in his econometric investigation. Kuh's estimation is carriedout by single-equation methods and his results seem to bear out thecapacity accelerator theory of investment and the homogeneous Lintnerhypothesis on dividends.


A more recent study of investment behavior is that of Jorgenson [9].His formal model considers the firm as determining its investment policythrough maximization of net worth, defined as the present discountedvalue of current receipts minus outlays, where the latter includes, inaddition to remuneration of the variable factors, expenditures on capitalgoods. However, his operational model through the convenient assump-tion of the Cobb-Douglas production function essentially reduces tothe accelerator model. His major innovation is the introduction of gen-eral lag structures between the demand for additional capital stock, asarising because of a discrepancy between desired and actual capitalstock, and the augmentation of the latter through deliveries of capitalequipment. Another innovation consists of the systematic treatment ofthe replacement demand for capital. His sample consists of quarterlyobservations on certain classes of industrial aggregates.

Concerning the dividend behavior of firms, in addition to the workof Kuh [13] referred to above, one should mention the pioneering workof Lintner [15, 16]. The same approach is more or less followed byDarling [2], whose innovation consists in making the dividend functiondepend on the change in sales; the apparent success of this is probablya reflection of the type of dependence we wish to investigate here. Inthis connection it should be pointed out that we [3] have determineda substantial impact of investment on the dividend decision. Anotherrecent study of dividend behavior is that of Brittain [1]. He follows theLintner model quite closely, but reports that profits gross of deprecia-tion is a better explanatory factor of dividend behavior than is profitsnet of depreciation.

The question of internal finance has not been systematically treatedin the econometric literature, save for the work of Kuh which has aslightly different orientation from the one pursued in this paper. Thus,although frequent allusions are made in the literature to the inter-dependent character of these three aspects of the firm's behavior, ratherlittle has been done in carrying out the implications of this view to theform of the econometric model to be estimated. One way of rational-izing this is to consider the current view in the literature as holdingthat the problem is being resolved in a sequential manner. Thus, invest-ment decisions are implicitly viewed as being taken solely (or chiefly)on the basis of "real" considerations, and then once this decision isresolved the question of how to finance this undertaking is considered.

In the case of dividend behavior, the main work on the subject, viz.,Lintner's [15, 16], views dividend disbursals as totally divorced frominvestment considerations. Yet, as noted above, we have determined a


significant dependence of the dividend policy of electric utility firms onactual and anticipated investment. As an empirical matter Kuh andMeyer [14] report that 75 per cent of investment in manufacturing isinternally "financed." The same authors in a previous work [17] list animpressive catalogue of reasons why such a preference might exist.While this phenomenon may, in some part, be due to the peculiaritiesof the tax structure of the United States it may also reflect imperfectionin the capital market.

The view taken in this paper is rather simply put as follows. Quitegenerally a firm faces an outflow of funds represented by its variableand fixed costs, tax and dividend payments, as well as by its investmentactivities. On the other hand, it can rely on an inflow of funds repre-sented chiefly by its sales and the proceeds through various forms ofexternal finance, viz., by bond or stock flotation. To the extent that aplausible objective for a firm is to grow, provided its operations areprofitable, and that the capital market is less than perfect, it wouldfollow that investment and dividend outlays are quite clearly competitive.

On the other hand, the rather marked reliance on internal fundssignifies a strong aversion to the use of the capital market. Thus, itwould seem quite reasonable to suppose that the three decisions—toinvest, to pay dividends, and to resort to external funds—are mutuallydetermined. Hence, it is desirable to investigate this problem in thecontext of a simultaneous-equation model. If our conjecture about themodus operandi of the system is correct, then we should expect thatthe coefficients of the jointly determined variables—investment, divi-dends paid, and external finance—would be significant, at least inseveral instances, where they serve as explanatory variables.

The ModelTHE STRUCTURE OF THE MODEL

In the discussion of the previous section it was pointed out that inour view a deficiency in the econometric investigations of the invest-ment aspect of the firm's behavior is that the interaction of investmentwith certain financial variables was substantially overlooked.

Our objective here is to present a somewhat general equilibriummodel of the firm's activities which attempts to remedy this deficiencyand at the same time is sufficiently simple so as not to obscure the essen-tial elements of the problem. Thus, at this stage, we shall not explicitlyconsider the technological constraint on the firm's activities in the formof a production function, nor the institutional characteristics of the


factor and product market in which it operates, in the form of firm orindustry factor supply and product demand functions.

We shall rather assume that production and marketing decisionspredate the decision process we wish to study. The view we implicitlytake is that investment affects output (and profits) only with a lag sothat current investment affects only future output, and hence profits,and that current output and hence current profits are not affected bycurrent investment.

While this may not be a perfectly valid view, it is, we believe, suffi-ciently accurate to serve at least as a rather useful simplification. Wewould think that the gains in simplicity obtained thereby more thanoffset any increased incidence of simultaneous-equation bias. Beyondthat, it has of course been traditional to treat sales and profits, bothcurrent and lagged, as valid predetermined variables in cross-sectionstudies on investment.

The general (schematic) structure of the model is as follows:

Ii = D, EF1, EF2; X1, X2, . . .

12 f2(11, D, EFI, EF2; X1, X2, . . .

D = f3(I1, '2, EF1, EF2; X1, X2, . . . (1)

EF1 = f4(I1, '2, D, EF2; X1, X2, . . .

EF2 = f5(I1, '2, D, EEl; X1, X2, .. .

Where is investment in fixed assets; '2 is inventory and other shortterm investments; D is common stock dividends paid; EF1 is (net)external finance obtained by borrowing; EF2 is (net) external financeobtained by stock flotation; the are predetermined variables, I = 1, 2,• . . n. The predetermined variables may include profits, depreciation,sales, long-term debt outstanding, etc., and will be introduced explicitlyas the occasion arises.

In addition, the firm faces the "budget constraint"

Ii+12=EF1+EF2+P—D+Dep, (2)

where P and Dep denote, respectively, profits and depreciation allow-ances.

If we use the constraint in (2), we can eliminate one of the (endog-enous) variables Of the system in (1). We have chosen to eliminateEF2; this was done chiefly because data on this variable were verydifficult to obtain with any reasonable degree of reliability. At any ratestock flotation as a source of external finance, while not negligible, is

TAB

LE 1

Sour

ces o

f Gro

ss E

xter

nal F

inan

ce in

Man

ufac

turin

g, M

inin

g,C

omm

erci

al, a

nd O

ther

Firm

s, 19

54 -5

6

Sou

rce:

Sta

tistic

alB

ulle

tin, S

ecur

ities

and

Exc

hang

e C

omm

issi

on, F

ebru

ary

1957

.

0 0

Type

of F

irm

Bon

ds

Mill

ion

Dol

lars

Per

Cen

t

Pref

erre

d

Mill

ion

Dol

lars

Stoc

k

Per

Cen

t

Com

mon

Mill

ion

Dol

lars

Stoc

k

Per

Cen

t

Tota

l(m

illio

ndo

llars

)

1956

Man

ufac

turin

gM

inin

gC

omm

erci

al &

oth

erTo

tal

3,00

030

025

93,

559

80.5

63.3

79.4

78.6

164 17 8

189

4.4

3.6

2.5

4.2

562

157 59 778

15.1

33.1

18.1

17.2

3,72

647

432

64,

526

1955

Man

ufac

turin

gM

inin

gC

omm

erci

al &

oth

erTo

tal

2,04

319

834

02,

581

68.2

47.8

76.7

67.0

165 10 21 196

5.5

2.4

4.8

5.1

786

206 82

1,07

4

26.3

49.8

18.5

27.9

2,99

441

444

33,

851

'954

Man

ufac

turin

gM

inin

gC

omm

erci

al &

oth

erTo

tal

1,87

734

731

72,

541

82.8

64.4

75.3

78.7

228 14 62 304

10.0 2.6

14.7 9.4

163

178 42 383

7.2

33.0

10.0

11.9

2,26

853

942

13,

228


of minor significance compared with bond flotation for most years andindustries in our analysis. Table 1 will bear this out. Since our samplewill consist largely of manufacturing and retail trade firms with a rathersmall representation of mining firms, it follows that our selection ofbond finance as the principal source of external funds to be studied isnot likely to lead to serious deficiencies.

Finally, we have chosen to regard short-term investment as a pre-determined variable, so that our final system of equations to be esti-mated is reduced to three, viz., the dividend, the (fixed) investment,and the external finance equations. This last decision carries with it,in principle, serious deficiencies.

There are, however, several ameliorating circumstances: first, to theextent that a component of short-term investment is unintended inven-tory accumulation, we are clearly not committing a specification error;secondly, to the extent that the data for this series includes creditadvanced to clients in the normal conduct of business (accounts receiv-able) this does not constitute a misspecification either; similarly foraccounts payable; but to the extent that the series contains a componentof intended inventory accumulation and short-term securities holdingsfor other than liquidity-transaction purposes, we are clearly committinga specification error. Unfortunately no detailed breakdown of the short-term investment series was available, so we have chosen to treat it as apredetermined variable. Thus, the model finally estimated is of the form

D = g1(I,EF1;X1,X2,...

I g2(D, EEl; X1, X2,... , (2)

EF1 = g3(D, I;X1,X2, . . .

GENERAL COMMENTS ON THE FORM OF THE EQUATIONS

The Dividend Equation. One can look upon dividend disbursals asconveying information to the market on the inherent profitability of thedisbursing firm as Modigliani and Miller [19] inter alios have argued.In fact they would contend that the dividend series contains "moreinformation" than the profit series. Hence, it would appear that it is thepolicy of firms to maintain a steady dividend per share and to adjust itupward and downward only when a "permanent" change in their eco-nomic environment has taken place. As a matter of fact, it is more orless common for firms to maintain a constant dividend per share. Butthis in no way implies constancy in the dividend-profit ratio.

It is reasonable to suppose that dividend disbursals will depend onthe rate of profit of the firm, its investment plans, and the external


finance obtained through the bond market; the rationale for this lastvariable would be that external finance will enable the firm to carry outits planned dividend disbursals even when the rate of profit is low andinvestment programs are extensive.

The Investment Equation. The foundations of investment theory inthe theory of the firm are too well-known to require repetition here.Clearly from this point of view investment would depend either onchanges in the volume of output or on its rate of profit, which may betaken to lead to changes in the expected profitability of new investment.These two considerations are not totally unrelated, especially if the firmis assumed to operate with a neoclassical production function allowingsubstitution; if no substitution is allowed, then it is not clear that therate of profit has any place in the investment function.

Our innovation here consists in introducing the other two jointlydependent variables, viz., dividends and external finance. We havealready given some indication as to why we consider these variablesrelevant.

Clearly dividend disbursals and investment outlays represent compet-ing demands on the resources available to the firm; thus it would bequite plausible to suppose that the investment activities of the firm willbe affected by its dividend activities; postponement or curtailment ofinvestment could conceivably result because of inability of the firm tocarry out a given investment program, "optimally" determined by some"rational" criteria, and at the same time continue to make "satisfactory"dividend payments. To call such action "irrational" may be giving acorrect label to the phenomenon but will not alter the facts—if, indeed,such is the factual state of affairs.

it would also be of interest to inquire whether such variables asdepreciation (on this, see Meyer and Kuh [17]) are signfficant determi-nants of investment; if depreciation is an accurate index of deteriora-tion of the capital stock due to its employment in the productiveprocess, then depreciation would describe accurately that part of invest-ment undertaken for replacement purposes. There are good reasons tobelieve, however, that depreciation does not accurately measure theusing up of capital, and hence its introduction in the investment equa-tion would only serve to portray more accurately the resources avail-able to the firm for investment and dividend outlays. In addition, thereis the question of the proper lags operating in the investment process;thus, it would be of interest to ascertain whether lagged rates of changeof sales or past rates of profit significantly affect the decision to invest.


The introduction of the bond finance variable here has a motivationbest understood in terms of imperfect capital markets. Thus, if in agiven universe all firms belong to more or less the same uncertaintyclass, then market discrimination might be expected to take the formof restricting the amount a firm can borrow without raising the cost ofobtaining long-term funds. Hence, we may conjecture, ceteris paribus,that the easier the access to this market—either in the amounts or inthe terms on which the loans are granted—the larger the investmentprogram a firm may undertake. Thus, in the investment equation divi-dends may be expected to have a negative impact, while externalfinance will have a positive one.

The External Finance Equation. Enough has been said in connectionwith the other two equations to make the hypothesized form of theexternal finance equation clear. One would expect to have this variabledepend positively on investment, negatively on the market interest rate,and negatively on depreciation and profits. The relationship of externalfinance to dividends, however, is not very clear-cut. Thus, it is possibleto argue that essentially because of a budgetary constraint, more divi-dends, other things being equal, mean more borrowing. But it is equallyplausible to argue that for firms that are no longer growing rapidlymore dividends need not induce further borrowing simply because theirinvestment activities are somewhat restricted. Thus, there should not beany feedback from dividends to external finance.

Finally, it would be interesting to inquire whether there is anyempirical merit to Kalecki's principle of increasing risk [11], i.e.,whether the rate of borrowing depends on the magnitude of long-termdebt already outstanding relative to the size of equity claims.

SOME ARGUMENTS FOR SIMULTANEOUS ESTIMATION OF THE MODEL

The question may perhaps arise as to why one might wish to applysimultaneous estimation techniques in estimating the model presentedabove. It might be argued that, since the firm represents a single unit ofdecision, the proper manner of looking at its decision-making processis one that views the firm as reacting to its economic environment, i.e.,as making decisions in terms of the exogenous variables confronting it.Hence the appropriate econometric model should be a reduced form.After all we do not construct simultaneous-equation models of a house-

- hold's consumption decisions.This view, however, overlooks the simple institutional fact that the

modern firm is a complex organization with a considerable degree ofdecentralization. The decisions made by one department have an impact


on those made by another. Hence each department may be viewed asoperating with certain decision rules which depend not only on exog-enous variables but also on decisions arrived at by other departments,i.e., on variables which are endogenous to the firm. But in addition afirm has also a certain cohesiveness and such decisions must in somesense be made consistent, or the organism may not survive.

In this view of the firm's operations, the role of the highest executiveechelons, say, the president and the board of directors, is rather unique.Their function is not to be the actual decision makers in the firstinstance, but rather to receive proposals, examine priorities, attempt tocoordinate proposals, and make departmental decisions consistent witheach other. In other words, it is their function to set the decision rulefor each department and insure the proper and consistent execution ofsuch rules. If one views the firm in this context, then it is quite clearthat one should wish to employ simultaneous-equation techniques inestimating the parameters of the model given above. Thus, given thatwe hypothesize the existence of a meaningful structure of intrafirmdecision-making, it is apparent that we should wish to estimate it bythe most efficient econometric techniques. Furthermore, in so doing wemay gain greater insight into the operations we wish to study, an insightthat would be foregone if we simply confined ourselves to what in thecontext of our model would be reduced-form estimation.

Empirical ImplementationSAMPLE CHARACTERISTICS AND DEFINITION OF VARIABLES

The sample on which our study is based consists of 181 firms forwhich a continuous satisfactory record exists between 1947 and 1960.These firms are largely manufacturing and retail trade ones, althoughseveral are chiefly engaged in mining activities.

The sources of our data are the balance sheets and income statementsof individual firms appearing in various issues of Moody's Manuals,although in some cases (about twenty) where data ambiguities orinconsistencies arose, the matter was rectified by direct correspondencewith the firms involved.

It seems desirable at this stage to give a brief description of thecharacteristics of the firms in our sample, in terms of the type of activitypursued and of their size. Because the firms we used belong to numer-ous industrial subclassifications, we have stratified our sample throughreclassification of firms into nine industrial classifications. This wasdone on the basis of the type of activity pursued by the individual firms.


TABLE 2

Classification Scheme

VariableDescription ofIndustrial Classification

Numberof Firms

Per Centof Total

C1 Transportation equipment 17 9.32C2 Retail trade stores 21 11.61C3 Alcohol, tobacco, and food 21 11.61C4 Rubber, petroleum, and chemical

products 30 16.58C5 Machine tools, agricultural equipment,

and accessories 20 11.08C6 Electrical equipment and appliances 13 7.19C7 Building materials and equipment 18 9.95C8 Te,ctiles, glass, pulp, and paper

products 25 13.82C9 Mining, steel, steel products, and

nonferrous metals 16 8.85

Total 181 100.00

TABLE 3

Size Distribution of Sample Firms, Ranked by Book Valueof Capital Stock, 1956

Size of Firm(million dollars)

Nof

umberFirms

Per Centof Total

0-49.9 18 9.9550-99.9 41 22.65

100-149.9 14.36150-199.9 15 8.29200-299.9 18 9.95300-499.9 17 9.39500-999.9 20 11.05

1,000 and over 14.36

Total 181 100.00

— - -- -.

Finanée Behavior of Firms 439

A complete enumeration of the firms in our sample, grouped byindustrial classification, as just explained, appears in the Appendix atthe end of this paper. To reflect the sample stratification, we resortedto the standard analysis of covariance device of introducing (dummy)variables assuming the value 1 if a firm belongs to, say, the 1thi classi-

fication and zero otherwise. The variables corresponding to these classi-fications appear in Table 2.

It would appear from Table 2 that we have a very good cross sectionof mining and manufacturing firms, and quite a number of large retailstores. Our sample does not weigh very heavily any particular classifi-cation, although our firms tend to be mostly medium-sized and largeones. This is quite evident from Tables 3 and 4.

Table 3 classifies the sample firms by size of the (book value of the)capital stock. The median firm has capital stock of $160.2 million,the smallest and largest firms having, respectively, $16.5 million and$10,265.2 million. The smallest firms have a rather weak representa-tion (9.95 per cent of the sample), while the largest firms have asomewhat stronger one (14.36 per cent of the sample). Essentially thesame situation emerges if we classify the sample firms by size of sales.This classification appears in Table 4.

In this ranking, the median firm has sales of $438.7 million, thesmallest and largest firms having, respectively, $38.2 million and$12,736.0 million. Again, the smallest firms are somewhat more weakly

TABLE 4

Size Distribution of Sample Firms, Ranked by Sales, 1956

Size of Firm(million dollars)

Nof

umberFirms

Per Centof Total

0-99.9 17 91.39

100-199.9 24 13.26200-299.9 29 16.02300-499.9 28 15.47500-749.9 24 13.26750-999.9 16 8.84

1,000-1,499.9 20 11.051,500- 23 12.71

Total 181 100.00


represented (9.39 per cent of the sample) than the largest firms(12.71 per cent). Thus, our sample appears to be well stratified, interms of size as well as the type of activity pursued.

At this stage it is desirable to catalog and explain briefly the natureof the variables entering into our investigation. We have already given,in Table 2 above, the dummy variables associated with the industrialclassifications. In addition, the following basic variables are employed(this list will also serve to define the symbolism employed in the sub-sequent development of our discussion):

St—sales at time t, undeflated.

(EF1 ) i—long-term borrowing-external finance-at time t; this is sim-ply the first difference of the book value of long-term debt outstandingand thus represents net current long-term borrowing; it should beremembered that this measure is somewhat biased by the transfer ofmaturing long-term debt to the short-term category.

Dc—dividends (common) paid at time t.Is—gross fixed investment at time t.Ks—book value of the capital stock at beginning of time 1.Ps—net profits (after taxes) at time t, undeflated.(LTD) c—net long-term debt outstanding at time t, in nominal terms.(Dep) t—depreciation allowances at time t.Nt—net current position of the firm at time t, defined as the excess

of inventories, cash, short-term securities, and accounts receivable overaccounts payable and other short term liabilities.

Re—interest payments at time t, on long-term debt outstanding. Thisis admittedly a very poor measure of the relevant interest rate but it isthe only one available.

In actually carrying out the empirical implementation of the model,we have chosen to normalize the jointly dependent variables by Thiswas done for two reasons: first, it tends to reduce heteroscedasticityand hence make the stochastic characteristics of our sample correspondmore closely to the standard specification of the simultaneous-equationmodels; second, since our objective is to isolate the. determinants of theinvestment-dividend-external finance decision process, this procedureprevents our results from being unduly influenced by large firms simplybecause of sheer size. Another related reason is the fact that one wouldnot expect the relation between investment and the appropriate accelera-tor variable to be identical in the case of a retail store and an aircraftmanufacturer. By relying on "intensive" variables, one tends to over-come such problems.


A list of the predetermined variables actually employed is givenbelow:

N/K enters the model as a consequence of the use of the budgetconstraint to eliminate one of the equations of the system; the normal-ization employed here is to some extent motivated by portfolio theoryconsiderations.

Dep/K represents the portion of the book value of the capital stock

TABLE 5

Simple Correlation Matrices, a Endogenous Variables, 1951 -60

Variable D/S I/S EFI/S D/S I/S EF1/S

1951 1956

fl/S 1.0000 .3018 —0.0854 1.0000 .3238 .0661I/S 1.0000 .3735 1.0000 .4932

EF1/S 1.0000 1.0000

D/S 1.0000

1952

.2906 .0610 1.0000

1957

.3871 .1522

I/S 1.0000 .6649 1.0000 .7414

EF1/S 1.0000 1.0000

D/S 1.0000

1953

.2793 .1630 1.0000

1958

.4537 —0.0279f/S 1.0000 .3346 1.0000 .3085EF1/S 1.0000 1.0000

fl/S 1.0000

1954

.1410 —0.0460 1.0000

1959

.3049 —0.0316I/S 1.0000 .3361 1.0000 .1012EF1/S 1.0000 1.0000

D/S 1.0000

1955

.2816 —0.0499 1.0000

1960

.3752 .0116f/S 1.0000 —0.0200 1.0000 .0274EF1/S 1.0000 1.0000

Critical values of In for rejecting the hypothesis of zero correlation at the5 and 1 per cent levels, respectively, are approximately .149 and .218.


written off as depreciation charges; its form is related to the followingbasic variable.

P/K is the rate of profit; it would have been better perhaps to havedefined the numerator of this fraction as profits plus depreciation plusinterest charges on the ground that, since it measures the (average)rate of return on the firm's capital resources, this ought to be measuredgross of irrelevant bookkeeping items such as depreciation and interestcharges.

— is the usual accelerator variable exceptthat it is normalized by a lagged value of sales. It was felt, however,that it is the pressure of sustained relative increases in sales that affectsinvestment.

LTD/(K — LTD) is the leverage variable employed to test the prin-ciple of increasing risk. It is probably not a very accurate one, therationale behind it being that businessmen are influenced by bookrather than market value considerations.

EMPIRICAL RESULTS

It is instructive before we proceed to have a brief look at the matrixof simple correlation among the jointly dependent variables. This isgiven in Table 5.

Two features of these results are of particular interest; first, wheneversignificant, such correlations between dividends and investment are uni-

TABLE 6

Cyclical Peaks and Troughs, Identified byQuarter, 1948 - 62

Year QuarterPeak orTrough

1948 4th Peak1949 2nd Trough1953 2nd Peak1954 2nd Trough1957 3rd Peak1958 1st Trough1960 2nd Peak1961 1st Trough1962 2nd Peak


formly positive, a result that is somewhat surprising and difficult toaccept at face value. Intrinsically these two quantities would not beexpected to vary in the same direction. It will indeed be found that, inthe proper structure to be estimated below, the impact of dividendson investment and vice versa is (contemporaneously) a negative one.

Secondly, it turns out that the simple correlation between dividendsand investment is insignificant only for 1954 (a trough year as indicatedby Table 6), that between investment and external finance is insignifi-cant for 1955, 1959, and 1960 (in Table 6, all upswing or peakyears), while the correlation between dividends and external finance issignificant only for 1953 and 1957 (both peak years).

To the extent that one might wish to generalize on the basis of theseresults, it would appear that they simply demonstrate the substantialreliance of firms on internal financing. Apparently when profits increaserapidly in an upswing they are quickly absorbed into investment, obvi-ating the need to resort to the capital market. On the other hand, whenat the peak profits begin to get squeezed, we get a significant correlationof dividends with external financing, indicating that another sort ofpressure is brought to bear on the firm's resources that forces increasedreliance on the capital market.

THE SCOPE OF THE INVESTIGATION AND SOMETYPICAL SINGLE-EQUATION ESTIMATES

As stated earlier, it is our purpose to investigate the dividend, invest-ment, and external finance decision-making process of the firm. Ourcontention is that these decisions are jointly determined and hence thatthe proper method of estimation is a simultaneous-estimation approach.Moreover, it is desirable to test the stability of structure as estimated.For one may argue that such simultaneity established by means of onlyone cross-sectional investigation is at best only a weak demonstration(in this connection, see Meyer and Kuh [12]). If, however, the struc-ture so estimated can be shown to persist through time, then this wouldconstitute additional support for our contention. Thus our procedurewas to construct a model of this triad of decisions for the year 1956and, after some experimentation, to decide on the variables that cansignificantly serve as explanatory variables. Then the model was esti-mated successively for the years 1951—60.

We first give (Tables 7, 8, and 9) what the statistical results wouldbe if single-equation techniques were employed. Notice that the rela-tionships thus estimated are significant if judged by the conventional

although the correlation coefficients indicate that economically

TAB

LE 7

Sing

le-E

quat

ion

Estim

ates

, Div

iden

d Eq

uatio

n, 1

951

-60

Var

iabl

e19

5119

5219

5319

5419

5519

5619

5719

5819

5919

60

C0

0.15

70.

0042

.003

0.0

013

—0.

0023

.006

7.0

023

.001

8.0

059

.0081

(3.2

54)

(.892

)(.6

18)

(0.2

23)

(—0.

418)

(1.2

66)

(.463

)(.3

52)

(1.0

03)

(1.5

76)

C2

—0.

0027

.996

9.9

988

.010

0.0

109

.002

4.0

050

.004

3.0

021

—0.

0014

(—0.

447)

(1.1

60)

(1.4

77)

(1.408)

(1.5

90)

(.366)

(.803)

(.668)

(.305)

(—0.

244)

C3

.0001

.0121

.009

6.0125

.0145

.0058

.008

1.0

039

.005

6.0

013

(.015)

(1.8

55)

(1.4

73)

(1.633)

(2.022)

(.833

)(1.232)

(.581)

(.759)

(.210)

C4

.0156

.0306

.0325

.0355

.0320

.0252

.0250

.0214

.0259

.0192

(2.594)

(5.3

97)

(5.679)

(5.486)

(5.030)

(3.865)

(4.056)

(3.3

53)

(3.6

88)

(3.127)

C5

.002

8.0

109

.013

3.0

178

.014

6.0

050

.008

1.0

141

.012

6.0

131

(.447

)(1

.813

)(2

.236

)(2

.545

)(2

.167

)(.7

38)

(1.2

79)

(2.1

53)

(1.7

42)

(2.1

24)

C6

—0.

0028

.004

8.0

075

.012

3.0

131

.007

2.0

061

.006

1.0

036

.000

4(—

0.39

4)(.7

23)

(1.1

34)

(1.5

78)

(1.7

41)

(.962

)(.8

57)

(.833

)(.4

54)

(.067

)C

7.0

128

.025

6.0

302

.033

5.0

261

.022

0.0

250

.026

8.0

277

.0222

(1.915)

(4.0

89)

(4.8

44)

(4.6

49)

(3.5

74)

(2.8

93)

(3.4

84)

(3.807)

(3.638)

(3.350)

C8

.010

1.0

195

.021

3.0

252

.023

6.0

180

.021

8.0

195

.016

3.0

108

(1.6

72)

(3.4

02)

(3.6

78)

(3.7

49)

(3.5

68)

(2.7

47)

(3.472)

(3.059)

(2.329)

(1.795)

C9

.0254

.039

0.0

367

.042

0.0

379

.034

2.0

343

.032

7.0338

.026

8(3

.680

)(5

.690

)(5

.588

)(5

.655

)(5

.243

)(4

.662

)(4

.821

)(4

.363

)(4

.197

)(3

.831

)P/

K.0

397

0.65

0.0

473

.050

0.0

824

.056

7.0

778

.091

2.0

716

.089

4(3

.420

)(4

.054

)(3

.431

)(3

.993

)(6

.325

)(5

.669

)(6

.880

)(6

.313

)(5

.335

)(6

.899

)N

/K—

0.00

23—

0.00

43—

0.00

22—

0.00

17—

0.00

4—

0.00

28—

0.00

48—

0.00

51—

0.00

41—

0.00

44(—

1.21

7)(—

2.06

6)(—

1.18

3)(—

0.88

6)(—

2.34

7)(—

1.38

4)(—

2.18

8)(—

2.11

0)(—

1.63

9)(—

1.77

1)f/S

.073

7.0

571

.003

0.0

265

.074

9.0

634

.087

1.1

567

.063

4.1

027

(2.6

48)

(2.2

12)

(.115

)(0

.973

)(2

.259

)(2

.064

)(3

.088

)(4

.231

)(1

.919

)(2

.672

)EF

I/S—

0.06

88(—

2.67

3)—

0.05

89(—

2.61

1).0

863

(2.6

11)

—0.

0014

(—0.

036)

—0.

0127

(—0.

311)

—0.

0171

(—0.

390)

—0.

0646

(1.5

32)

—0.

0749

(—2.

099)

—0.

0330

(—0.

552)

.023

5(.9

14)

R2

.274

7.3

328

.297

2.2

710

.353

6.3

720

.437

6.4

336

.308

3.4

166

F-st

atis

tic6.

6823

8.48

087.

3440

6.57

679.

2046

9.83

7312

.670

512

.481

97.

6490

11.7

029

Not

e: F

•05

l28

t.10

1.65

Num

ber i

n pa

rent

hese

s is t

-rat

io.

1.45

1.97

2.58

TAB

LE 8

Sing

le-E

quat

ion

Estim

ates

, Inv

estm

ent E

quat

ion,

195

1 -6

0

Var

iabl

e19

5119

5219

5319

5419

5519

561957

1958

1959

1960

C0

.019

4.0

286

.025

4.0

655

.025

5.0

407

.019

7.0

320

.037

0.0

416

(1.307)

(1.561)

(1.286)

(3.8

49)

(2.1

23)

(3.0

04)

(1.3

61)

(3.0

88)

(2.7

16)

(4.3

59)

C2

.0038

-0.0085

—0.0019

—0.0292

—0.0000

—0.

0079

.010

2—0.0067

—0.0031

—0.0127

(.219)

(—0.418)

(—0.087)

(—1.430)

(—0.005)

(—0.458)

(.607)

(—0.521)

(—0.185)

(—1.060)

C3

—0.

0023

—0.

0009

.001

7—

0.01

93.0

049

—0.

0018

.017

9—

0.00

14.0

057

.000

3(—

0.12

8)(—

0.04

3)(.0

76)

(—0.

875)

(.314)

(—0.

100)

(1.0

00)

(—0.

109)

(.328

)(.0

28)

C4

.0390

.0313

.0603

.0334

.027

5.0

338

.049

0.0

310

.044

9.0

318

(2.431)

(1.618)

(2.945)

(1.661)

(1.866)

(1.906)

(2.921)

(2.370)

(2.665)

(2.613)

C5

—0.0099

—0.0176

.0115

—0.0094

.0213

—0.0011

.0035

—0.0071

—0.0023

—0.0013

(—0.591)

(—0.932)

(.568

)(—

0.44

9)(1

.413

)(—

0.07

0)(—

0.54

5)(—

0.13

7)(—0.110).

C6

—0.0021

—0.

0019

• 0079

—0.0188

—0.0016

—0.0145

.O1S1

.0029

—0.0037

.0100

(—0.110)

(—0.

094)

(.353

)(—

0.84

6)(—

0.10

0)(—

0.76

4)(1

.013

)(.2

02)

(—0.

198)

(.739

)C

7.0525

.0388

.0424

.0231

.0556

.066

6.0

780

.028

1.0304

.0311

(2.9

61)

(1.9

23)

(1.978).

(1.054)

(3.475)

(3.429)

(4.186)

(1.921)

(1.619)

(2.312)

C8

.002

1.0

191

.027

5—

0.00

82.0

118

—0.

0017

.039

0.0048

.012

5.0

131

(.132

)(1

.001

)(1

.314

)(—

0.40

7)(.7

88)

(—0.

092)

(2.3

26)

(.368

)(.7

48)

(1.0

75)

C9

.035

4.0

296

.074

3.0

032

.005

4.0

195

.0486

.0388

.0358

.0319

(1.8

54)

(1.2

95)

(3.4

19)

(0.1

41)

(0.3

13)

(.956

)(2

.490

)(2

.502

)(1

.808

)(2

.241

)(P

/K)..

.1—

0.02

96.0

148

.015

5—

0.05

92—

0.04

29—

0.03

36—

.0.0

179

—0.

0234

—0.

0168

—0.

0347

(—1.

097)

(.488

)(.3

09)

(—1.

179)

(—1.

482)

(—0.

918)

(—0.

680)

(—0.

867)

(—0.

377)

(—1.

210)

S*2

.032

9.0

016

.005

8.0

264

.079

5.0

404

.021

6.0

101

.013

5.0

305

(2.1

57)

(.145

)(.5

18)

(1.7

78)

(5.0

49)

(2.5

27)

(1.1

48)

(.926

)(.7

48)

(2.1

99)

N/K

—0.

0051

—0.

0123

—0.

0104

—0.

0079

—0.

0058

—0.

0084

—0.

0135

—0.

0114

—0.

0137

—0.

0119

(—1.

045)

(—2.

407)

(—1.

707)

(—1.

276)

(—1.

303)

(—1.

561)

(—2.

452)

(—2.

508)

(—2.

145)

(—2.

311)

D/S

.5755

.459

8.0

157

.155

4.3

460

.416

4.6

386

.659

9.3

760

.419

7(2

.879

)(2

.075

)(.0

67)

(0.7

03)

(2.1

96)

(2.2

26)

(3.1

35).

(4.1

28)

(1.8

33)

(2.8

31)

SF1/

S.3

302

.554

0.4

695

.482

4—

0.02

51.6

579

1.09

08.3

197

.216

4.0

477

(5.0

48)

(10.503)

(5.194)

(4.504)

(—0.

283)

(6.9

42)

(14.

556)

(4.5

82)

(1.5

69)

(.947

)R

2.3

887

.561

6.3

179

.276

2.3

738

.501

2.7089

.463

9.2

470

.384

4F-

stat

istic

9.80

5418

.733

17.

4546

6.28

489.

2636

14.8

345

34.7

193

12.9

828

5.51

599.

6451

Not

e: N

umbe

r in

pare

nthe

ses i

s t-r

atio

.

TAB

LE 9

Sing

le-E

quat

ion

Estim

ates

, Ext

erna

l Fin

ance

Equ

atio

n, 1

951

-60

Var

iabl

e19

5119

5219

5319

5419

5519

5619

5719

5819

5919

60

C0

.018

6.0

084

.033

6.0

223

.020

6.0

132

.007

0.0

275

.011

4.0

382

(1.5

09)

(.702).

(1.5

09)

(1.1

77)

(2.3

67)

(.914

)(.7

79)

(2.6

21)

(1.5

64)

(2.5

44)

LTD

/(K-L

TD) —

0.01

18—

0.00

08.0

360

.016

5.0

012

.010

5—

0.00

08—

0.00

00—

0.00

22.0

007

(—2.829)

(—4.

551)

(1.7

04)

(0.9

24)

(.602

)(.8

10)

(—0.

687)

(—0.

056)

(—2.

595)

(.544

)H

/LTD

—0.

1229

—0.

0010

—0.

0099

—0.

1512

—0.

1966

—0.

1577

—0.

0639

—0.

1375

—0.

0609

—0.

9807

(—0.

551)

(—0.

006)

(—0.

091)

(—1.

845)

(—1.

435)

(—1.

305)

(—0.

721)

(—0.

863)

(—7.

446)

DEP

/K.2

151

—0.

2218

—0.

0706

—0.

1205

—0.

1241

.015

2--

0.26

90—

0.28

71—

0.07

97—

0.01

12(1

.946

)(—

2.10

5)(—

0.92

4)(—

1.34

0)(—

1.31

0)(.1

51)

(—2.

737)

(—2.

360)

(—0.

977)

(—0.

069)

P/K

—0.

0528

.043

8—

0.00

91.0

138

—0.

0035

—0.

0098

.040

3.0

229

—0.

0045

—0.

0049

(—2.

121)

(1.2

39)

(—0.

384)

(0.7

15)

(—0.

182)

(—0.

665)

(2.2

76)

(.937

)(—

0.32

1)(—

0165

)D

/S—

0.52

43—

0.64

41.1

382

—0.

1986

—0.

1258

—0.

1672

—0.

5218

—0.

5361

—0.

1126

—0.

2749

(—2.

532)

(—2.

897)

(.859

)(—

1.55

2)(—

0.94

5)(—

1.28

6)(—

4.12

8)(—

3.51

0)(—

1.26

5)(—

1.41

6)I/S

.392

5.7

300

.219

3.2

246

.016

4.3

048

.488

6.3

973

.067

2.0

769

(5.6

90)

(11.

722)

(3.9

94)

(4.8

02)

(.277

)(6

.997

)(1

5.17

5)(5

.487

)(1

.736

)(.8

29)

R2

.223

9.5

138

.109

5.1

222

—0.0059

.2392

.5862

.1435

.0340

.2198

F-st

atLs

tlc9.

6527

32.7078

4.6875

5.1756

.8249

10.3788

43.5010

6.0269

2.0490

9.4526

Not

e: N

umbe

r in

pare

nthe

ses i

s t-r

atio

.


the explanatory power of such equations is not very great. The latter,however, is quite common in cross-section investment studies.

Consider the dividend equation first. In this connection, it should benoted that C0 indicates the constant term, while the i = 2, . . 9,indicate the usual analysis of covariance contrasts; thus they may beinterpreted as measures of the difference in the constant term amongthe various industrial classifications. In addition, it should be remarkedthat at the end of Table 7 we give the critical values of the F and tstatistics at various levels of significance; the degrees of freedom of theF-statistic vary from equation to equation ranging from 165 to 180;the critical value for the t-statistic pertains to the conventional two-tailed test and the relevant degrees of freedom are approximately 165,though again this varies moderately from equation to equation. We shallnot repeat this information for subsequent tables.

Returning now to the substantive results obtained, it is interesting tonote that there is apparently quite a consistent difference in the dividendpolicies of firms in various industrial classifications as evidenced by thefact that the contrasts for i = 9, 8, 7, 4, turn out to be uniformlysignificant, while for i 6, 5, 3, and 2 they are nearly uniformly insig-nificant. Thus the result seems to imply that, ceteris paribus, firms inmining, steel, etc. (9), textiles, glass, etc. (8), building materials andequipment (7), rubber, petroleum, etc. (4), tend .to pay higher divi-dends (per dollar sales) than is the case for the remaining firms in oursample. The rate of profit variable, as was to be expected, turns out tobe the most consistent factor in explaining dividends over the sampleperiod. Another result of interest is that the investment variable turnsout to be significant quite frequently (in eight cross sections) andappears to exert a positive influence on dividends. Now one might arguethat investment, with a lag, affects the profitability of the firm's opera-tions and hence would be expected to lead to greater dividend pay-ments. On the other hand, it is difficult to rationalize this result in thecontext of our present investigation. The sign of the external financecoefficient, whenever significant, turns out to be negative. In both thecase of investment and external finance, their coefficients appear to bein conformity with the signs of their respective simple correlationswith the dividend variable.

Turning now to the investment equation, four features are worthnoting. The rate of change in sales variables turns out to be significantin only five cross sections (1951, 1954, 1955, 1956, and 1960; withthe exception of 1954 all are upswing or peak years); the rate of profitvariable turns out to be uniformly insignificant. Of the contrasts, only


for i = 7 and 4 turn out to be more or less consistently significant;notice, however, that the general constant, C0, turns out to be morefrequently significant in comparison with the general constant of thedividend equation. The external finance variable is nearly uniformlysignificant and its impact on investment is uniformly positive.

The astonishing result in the investment equation is the sign andnearly consistent significance of the coefficient of the dividend variable.While in discussing the dividend equation one could provide somerationalization for the positive sign of the coefficient of the investmentvariable, it is difficult to rationalize the positive sign of the dividendcoefficient which points to the potential pitfalls of the single-equationapproach and the ordinary tests of significance.

The external finance equation displays two interesting characteristics;there is no evidence that the firm's behavior in this aspect is affected bythe type of activity it is engaged in, since the contrasts turned out to beuniformly insignificant and thus were omitted from the reported results.There is little evidence for the validity of the principle of increasingrisk, although whenever the leverage variable is significant its coefficientturns out to be negative in conformity to the principle's implications.Finally it would appear that the most consistent determinant of externalfinance is investment, as one might expect.

The single-equation results in this section are quite instructive in anegative way. First, they demonstrate how dangerous it is to use single-equation techniques indiscriminately, when quite clearly the problemsunder consideration are fundamentally interdependent. Secondly, theyshow that the criterion of inclusion or exclusion of a variable from aregression based on the magnitude of the resulting multiple correlationcoefficient as advocated by Hotelling [5] or Theil [1412 is a very delicatecriterion and should be used very cautiously indeed. In this context, ifone is determined to use single-equation techniques, it is best to use asexplanatory variables those which are truly exogenous, even thoughsome explanatory power is lost thereby.

Estimation of the Dividend-investment-External FinanceStructure by Simultaneous-Equation Techniques

THE NEED FOR FULL INFORMATION METHODS

We have seen in the previous section that single-equation methodsare beset by serious deficiencies in meaningfully tracing the interde-

2 Pp. 206 if.


TABLE 10

Correlation Matrix of Two-Stage Least-Squares Residuals, 1951 -60

Eq.(1) Eq.(2) Eq.(3) Eq.(1) Eq.(2) Eq.(3)

1951 1956

Eq. (1)Eq. (2)Eq. (3)

1.0000 .67561.0000

.1366—0.1478

1.0000

1.0000 .75171.0000

—0.2722—0.0492

1.0000

1952 1957

Eq. (1)Eq. (2)Eq. (3)

1.0000 .84701.0000

—0.3927—0.7309

1.0000

1.0000 .90991.0000

—0.0384.2152

1.0000

1953 1958

Eq. (1)Eq. (2)Eq. (3)

1.0000 .89641.0000

—0.5025—0.1663

1.0000

1.0000 .86031.0000

.1955

.11911.0000

1954 1959

Eq. (1)Eq. (2)Eq. (3)

1.0000 .90301.0000

—0.0715—0.4463

1.0000

1.0000 .76081.0000

—0.5608—0.1303

1.0000

1955 1960

Eq. (1)Eq. (2)Eq. (3)

1.0000 .69961.0000

—0.2867—0.0441

1.0000

1.0000 .86821.0000

.0169—0.2147

1.0000

pendence of the decision making processes we wish to investigate. Thus,simultaneous-equation techniques are clearly indicated. The questionthen naturally arises as to whether limited information methods, suchas two-stage least squares, are appropriate or whether full informationtechniques are indicated.

It is, of course, the case that full information methods are moresensitive to errors of specification than limited information methods are.Nonetheless, one may suspect that the information content of the non-estimated equations is sufficiently high so that the gain in efficiency wouldoutweigh the effects of propagating specification errors. This is so sincewe are dealing with a cross-sectional sample and the three decisions


studied here are made more or less concurrently by the same group ofpeople possessing substantially overlapping or similar information.

But, of course, the ultimate test as to whether full informationmethods are indicated is related to the correlation matrix of the, say,two-stage least-squares residuals of the three equations. In Table 10,we present the relevant correlation matrixes for the years 1951—60.

In Table 10, the dividend equation is termed equation (1), theinvestment equation is equation (2), and the external finance equationis equation (3). The model on which these correlations are based isthat presented in Tables 7, 8, and. 9. The impression is unmistakable,from Table 10, that full information methods are indicated in thepresent case.

The proper significance tests for this type of correlation coefficientare not adequately dealt with in the econometric literature. However,in view of the consistency of the two-stage least-squares estimates, it isnot amiss to treat the residuals, at least asymptotically, as observationson the (unobservable) disturbances. This, of course, assumes no speci-fication errors. If we take this view, then the proper test would bebased on the distribution of the usual sample correlation coefficient.The appropriate number of degrees of freedom would be the numberof observations less the number of estimated parameters in the twoequations of interest. In our case, the degrees of freedom are not lessthan 150. On the basis of these considerations, it would appear that thecritical values for rejecting the hypothesis of no correlation areIn = .159 and .208, employing a two-tail test at the 5 and 1 percent levels of significance, respectively.

Table 10 shows that the hypothesis is to be accepted only in thefollowing instances; between dividend and external finance residuals for1951, 1954, 1957, and 1960. In addition, the hypothesis is to beaccepted also for the years 1955, 1959, and 1960 between the invest-ment and external finance residuals. Interestingly enough, these areyears of upswings or peaks. Thus, with the possible exception of upswingyears, where apparently profits outrun investment and dividend require-ments so that there is no effective (resource) constraint on the firms'operation, it would appear that the correlation coefficients are suffi-ciently high to warrant the conclusion that the efficiency of the param-eter estimators would be appreciably increased by the use of full infor-mation methods. It might be instructive, however, before we presentthe main empirical results of our study, to examine briefly the limitedinformation (two-stage least-squares) estimates of our model. These arepresented in Tables 11, 12, and 13.

TAB

LE 1

1

Lim

ited

Info

rmat

ion

(2SL

S) E

stim

ates

, Div

iden

d Eq

uatio

n, 1

951

-60

Var

iabl

e19

5119

5219

5319

5419

5519

5619

5719

5819

5919

60

C0

.027

7.0

172

.015

7.0

132

.010

0.0

196

.022

2.0

093

.009

4.0

259

(2.7

999)

(1.3

136)

(1.0

747)

(1.5

090)

(1.2

304)

(1.8

840)

(1.5

188)

(1.1

085)

(.866

8)(2

.133

3)C

2—

0.00

47.0

044

.002

8.0

042

.004

7.0

056

.009

5.0

035

.003

0—

0.00

46(—

0.63

01)

(.483

2)(.2

584)

(.501

7)(.5

252)

(.484

4)(.7

920)

(.513

8)(.3

892)

(—0.

5898

)C

3—

0.00

24.0

135

.002

8.0

094

.008

5.0

105

.017

3.0

045

.006

6.0

044

(—0.

3083

)(1

.375

5)(.2

380)

(.986

4)(.9

249)

(.779

2)(1

.327

1)(.6

377)

(.816

0)(.5

461)

C4

.028

9.0

478

.047

0.0

417

.042

9.0

474

.064

1.0

317

.032

2.0

378

(2.4

175)

(2.5

982)

(2.1

911)

(4.7

282)

(4.9

359)

(3.0

207)

(2.8

057)

(2.9

074)

(2.3

465)

(2.8

913)

C5

.001

3.0

053

.016

0.0

151

.014

9.0

082

.013

1.0

143

.014

9.0

155

(.177

8)(.5

500)

(.971

3)(1

.885

7)(1

.800

3)(.8

785)

(1.1

106)

(2.0

966)

(1.8

820)

(1.9

992)

C6

—0.

0018

.005

0.0

040

.010

4.0

086

.005

2.0

179

.007

8.0

021

.006

9(—

0.20

'70)

(.492

0)(.3

480)

(1.0

810)

(.912

1)(.5

018)

(1.2

377)

(1.0

087)

(.240

9)(.7

486)

C7

.030

7.0

456

.044

5.0

384

.043

4.0

543

.086

6.0

370

.030

7.0

419

(2.1

275)

(2.3

471)

(2.6

646)

(4.0

349)

(3.9

680)

(2.7

453)

(2.3

678)

(3.2

034)

(2.5

173)

(3.0

755)

C8

.0.1

50.0

306

.025

9.0

234

.024

5.0

248

.052

2.0

233

.017

0.0

212

(1.8

553)

(2.3

904)

(2.1

047)

(3.0

610)

(2.6

935)

(2.0

607)

(2.7

126)

(3.0

333)

(1.9

699)

(2.2

348)

C9

.042

8.0

549

.065

0.0

446

.044

9.0

524

.080

2.0

451

.035

2.0

485

(3.0

808)

(2.4

140)

(2.4

184)

(4.7

073)

(4.8

429)

(3.6

863)

(2.7

637)

(3.2

7.43

)(2

.554

1)(3

.458

3)P/

K.0

343

.078

7.0

782

.040

7.0

769

.066

6.0

881

.098

8.0

778

.093

6(2

.217

1)(3

.021

1)(2

.628

1)(2

.756

8)(4

.834

)(3

.186

7)(3

.925

6)(5

.961

6)(5

.176

5)(5

.757

0)N

/K—

0.00

38—

0.00

96—

0.00

85—

0.00

29—

0.00

64—

0.00

78—

0.01

42—

0.00

83—

0.00

66—

0.01

11(—

1.38

56)

(—1.

9510

)(—

1.72

14)

(—1.

2093

)(—

2.40

54)

(—1.

8377

)(—

2.28

96)

(—1.

6282

)(—

2.31

44)

I/S—

0.18

12—

0.33

93—

0.40

08—

0.14

81—

0.17

73—

0.30

71—

0.47

98—

0.04

17—

0.03

07—

0.29

31(—

0.96

24)

(—1.

0504

)(—

1.12

98)

(—1.

5108

)(—

1.74

33)

(—1.

3858

)(—

1.63

44)

(—0.

2445

)(—

0.14

76)

(—1.

2723

)EF

1/S

—0.

0656

.181

5.6

296

.181

4.2

483

.350

5.4

206

—0.

0020

.303

9.1

151

(—0.

6753

)(1

.078

1)(2

.153

9)(.8

808)

(1.2

511)

(.624

8)(1

.083

6)(—

0.02

13)

(1.0

728)

(1.7

823)

Not

e: N

umbe

r in

pare

nthe

ses i

s t-r

atio

.

TAB

LE 1

2

Lim

ited

In fo

rmat

ion

(2SL

S) E

stim

ates

, Inv

estm

ent E

quat

ion,

195

1 -6

0

Var

iabl

e19

5119

5219

5319

5419

5519

5619

5719

5819

5919

60

C0

.036

2.0

427

.046

2.0

636

.029

3.0

494

.057

4.0

452

.061

7.0

466

(1.7

394)

(2.0

186)

(1.7

199)

(2.0

680)

(.030

9)(2

.147

8)(2

.308

6)(3

.172

3)(2

.613

6)(4

.420

4)C

2.0

025

—0.

0073

—0.

0097

—0.

0019

.012

4—

0.00

33.0

061

.001

3.0

079

—0.

0078

(.142

2)(—

0.35

03)

(—0.

4062

)(—

0.05

84)

(.635

2)(—

0.13

96)

(.272

6)(.0

797)

(.327

5)(—

0.59

40)

C3

—0.

0018

.002

1—

0.00

28.0

329

.020

3.0

061

.019

7.0

163

.017

0.0

089

(—0.

0991

)(.0

978)

(—0.

1162

)(.8

598)

(.980

4)(.2

411)

(.833

5)(.8

705)

(.683

9)(.6

167)

C4

.057

8.0

627

.080

3.1

457

.072

9.0

777

.105

0.0

858

.119

6.0

596

(2.4

587)

(2.3

626)

(2.7

334)

(2.6

593)

(3.2

030)

(2.3

379)

(3.1

317)

(3.1

028)

(2.3

611)

(2.7

069)

C5

—0.

0055

—0.

0084

.007

5.0

311

.041

5.0

134

.014

4.0

095

.032

2.0

114

(—0.

3063

)(—

0.41

15)

(.336

4)(.8

718)

(2.1

673)

(.630

2)(.6

211)

(.537

0)(.9

827)

(.740

8)C

6—

0.00

21.0

003

—0.

0052

.027

9.0

139

.001

8.0

267

.018

9.0

187

.015

6(—

0.10

48)

(.017

6)(—

0,20

11)

(.733

6)(.6

891)

(076

7)(1

.039

4)(.9

613)

(.647

0)(1

.056

6)C

7.0

745

.063

6.0

650

.136

7.0

966

.117

3.1

585

.094

4.1

1.12

.061

5(2

.910

3)(2

.564

9)(2

.187

7)(2

.495

0)(4

.228

4)(2

.825

2)(3

.538

1)(3

.021

4)(2

.082

7)(2

.523

8)C

8.0

209

.035

1.0

357

.067

1.0

475

.041

6.0

788

.048

2.0

598

.033

4(.9

268)

(1.6

274)

(1.4

860)

(1.5

482)

(2.0

668)

(1.0

849)

(2.7

615)

(2.0

881)

(1.6

871)

(1.8

133)

C9

.071

70.

739

.108

2.1

491

.060

7.0

748

.134

8.1

13.1

078

.066

5(2

.027

8)(2

.058

6)(2

.974

8)(2

.274

2)(2

.223

6)(1

.825

1)(2

.932

6)(3

.349

4)(2

.229

6)(2

.378

1)(P

/K)..

1—

0.01

19.0

473

.124

2.2

040

.036

6.0

498

.097

3.1

439

.293

8.0

458

(—0.

3680

)(1

.261

4)(1

.210

7)(1

.612

5)(.8

487)

(.767

4)(1

.527

7)(1

.901

2)(1

.488

0)(.7

561)

S*2

.033

7.0

000

—0.

0054

—0.

0153

.072

5.0

409

—0.

0135

—0.

0103

—0.

0753

.010

0(2

.025

2)(.0

071)

(—0.

3603

)(—

0.57

59)

(3.7

398)

(2.0

248)

(—0.

4020

)(—

0.44

77)

(—1.

2534

)(.5

051)

N/K

—0.

0064

—0.

0142

—0.

0177

—0.

0216

—0.

0102

—0.

0141

—0.

0243

—0.

0244

—0.

0398

—0.

0191

(—1.

2329

)(—

2.63

78)

(—2.

0275

)(—

1.98

65)

(—1.

8737

)(—

2.00

68)

(2.8

265)

(—3.

0674

)(—

2.13

92)

(—2.

6342

)D

/S—

0.25

53—

0.55

61—

1.25

83—

2.91

85—

0.88

33—

0.81

74—

1.27

44—

1.09

97—

2.00

07—

0.43

23(—

0.34

05)

(—0.

8720

)(—

1.16

88)

(—1.

9967

)(—

1.92

63)

(—1.

0164

)(—

1.43

65)

(—1.

5189

)(—

1.40

55)

(—0.

7581

)EF

1/S

.085

1.4

695

.802

91.

4809

—0.

1375

.203

9.5

695

.165

01.

0848

.129

5(.3

719)

(3.4

898)

(2.1

078)

(2.7

006)

(—0.

3116

).(.2

317)

(.921

3)(.5

679)

(.928

9)(1

.051

6)

Not

e: N

umbe

r in

pare

nthe

ses i

s t-r

atio

.

TAB

LE 1

3

Ii

Lim

ited

Info

rmat

ion

(2SL

S) E

stim

ates

, Ext

erna

l F'in

ance

equ

atio

n, 1

951

-60

I

Var

iabl

e19

5119

5219

5319

5419

5519

5619

5719

5819

5919

60

C0

.018

6.0

051

.043

8.0

259

.024

1.0

195

.021

3.0

324

.011

8.0

728

(1.0

536)

(.268

8)(1

.777

7)(1

.236

1)(2

.289

9)(1

.293

6)(1

.953

6)(2

.694

0)(1

.474

1)(3

.326

1)LT

D/(K

—LT

D) —

0.01

18—

0.00

05.0

337

.011

1.0

011

.009

1—

0.00

08—

.0.0

000

—0.

0021

—0.

0004

(—2.

8772

)(—

1.99

15)

(1.5

071)

(.599

6)(.5

416)

(.675

5)(—

0.59

46)

(—0.

0806

)(—

2.51

68)

(—0.

2556

)H

/LTD

—0.

1489

—0.

1611

.00.

16—

0.17

01—

0.26

41—

0.18

71—

0.10

81—

0.18

57—

0.06

13—

1.17

84(—

0.56

81)

(—9.

6545

)(.0

128)

(—2.

0045

)(—

1.74

69)

(—1.

3453

)(—

1.10

51)

(—1.

5837

)(—

0.87

54)

(—6.

8722

)D

FP/K

.203

0—

0.35

40—

0.03

27—

0.20

35—

0.17

72.0

029

—0.

3236

—0.

3740

—0.

0636

—0.

3391

(1.7

329)

(—2.

2743

)(—

0.31

77)

(—2.

0577

)(—

1.67

25)

(.023

8)(—

2.90

77)

(—2.

6794

)(—

0.73

21)

(—1.

4703

)P/

K—

0.04

36.1

607

—0.

0434

.042

4.0

161

—0.

0096

.044

5.0

559

—0.

0109

.110

6(—

1.30

47)

(2.0

572)

(—1.

0145

)(1

.764

3)(.6

112)

(—0.

4048

)(1

.557

1)(1

.557

8)(—

0.59

12)

(1.7

708)

D/S

—0.

6930

—1.

9653

.275

0—

0.67

88—

0.38

43—

0.26

18—

0.74

63—

0.93

94—

0.04

37—

1.81

34(—

1.10

98)

(—1.

9920

)(.4

919)

(—2.

8479

)(—

1.32

99)

(—0.

7325

)(—

2.30

16)

(—2.

7185

)(—

0.25

20)

(—2.

6671

)I/S

.471

01.

3034

—0.

0497

.345

5.1

287

.259

3.4

516

.598

0.0

153

.606

5(2

.430

8)(4

.122

0)(—

0.19

72)

(3.8

562)

(1.0

669)

(2.2

752)

(4.8

456)

(3.1

997)

(.148

6)(1

.946

9)

'I I I

Not

e: N

umbe

r in

pare

nthe

ses i

s t-r

atio

.

—I


At least two aspects of these results are worthy of comment. First,notice the reversal of sign of the coefficients of the investment variablein the dividend equation and that of the dividend variable in the invest-ment equation. This does indeed demonstrate the type of erroneousconclusions one might be led to by using inappropriate econometrictechniques.

The second interesting feature is that the single-equation resultsseriously understate the role of the rate of profit as a determinant ofinvestment. The two-stage least-squares results show that variable to beoccasionally significant in the investment equation, while the single-equation results never show it to be significant (often with an unac-ceptable sign).

No further comments seem to be warranted at this juncture, sincewe shall comment at length below on the structure of this triad ofdecision-making processes, when its parameters have been estimatedby the appropriate econometric procedure, which our results so farsuggest should be a full information one. The parameter estimates,however, do not differ markedly in two- and three-stage least squares,though the latter results are more efficient.

MAIN EMPIRICAL RESULTS

In dealing with the full information results of our estimation, it iswell to bear in mind that while there are cogent a priori reasons whyone might expect that such decisions as we are studying here areresolved simultaneously, nonetheless one cannot assert with any confi-dence that the feedbacks are equally strong in all directions. Hence oneof the issues to be resolved by the empirical results is whether a fullysimultaneous model, such as the one presented in the previous section,is the most appropriate or whether some other form such as a recursiveone would be a more suitable version for the model. Admittedly, thereis no clear-cut criterion by which to choose. We have tended to makethis determination in terms of the interpretability of the results yieldedin the context of the appropriate theoretical considerations. On thisbasis, it would appear that the fully simultaneous model gives moremeaningful results and we shall, therefore, proceed to analyze itsempirical implementation in some detail. The empirical results arecontained in Tables 14, 15, and 16.

Consider the dividend equation first. One noteworthy aspect is therather remarkable stability of both the significance pattern and magni-tudes of the contrasts. While it might be difficult to argue definitivelyagainst the observation that this may be due entirely to the normaliza-

TAB

LE 1

4

In fo

rmat

ion

(3SL

S) E

stim

ates

, Div

iden

d Eq

uatio

n, 1

951

-60

Var

iabl

e19

5119

5219

5319

5419

5519

5619

5719

5819

5919

60

C0

.028

9.0

268

.005

3.0

138

.008

7.0

186

.030

8.0

147

.009

1.0

328

(3.0

240)

(2.6

695)

(.472

3)(1

.661

9)(1

.117

9)(1

.880

4)(2

.428

9)(1

.885

8)(.9

025)

(3.3

014)

C2

—0.

0049

.004

1.0

035

.004

1.0

068

.002

6.0

075

.002

9.0

023

—0.

0059

(—0.

6628

)(.4

939)

(.373

8)(.4

937)

(.777

8)(.2

457)

(.636

9)(.4

396)

(.326

6)(—

0.77

00)

C3

—0.

0022

.012

9.0

034

.009

2.0

106

.006

9.0

156

.005

5.0

062

.005

9(—

0.28

40)

(1.4

304)

(.329

9)(.9

858)

(1.1

872)

(.556

4)(1

.238

3)(.7

859)

(.820

1)(.7

472)

C4

.030

1.0

615

.030

9.0

424

.045

6.0

435

.072

2.0

389

.030

1.0

470

(2.5

859)

(4.1

939)

(2. 1

453)

.008

4'(5

.501

0)(5

.438

9)(3

. 195

9)(3

.854

2)(3

.872

2)(2

.321

7)(4

.290

5)C

5.0

014

.011

0.0

154

.017

1.0

073

.014

8.0

134

.012

2.0

164

(.191

4)(1

.235

8)(.8

782)

(1.9

273)

(2.1

397)

(.822

6)(1

.270

4)(1

.994

9)(1

.640

2)(2

.134

4)C

6—

0.00

24.0

065

.003

7.0

102

.010

4.0

061

.018

2.0

089

.004

4.0

101

(—0.

2808

)(.6

968)

(.362

7)(1

.083

2)(1

.138

4)(.6

222)

(1.1

710)

(.536

4)(1

.136

1)C

7.0

315

.055

3.0

294

.038

8.0

4413

.049

8.1

039

.044

2.0

294

.050

3(2

.246

4)(3

.555

9)(2

.421

6)(4

.533

1)(4

.300

7)(2

.857

0)(3

.500

5)(4

. 131

9)(2

.543

8)(4

.352

2)C

8.0

137

.036

4.0

173

.023

7.0

274

.026

7.0

566

.025

8.0

168

.026

4(1

.709

4)(3

.360

4)(1

.745

7)(3

.125

2)(3

.113

0)(2

.292

7)(3

.443

1)(3

.468

3)(2

.060

8)(3

.056

4)C

9.0

423

.076

7.0

411

.045

3.0

479

.049

9.0

945

.054

2.0

361

.057

5(3

.102

5)(4

.a83

)(2

.349

5)5.

120

(5.2

868)

(3.8

895)

(3.9

039)

(4.2

770)

(2.7

52.1

)(4

.753

7)P/

K.0

332

.049

1.0

721

.040

9.0

634

.066

0.1

012

.070

2.0

751

(2.4

752)

(2.3

130)

(2.7

414)

(2.8

356)

(5.2

616)

(3.4

194)

(3.3

412)

(6.3

468)

(5.0

082)

(520

06)

N/K

—0.

0040

—0.

0092

—0.

0043

—0.

0030

—0.

0073

—0.

0065

—0.

0138

—0.

0103

—0.

0053

—0.

0114

(—1.

6576

)(—

2.19

27)

(—1.

2474

)(—

1.44

19)

(—2.

8687

)(—

1.80

40)

(—2.

5332

)(—

3.06

66)

(—1.

3948

)(—

2.83

24)

I/S—

0.19

00—

0.57

52—

0.11

06—

0.16

25—

0.19

96—

0.27

27—

0.61

00—

0.16

63—

0.01

24—

0.45

18(—

1.06

67)

(—2.

3977

)(—

0500

1)(—

2.54

32)

(—2.

1176

)(—

1.51

24)

(—2.

7209

)(—

1.10

17)

(—0.

0633

)(—

2.59

95)

EF1/

S—

0.07

70.2

646

.689

0.2

087

.360

4.3

289

.427

2—

0.02

23.3

128

.123

1(—

0.83

88)

(1.9

407)

(2.9

665)

(1.3

556)

(1.8

633)

(.671

5)(1

.199

0)(—

0.23

95)

(1.1

293)

(1.9

467)

Not

e: N

umbe

r in

pare

nthe

ses i

s t—

ratio

.

TAB

LE 1

5

Full

Info

rmat

ion

(3SL

S) E

stim

ates

, Inv

estm

ent E

quat

ion,

195

1 -6

0

Var

iabl

e19

5119

5219

5319

5419

5519

5619

5719

5819

5919

60

C0

.056

2.0

360

.035

2.0

779

.031

3.0

727

.046

3.0

483

.065

7.0

493

(2.9

868)

(2.3

605)

(1.7

951)

(2.9

383)

(2.1

768)

(3.4

825)

(2.0

629)

(3.4

706)

(2.9

005)

(4.8

220)

C2

.000

9.0

129

.007

6.0

062

.017

2—

0.00

80.0

072

.002

3.0

045

—0.

0097

(.050

7)(.9

951)

(.375

9)(.2

096)

(.888

5)(—

.0.3

568)

(.323

8)(.1

461)

(.198

0)(—

0.77

71)

C3

—.0

024

.021

9.0

133

.020

1.0

261

.003

2.0

233

.014

8.0

220

.006

0(—

.130

6)(1

.593

4)(.5

967)

(.585

5)(1

.288

8)(.1

358)

(1.0

010)

(.815

1)(.9

458)

(.444

8)C

4.0

822

.095

8.1

135

.185

7.0

918

.130

0.0

942

.098

9.1

344

.061

8(3

.971

9)(5

.143

2)(4

.798

3)(4

.670

2)(4

.675

9)(5

.058

2)(3

.210

4)(3

.904

4)(2

.814

8)(3

.505

9)C

5.0

014

.031

4.0

262

.060

4.0

481

.028

1.0

146

.020

2.0

328

.011

9(.0

849)

(2.2

174)

(1.3

235)

(1.9

267)

(2.5

908)

(1.3

977)

(.639

1)(1

.166

9)(1

.066

7)(.8567)

C6

.000

2.0

120

.009

8.0

199

.019

8.0

217

.021

2.0

190

.026

1.0

155

(.014

9)(.8

320)

(.448

3)(.5

746)

(.986

6)(.9

607)

(.832

4)(.9

917)

(.955

6)(1

.098

0)C

7.1

011

.078

1.0

940

.163

9.1

147

.186

8.1

689

.119

6.1

275

.061

0(4

.452

7)(4

.701

7)(3

.900

9)(3

.994

8)(5

.671

3)(5

.719

8)(4

.156

8)(4

.199

9)(2

.540

1)(3

.145

2)C

8.0

462

.0544

.0606

.1014

.0623

.1022

.0709

.072

1.0

721

.029

8(2

.250

4)(3

.992

3).

(2.9

593)

(2.9

762)

(2.9

613)

(3.3

186)

(2.7

200)

(3.3

576)

(2.1

713)

(1.9

612)

C9

.1186

.1314

.1413

.1918

.0844

.1422

.142

8.1

628

.152

8.0

624

(3.9668)

(4.6803)

(5.0492)

(4.1124)

(3.6545)

(4.5389)

(3.4

699.

)(5.4355)

(3.3663)

(2.8230)

(P/K).1

.0234

.0042

.1340

.1842

.0806

.1238

.009

0.1402

.2418

—.0.0108

(.4335)

(.2297)

(1.9058)

(2.1315)

(2.4284)

(2.4919)

(.209

9)(2.1097)

(1.3015)

(—0.2809)

.0315

.0071

.0074

.0143

.0627

.0447

.0357

.0403

—0.

0076

.0241

(1.9

453)

(1.2

251)

(.938

0)(1

.422

0)(3

.695

2)(2

.664

:8)

(2.2

093)

(2.0

490)

(—0.

1372

)(1

.897

8)N

/K—

0.00

70—

0.00

95—

0.01

66—

0.01

50—

0.01

33—

0.01

93—

0.01

66—

0.02

17—

0.03

27—

0.01

19(—

1.52

60)

(—2.

5.34

5)(—

2.39

55)

(—1.

7010

)(—

2.64

45)

(—3.

0976

)(—

2. 1

921)

(—2.

9701

)(—

1.85

91)

(—2.

1800

)D

/S—

1.37

58—

1.10

43—

1.96

39—

4.20

04—

1.37

51—

2.36

47—

0.86

37—

1.69

44—

2.69

40—

0.42

51(—2.3716)

(—2.3017)

(—2.8860)

(—4.

9679

)(—

3.87

10)

(—4.

1750

)(—

1.24

35)

(—2.

6593

)(—

2.00

22)

(—1.

0417

)EF1/S

—.0943

.3976

1.0322

1.0480

—0.1781

—0.8921

.0912

—0.3760

.4722

.1064

(—.4472)

(3. 1588)

(3.0578)

(2.1

296)

(—0.

4063

)(—

1. 1

444)

(.162

6)(—

1.44

26)

(.424

6)(.9

507)

Not

e: N

umbe

r in

pare

nthe

ses i

s t-r

atio

.

TAB

LE 1

6

Full

Info

rmat

ion

(3SL

S) E

stim

ates

, Ext

erna

l Fin

ance

Equ

atio

n, 1

951

-60

Var

iabl

e19

5119

5219

5319

5419

5519

5619

5719

5819

5919

60

C0

.014

8—

0.01

21.0

250

.025

1.0

258

.017

4.0

217

.033

0.0

091

.060

5(.8

566)

(—0.

7917

)(1

.080

3)(1

.740

4)(2

.471

3)(1

.211

7)(1

.991

6)(2

.748

5)(1

.163

9)(2

.815

4)LT

D/(K

—LT

D) —

0.01

15—

0.00

08.0

149

.010

6.0

007

.003

2—

0.00

04—

0.00

03—

0.00

17.0

001

(—2.8806)

(—3.0161)

(.7424)

(1.1168)

(.3778)

(.2462)

(—0.

3631

)(—

0.29

22)

(—2.

1438

)(.0

889)

R/L

TD—

0.06

60—0.0145

—0.0321

—0.1591

—0.3001

—0.2066

—0.1374

—0.2086

—0.0889

—1.1594

(—0.

2556

)'(—

0.08

57)

(—0.

3369

)(—

2. 1

438)

(—2.

0968

)(—

1.52

73)

(—1.

4366

)(—

1.88

85)

(—1.

3550

)(—

6.76

65)

DEP

/K.2

191

—0.

2285

.003

5—

0.21

69—

0.19

76—

0.05

14—

0.31

57—

0.38

29—

0.00

73—

0.21

07(1

.941

1)(—

1.90

49)

(.039

6)(—

2.24

00)

(—1.

9292

)(—

0.43

76)

(—2.

8443

)(—

2.76

16)

(—0.

0859

)(—

0.93

38)

P/K

—0.

0406

.124

6—

0.07

01.0

436

.016

7—

0.00

93.0

430

.076

6—

0.03

47.1

252

(—1.

2166

)(1

.650

9)(—

1.75

43)

(1.8

287)

(.637

0)(—

0.39

25)

(1.5

073)

(2.1

475)

(—1.

9325

)(2

.008

3)D

/S—

0.68

02—

1.05

31.6

903

—0.

6702

—0.

3956

—0.

2575

—0.

7333

—1.

2043

.325

5—

1.86

97(—

1.09

40)

(—1.

1633

)(1

.361

7)(—

2.81

98)

(—1.

3703

)(—

0.72

54)

(—3.

5257

)(1.9936)

(—2.

7528

)I/S

.478

51.

0682

—0.

2174

.350

2.1

322

.259

6.4

482

.718

4—

0.14

49.6608

(2.4

752)

(3.5

622)

(—0.

9311

)(4.0191)

(1.0

985)

(2.2

882)

(4.8

168)

(3.8

733)

(—1.

4538

)(2

.124

2)

Not

e: N

umbe

r in

pare

nthe

ses i

s t-r

atio

.


tion we have chosen to impose on our endogenous variables, nonethe-less such inference does not seem valid to us. It might be observedthat the stability in the contrasts just noted seems entirely in accordwith the single-equation conclusions based on Table 7. A rather impor-tant consequence of this finding relates to the well-known studies ofLintner [15, 16], as well as to those of others. We seem to find thatthe dividend decision function varies from one industrial classificationto another, at least with respect to the constant term. Hence, it wouldnot appear proper to deal with this relationship in simple aggregativeterms. At least this aspect of nonhomogeneity must be taken account of.

The coefficients of the exogenous variables exhibit a substantialdegree of stability over time. The coefficient of the net current position(N/K) is uniformly small and negative; it turns out to be insignificantfor 1953, 1954, and 1959. It may be remarked that these were years ofrelatively low (corporate) inventory investment, and thus the iásignifi-cance of the net current position is best understood in terms ofdominance of this variable by its inventory component. The result isthen perfectly intelligible in view of the uniformly negative impact ofinvestment on the dividend policy of firms.

The coefficient of the rate of profit variable is consistently positiveand significant. There is some slight evidence of an upward trend in itsmagnitude, and while one might tend to attribute it to the rather pro-longed expansionary behavior of the economy in the postwar years, theevidence is too slight to permit this interpretation to be pressed toovigorously.

The behavior of the investment and external finance coefficients isof considerable interest. First, comparing them with the single-equationresults, we find a complete reversal of their sign. Beyond that, the fullinformation results appear to indicate an interesting pattern of relation-ships. Thus, the external finance variable appears to be significant onlyin periods of upswings (1952, 1955) or peaks (1953, 1960). On theother. hand, the coefficient of the investment variable appears to besignificant (and negative) chiefly during the upturn and peak years,although this is not a strictly consistent phenomenon. These resultsclearly demonstrate the significant dependence of the dividend policyof firms on their investment decision, a result obtained also in connec-tion with our study of the dividend policy of electric utility firms [3]in a slightly different context. In the light of these results, one oughtto have serious reservations concerning the specification of Lintner'sdividend relation. Our findings would lead us to believe that one couldbest interpret his relation as simply the reduced form of the appropri-


ate dividend function. Thus it would appear that several useful variableshave been omitted from consideration and the consequent incidence ofspecification bias would tend to vitiate his otherwise excellent andpioneering work. Incidentally a similar interpretation of Darling's [2]work would tend to explain why changes in sales tend to serve asa useful explanatory variable for the dividend behavior of firms.

Finally, one might remark generally that the behavior of the esti-mates of the coefficients of the dividend equation over the entire sampleperiod is rather stable and relatively free of cyclical characteristics.

Turning now to the investment equation, we first note that certaincontrasts turn out to be consistently significant, viz., for i 9, 8, 7,and 4. The remaining contrasts tend, on the whole, to be insignificant.This is a result that departs from the single-equation findings. It mightalso be pointed out in passing that these are the same industrial classi-fications for which significant contrasts were also found in the dividendequation. The significance of these contrasts indicates some degree ofnonhomogeneity in the investment behavior of firms and hence severalcross-section studies, such as those of Eisner [6], commit a specificationerror when they neglect this aspect. It might also be remarked that themagnitude of these estimates remains relatively stable over the period.

The pattern of influence of the predetermined variables on the invest-ment decision is of considerable interest. First, note that the net currentposition (N/K) variable is nearly uniformly significant and negative inimpact. This is best understood in terms of the cash flow constraintreferred to above (equation 2) and the remarks made in connectionwith this variable when discussing the dividend equation.

Beyond this, the results provide some basis for appraising the"profits" and "accelerator" versions of investment theory. At firstglance, the results seem to favor slightly the accelerator version. The(two-year average) change of sales variable (S*_2) appears to exerta significant influence on investment during the long period of postwarinvestment boom (following 1955). During that period it is insignificantonly for 1959. Its coefficient is also significant in 1951, during theKorean war boom. The (lagged) rate of profit variable (our view ofthe profits version is somewhat different from the conventional one)appears to be insignificant in upswing (1951, 1952, 1959) and peakyears (1957 and 1960). Moreover, the coefficient of this variable, whensignificant, does not appear to exhibit as much stability as the coefficientof the change of sales variable.

One might be tempted to argue that the "profits" version of invest-ment theory is less pertinent than the accelerator version. This is, e.g.,


the conclusion of Eisner [6] and Kuh [13], but it is a misleading one.First, their results are based on single-equation techniques and thus donot adequately reveal the structure of the decision-making process theywish to study. (In this respect our own single-equation results tend tocorroborate theirs.) Secondly, in choosing between the two alternativeversions, it is incorrect to rely only on the direct role each of the rele-vant variables is empirically determined to play in the investment equa-tion. On the contrary, our results seem to imply that the "profits" ver-sion is a very consistent and useful way of looking at the empiricalbehavior of investment. For the rate of profit variable is a very con-sistent and very significant determinant of the dividend behavior offirms, while the dividend variable appears also as a very useful andconsistent determinant of investment. Thus, profits seem to operate oninvestment in a rather complex way, both directly and indirectly throughthe impact they exert on the dividend and external finance behavior offirms. Thus, if we interpret the cross-section studies of investment byrecent authors, such as Kuh [13] and Eisner [6], as relating to thereduced form of the investment function, then we can plausibly arguethat their generally negative findings with respect to the profits versionis simply a reflection of the fact that the reduced form confounds thevarious links through which the rate of profit affects the investmentprocess, given the structure as set forth in our investigation.

The dividend impact on investment is quite pronounced and con-sistently negative and significant (except for 1957 and 1960, both peakyears). This is again a sharp departure from the results obtained bysingle-equation methods and supports the interdependent view of thistriad of decision processes. In conjunction with the similar situationfound for the dividend equation, it demonstrates the competitive char-acter of the investhent and dividend demands on the firm's resources.

The external finance variable appears to have been significant onlyin the pre-1955 period. Its coefficient, however, is not very stable,although it is positive, as one might expect. This might indicate thatequity financing has become relatively more significant since 1955.Beyond that, one is hard put to find any other interpretation.

Finally, turning to the external finance equation, we should note thatcontrasts were determined to be insignificant and hence were omittedfrom the final form of this equation. Beyond that, the results hereprovide considerable information on a number of rather minor butstill interesting issues. One of these is the empirical relevance of theso-called principle of increasing risk [11]. This has received scantattention in the literature. A recent study that seems to have examined


this issue is that of Meyer and Glauber [18]. Unfortunately, however,the manner in which they approach the problem is unsatisfactory; forthey merely limit themselves to examining the significance of the simplecorrelation between investment and an appropriate leverage variable.But, admittedly, the connection between investment and leverage is nota simple and direct one, and conclusions based on so crude a measureas simple correlations are bound to be quite tenuous. When the problemis examined in its proper perspective, viz., as influencing the externalfinance activity of the firm and only through that its investment deci-sion, then one sees that there is some validity to the notion of increasingrisk, especially in periods (the pre-Accord years3) when market ration-ing could not adequately take place through interest rate variation. Wesee that the leverage variable does turn out to be significant in 1951and 1952 as well as in 1959, and it is negative in sign. Its numericalcoefficient, however, is quite small. In conjunction with this, our interestvariable, although quite imperfect in construction, tends to be insignifi-cant in 1951, 1952, and 1953 (a period adjacent to the Accord), whileit is consistently significant and negative in impact thereafter. Again thisresult throws some light on a related question, viz., the impact of theinterest rate on investment, to which the traditional single-equationapproach has normally answered that the impact of the interest rate oninvestment is insignificant. In this structural context, however, we dosee such an impact in the positive and significant impact of the externalfinance variable in the investment equation in certain years.

The depreciation variable behaves in the expected fashion in con-formity with the implication of the flow-of-funds view. One might askwhy depreciation and profits should not be aggregated into one variablewhen dealing with the external finance equation, again in the context ofthe flow-of-funds view. As the results indicate, this would have beenstatistically an unsound procedure since the coefficients of the depre-ciation and (rate of) profit variables are generally of different sign.Moreover, the sign of the rate of profit variable is not stable and dis-plays a distinctly cyclical character. It tends to be positive duringperiods when economic activity slows down (1954, 1958), while ittends to be negative during upswing or peak years (1953, 1959). Thisis quite interesting and seems to be symptomatic of the following typeof behavior. It is a well-known and documented fact that firms displaymarked reliance on internal funds. During the upswing, when profitsare rapidly rising, the more profitable a firm is, the less likely it may be

Treasury-Federal Reserve Accord, March 4, 1951.


to rely on external finance. On the other hand, during periods of tightmoney or during the downswing, when profits are generally squeezed, thenfirms lacking in internal funds may be more inclined to resort to the capitalmarket. At such periods, however, it is also a well-established fact thatnoninterest related forms of credit rationing are relatively more preva-lent. Hence it is the more profitable firms that tend to have access tothe capital market. If we recall that we deal with a cross-sectionalsample, the change in sign of the coefficient of the profitability variableis easily understood.

As might have been expected, the investment variable turns out tobe a rather consistent determinant of external finance, although itscoefficient does not display much stability, even when significant. Thedividend variable seems to be a less consistent determinant of theexternal finance behavior of finns. Its coefficient in 1954, 1957, 1958,and 1960 is significant and negative, while in 1959 it is significant butpositive. Although it is difficult to give a useful interpretation to thisresult, other than the fact that stock flotation is in general an alternativeto bond finance, it at least serves to dispel one possible implication ofthe Lintner view of dividend behavior. Lintner seems to imply that firmsmanifest a strong desire to maintain a rigid dividend policy. Thus, itwould appear that in years when profits are unusually low this allegedrigidity should lead to added borrowing. But in the two recession yearsof our sample (1954 and 1958) the coefficient of the dividend variableis significantly negative, and hence the rigidity view does not appear tobe supported empirically.

Before concluding this section, it might be desirable to give a fewnegative results. As an alternative to the fully simultaneous modelpresented above, one might consider a recursive model. Thus, one mightpostulate that investment is independent of the dividend and externalfinance decisions, dividend decisions depend only on investment, whileexternal finance depends on both the dividend and investment decisions.The results of this estimation, not presented here in the interest of con-serving space, are generally similar to those obtained in the fully simul-taneous model, except that in the investment equation the rate of profitvariable is frequently insignificant, but whenever it is significant—inthree years—it has a negative sign. If we note that the dividend variablein the investment equation has a significantly negative coefficient andthat the rate of profit variable in the dividend equation is consistentlyvery significant with a positive coefficient, then the phenomenondescribed seems to indicate that the simultaneous model is the morerelevant of the two.


Another failure worth noting was connected with the introductionof lagged dividends (D/S)1 as an explanatory variable in the dividendequation. Aside from any other problems that might arise when laggedendogenous variables serve as predetermined variables, multicollinearitywas so severe as to preclude any meaningful results. Hence this wasabandoned. On the other hand, we performed some single-equationcomputations of the dividend equation using lagged dividends as anexplanatory variable. It turns out that its coefficient is highly significantand positive although the remaining explanatory variables still retaintheir significance. We should point out in this connection that we donot consider very meaningful the introduction of lagged endogenousvariables as explanal:ory variables in cross-section studies of the sortpursued here. Hence we are not particularly concerned with the severeincidence of multicollinearity in these equations.

ConclusionThe objective of this study was to determine the structure underlyingthe dividend-investment-external finance triad of decision-making pro-cesses. The aim was not so much to attain predictive sharpness butrather to aid our understanding of the complex of relations that bindthese policies together and, in conjunction with information exogenousto the firm, determine its actions. Moreover, we were concerned withdemonstrating the simultaneous character of these decision-makingprocesses.

The sample employed was a cross-sectional one involving 181 manu-facturing, mining, and retail trade firms for which we had a continuousrecord over the period 1947—60. The method of investigation con-sisted of estimating the structure of our model successively for the years195 1—60. The main findings are in brief the following:

1. The single equation approach obscures the character of thesedecision-making processes.

2. Full information (three-stage least-squares) estimation methodsare indicated since the correlation coefficients of the limited information(two-stage least-squares) residuals turn out to be quite significantlydifferent from zero.

3. There is a significant degree of interdependence between theinvestment and dividend decision-making processes, with the implica-tion that if dividend policies are very rigid as some allege, then thisrigidity may tend to hamper the investment activity of firms. On theother hand, our results tend to show that the investment requirementsof firms tend to have a significant effect on their dividend behavior.


This result is in accord with the findings of Dhrymes and Kurz [3], inthe case of electric utilities.

4. The external finance activity of firms seems to be strongly affectedby their investment policies, but less so by their dividend policies.

5. There is considerable evidence that elements of the acceleratorversion of investment theory are empirically quite relevant. On the otherhand, our results seem to indicate that this should not be interpretedas a denial of the usefulness of the profits (rate of profit in our view)version. The manner in which profits affect investment is quite complex.They may tend to register an impact in several directions through themanner in which they affect directly the investment, the dividend, andthe external finance decision-making processes.

6. Although our results differ markedly from the results of severalcited studies, nonetheless if we interpret their findings as pertaining tothe reduced form of the structure we have estimated here, then theirresults become compatible with ours. Thus, our results are best under-stood as generalizing and putting into proper perspective several previ-ous findings pertaining to the investment and dividend behavior of firms.

In conclusion, we should point out that we offer this study by wayof a preliminary exploration of our topic. Quite clearly much may bedone further to elucidate more clearly the cyclical variation of thestructure of these decision-making processes. The rather crude methodof successive cross sections used here, while it does very definitelypoint to the presence of some cyclical pattern, nonetheless does notadequately pinpoint it.

Beyond that there are some data problems relating to the measureof the capital stock, the market value of the firm, and the interest ratevariables that might have been handled more adequately. But, despitethese reservations, we think that several useful conclusions may bederived from this study.

Appendix(The list of firms comprising our sample is given below

by industrial classification.)INDUSTRIAL CLASSIFICATION I INDUSTRIAL CLASSIFICATION 1 (coNT.)

Boeing Airplane Company Alco Products IncorporatedCurtiss-Wright Corporation Pullman Incorporated

Baldwin-Lima-Hamilton Corp.Douglas Aircraft Company, Inc. Chrysler CorporationLockheed Aircraft Corporation General Motors CorporationNorth American Aviation, Inc. Studebaker-Packard Corp.United Aircraft Corporation American Motors Corp.American Car and Foundry Co. The White Motor Co.Martin Company Mack Trucks, Inc.


INDUSTRIAL CLASSIFICATION 2Bond Stores, Inc.J. C. Penney CompanyAllied Stores CorporationAssociated Dry Goods Corp.City Stores CompanyFederated Department Stores, Inc.Gimbel Brothers, Inc.R. H. Macy & Co., Inc.Marshall Field & CompanyThe May Department Stores Co.American Stores CompanyFirst National Stores Inc.The Kroger CompanySafeway Stores, Inc.W. T. Grant CompanyS. S. Kresge CompanyS. H. Kress & CompanyG. C. Murphy CompanyF. W. Woolworth CompanyMontgomery Ward & Co., Inc.Sears, Roebuck and CompanyINDUSTRIAL CLASSIFICATION 3The American Tobacco CompanyLiggett & Myers Tobacco CompanyR. I. Reynolds Tobacco CompanyP. Lorillard Company.U.S. Tobacco CompanyPhilip Morris, Inc.The Borden CompanyNational Dairy Products Corp.National Distillers Products Corp.Schenley Industries, Inc.Hiram Walker—Gooderham & Worts Ltd.Distillers Corp.—Seagrams LimitedGeneral Mills, Inc.Pillsbury Mills, Inc.Penick & Ford, Ltd., Inc.Armour and CompanyThe Cudahy Packing CompanySwift & CompanyWilson & Company, Inc.General Foods CorporationStandard Brands Inc.INDUSTRIAL CLASSIFICATION 4Abbott LaboratoriesMerck & Company, Inc.Parke, Davis & CompanyColgate-Palmolive CompanyThe Procter & Gamble CompanyGulf Oil CorporationPhillips Petroleum CompanyShell Oil Company

INDUSTRIAL CLASSIFICATION 4 (coNT.)Sinclair Oil CorporationSocony Mobil Oil Company, Inc.Standard Oil Company of CaliforniaStandard Oil Company—IndianaStandard Oil Company—New JerseyThe Texas CompanyThe Firestone Tire & Rubber CompanyThe B. F. Goodrich CompanyThe Goodyear Tire & Rubber CompanyUnited States Rubber CompanyGeneral Tire and Rubber CompanyEagle-Picher CompanyAmerican Home Products CorporationSterling Drug Inc.Vick Chemical CompanyAllied Chemical & Dye CorporationAmerican Cyanamid CompanyThe Dow Chemical CompanyDuPontHercules Powder CompanyMonsanto ChemicalUnion Carbide

INDUSTRIAL CLASSIFICATION 5

Allis-Chalmers Manufacturing Co.Deere & CompanyInternational Harvester CompanyJ. I. Case CompanyAmerican Chain & Cable Company, Inc.Blaw-Knox CompanyCombustion Engineering, Inc.Fairbanks, Morse & CompanyIngersoll-Rand CompanyLink-Belt CompanyWorthington CorporationThe Budd CompanyBendix Aviation CorporationDana CorporationEaton Manufacturing CompanyThe Electric Auto-Lite CompanyStewart-Warner CorporationThompson Products, Inc.The Black and Decker Mfg. Co.Ex-Cell-O Corporation

INDUSTRIAL CLASSIFICATION 6General Electric CompanyWestinghouse Electric Corp.Cutler-Hammer, Inc.The Electric Storage Battery Co.McGraw-Edison CompanyMinneapolis-Honeywell Regulator Co.General Cable Corporation

466 Producer DurablesINDUSTRIAL CLASSIFICATION 6 (coNT.)Admiral CorporationMotorola, Inc.Philco CorporationRadio Corporation of AmericaZenith Radio CorporationAvco Manufacturing Corp.


Johns-Manville CorporationNational Gypsum CompanyThe Ruberoid CompanyUnited States Gypsum CompanyGeneral Refractories CompanyAlpha Portland Cement CompanyArmstrong Cork CompanySmith-Corona Marchant, Inc.International Business Machine Corp.Lehigh Portland Cement CompanyAmerican Radiator & Standard Sanitary

CorporationCrane CompanyOtis Elevator CompanyLone Star Cement CorporationHarbison-Walker Refractories CompanyPenn-Dixie Cement CorporationYale and Towne Manufacturing Co.U.nited States Pipe and Foundry Co.


American Can CompanyContinental Can Company, Inc.Anchor Hocking Glass CorporationOwens-Illinois Glass CompanyMasonite CorporationLibbey-Owens-Ford Glass CompanyThe Champion Paper and Fibre Co.

INDUSTRIAL CLASSIFICATION 8 (coNT.)Container Corporation of AmericaCrown Zellerbach CorporationInternational Paper CompanyKimberly-Clark CorporationThe Mead CorporationRayonier Inc.St. Regis Paper CompanyScott Paper CompanyUnion Bag-Camp Paper Corp.West Virginia Pulp and Paper Co.Burlington Industries, Inc.Cannon Mills CompanyJ. P. Stevens & Company, Inc.United Merchants & Manufacturers, Inc.Beaunit Mills, Inc.American Viscose CorporationBigelow-Sanford Carpet Company, Inc.Simmons Company

INDUSTRIAL CLASSIFICATION 9The International Nickel Company of

Canada, Ltd.Bethlehem Steel CorporationInland Steel CompanyJones & Laughlin Steel CorporationNational Steel CorporationRepublic Steel CorporationUnited States Steel CorporationArmco Steel CorporationRevere Copper and Brass, Inc.Aluminum Company of AmericaKaiser Aluminum and Chemical Corp.Olin Mathieson Chemical Corp.Reynolds Metals CompanyThe Anaconda CompanyKennecott Copper CorporationPhelps Dodge Corporation

BIBLIOGRAPHY1. Brittain, J. A., "The Tax Structure and Corporate Dividend Policy,"

American Economic Review, May 1964.2. Darling, P. G., "The Influence of Expectations and Liquidity on Dividend

Policy," Journal of Political Economy, June 1957.3. Dhrymes, P. J., and Kurz, M., "On the Dividend Policy of Electric

Utilities," Review of Economics and Statistics, February 1964.4. Dobrovoisky, S. P., "Economics of

Financing," The Journal of Finance,5. Eisner, R., "A Distributed Lag Investment

January 1960.

Corporate Internal andMarch 1958, p. 35.

External

Function," Econometrica,

r -


6. , "Investment: Fact and Fancy," American Economic Review,Papers and Proceedings, May 1963.

7. Eisner, R., and Strotz, R., "Determinants of Business Investment," inImpacts of Monetary Policy, Commission on Money and Credit, Engle-wood Cliffs, N.J., 1963.

8. Hotelling, H., "The Selection of Variates for Use in Prediction withSome Comments on the General Problem of Nuisance Parameters,"Annals of Mathematical Statistics, September 1940, p. 271.

9. Jorgenson, D. W., "Anticipations and Investment Behavior," in J. Duesen-berry et a!. (eds.), S.S.R.C. Econometric Model of the U.S., Chicago,1965.

10. Kaldor, N., "Economic Growth and Cyclical Fluctuations," EconomicJournal, March 1954.

11. Kalecki, M., "The Principle of Increasing Risk," Economica, November1937.

12. Klein, L. R., Economic Fluctuations in the United States, 1921—1 941,New York, 1950.

13. Kuh, E., Capital Stock Growth: A Micro-Econometric Approach,Amsterdam, 1963.

14. Kuh, E., and Meyer, J. R., "Investment Liquidity and Monetary Policy,"in Impacts of Monetary Policy, Commission on Money and Credit, Engle-wood Cliffs, N.J., 1963.

15. Lintner, J., "The Determinants of Corporate Savings," in Savings in theModern Economy, W. Helter (ed.), Minneapolis, 1953.

16. , "Distribution of Incomes of Corporations Among Dividends,Retained Earnings and Taxes," American Economic Review, Papers andProceedings, May 1956.

17. Mayer, J. R., and Kuh, E., The Investment Decision, Cambridge, Mass.,1957.

18. Meyer, J. R., and Glauber, R. R., investment Decisions Economic Fore-casting and Public Policy, Graduate School of Business Administration,Boston, 1964.

19. Modigliani, F., and Miller, M. H.,- "The Cost of Capital, CorporationFinance and the Theory of Investment," American Economic Review,June 1958.

20. Roos, C. F., "A Dynamical Theory of Economics," Journal of PoliticalEconomy, October 1927.

21. Theil, H., Economic Forecasts and Policy, 2nd ed., Amsterdam, 1961.22. Tinbergen, J., "Statistical Evidence on the Acceleration Principle,"

Economica, May 1938.23. , Statistical Testing of Business Cycle Theories, Geneva, 1938.24. U.S. Congress, Joint Economic Committee, Variability of Private Invest-

- mens in Plant and Equipment, Part I: Investment and Its Financing,Washington, 1962.


COMMENTON DHRYMES-KURZ AND ANDERSON

BY HENRY A. LATANE, UNIVERSITY OF NORTH CAROLINA

There are two types of analysis used in dealing with the investmentand financing decisions of the firm. The first is the standard marginalanalysis based on certainty or point estimating models with modifica-tions to reflect uncertainty. The second is to approach the problemfrom the point of view of the actual decision maker. Dhrymes and Kurzstate that the objective of their study was to determine the dividend-investment-external finance triad of decision-making processes, thusapparently attempting to take the second approach. In any event I amgoing to use this approach in talking about the two papers.

Before making direct comments, I would like to take a few momentsto develop a frame of reference. The problem we are dealing with—business fixed investment and the financing and dividend policies of thefirm—are directly analogous to the types of decisions faced by therational portfolio manager in handling financial assets and liabilities.The rational portfolio manager decides how much to withdraw for livingand other expenses each year and how much to reinvest. This corre-sponds to the dividend decision. He also has two different types ofproblems. In the first place, he must form probability beliefs as toreturns from individual assets under various states of nature. In somegambling and insurance situations risk can be virtually eliminated bydiversification, but this is not true with most economic assets becauseof the covariance of yields. If one venture is profitable so will be othersand vice versa. Probability beliefs about distributions of returns fromvarious combinations of assets also are needed. In the second place,the portfolio manager must allocate his portfolio in the light of hisbeliefs about returns from available portfolios. The portfolio managerhas the option to borrow and lend within restrictions, and in order tomake rational choices he must have some trade-off function betweenreturns and risk.

The rational entrepreneur has virtually the same problems. He mustform probability beliefs not only about individual investment opportuni-ties but also about returns to the firm as a whole if various alternativeinvestment policies are adopted and various states of nature occur. Hemust decide upon proper financing or financial reserves. He too musthave some trade-off functions between risk and return in order to makerational decisions.

Comment 469

I wifi talk first about the variables used by Dhrymes and Kurz,secondly about variables omitted in both papers, and finally about thetrend in the cost of funds.

In the Dhrymes-Kurz model sales are used as a deflator for dividendsand investment. Dividends divided by sales turn out to be positivelycorrelated with investment divided by sales, and the coefficient of thedividend variable is significantly positive when used as one of theexplanatory variables to explain the investment variable. They say "theastonishing result in the investment equation is the sign and nearlyconsistent significance of the coefficient of the dividend variable . . . itis difficult to rationalize the positive sign of the dividend coefficientwhich points to the potential pitfalls of the single-equation approachand the ordinary tests of significance." However, there is a plausiblerationalization for this positive correlation. It is in the choice of salesas a deflator. Even if there is some small negative correlation betweendividends per dollar of book and investment per dollar of book, asindeed would be expected if dividends and investment are accepted asalternative uses of funds, there would be a positive correlation betweendividends per dollar of sales and investment per dollar of sales if thereare significant differences in sales per dollar of book value. Consider,for example, a meat packer and a public utility. The meat packer paysa dividend equivalent to 5 per cent of book and invests 12 per cent ofbook, the public utility pays 6 per cent of book and invests 10 per centof book. The dividend ratio is negatively correlated with the investmentratio. But the meat packer has $10 of sales for each $1 of book, whilethe public utility has only 25 cents of sales for each $1 of book, so thedividend-sales ratio for the meat packer is .5 per cent and the invest-ment-sales ratio is 1.2 per cent, while the corresponding ratios for thepublic utility are 24 and 40 per cent. The dividend-sales ratios arepositively correlated with the investment-sales ratios. There may wellbe other reasons for preferring the full information approach to thesingle equation—indeed I am sure of it—but the positive correlationbetween the dividends and investment ratios used by Dhrymes and Kurzdoes not seem to me to be one of them.

Let us now consider the variables omitted from consideration. Pos-sibly the most striking example of this is Anderson's omission of the 1957data. He says "Since I firmly believe that the continuation of the mid-1950's investment boom into 1957 was collective madness ex ante, notto mention ex post, I had little compunction about pulling the 1957observation and refitting the equations without it. The object of thiswas to increase the accuracy of the parameter estimates and not, of


course, to raise the R2." I too would accept "collective madness" as auseful variable. The airline order curve for new equipment is a classicexample of this sort. One airline ordered a new model and the race wason. I might agree with him that, in the absence of a good measure ofcollective madness, we should omit contaminated data. However, Iwould hesitate to claim that this procedure increased the accuracy ofthe' parameter estimates unless the limited population about which Iwas talking were very clearly specified.

Probably a more serious omission in both papers is the omission oflagged variables in the explanatory equations. There are sound econo-metric reasons for this, but no one can claim that a model withoutlagged variables adequately reflects the structure underlying the divi-dend-investment-external finance triad of decision-making processes.This is particularly true as it affects the dividend decision. What thecompany has been paying in the past is a matter of prime concern tothe directors who are deciding what to pay in the future. Not onlydoes the omission of lagged dividends break with the actual decisionprocess, but also it upsets the weights to be used with other estimatingvariables. For example, I have run simple cross-section regressions toexplain the prices of the thirty stocks in the Dow Jones IndustrialAverage, using dividends and retained earnings only as the explanatoryvariables—all variables being deflated by book value. The regressioncoefficients vary widely from year to year, but the median values for theeight years 1956—1963 were as follows:

If now we add the price last year to the explanation we get

= —2 + 5.6D + 3.0 RE +In three out of the eight years retained earnings carry more weight

than dividends. Since the lagged price is available to anyone who isinterested, it is clear that, in the actual decision-making context,retained earnings carry about as much weight as dividends. In the firstequation the dividend variable acts as a proxy to take up the weightwhich should have been placed on past price.1

Finally, Anderson states that there is a secular uptrend in the positionof the marginal cost of funds schedule. If he means. only the interestrate when he says marginal cost of funds, I would agree. Not only areinterest rates relatively high, but firms are economizing on their cash

1 See Irwin Friend and Marshall Puckett, "Dividends and Stocks Prices" Ameri-can Economic Review, September 1964, pp. 676—677.

Comment 471

balances, thus indicating that the cost of safe capital is relatively high.However, there is another type of capital—risk capital, whether derivedfrom retained earnings or stock issuance—and there is no convincingevidence that the cost of this type of capital has increased. Industrialstock prices now average over twice book value compared with 1.2times in 1950 and 1.4 times in 1957. Dividend yields have been cut inhalf. Part of this higher price-to-book ratio and lower dividend yieldsmay have resulted from smaller uncertainty surrounding expectedyields, but the reduction in uncertainty hardly has been enough to jus-tify the assumption that the marginal cost of equity capital has risen.

ON DHRYMES-KURZ AND ANDERSON

BY R. W. RESEK, UNIVERSITY OF ILLINOIS

In past years economists discussed the problems of business invest-ment which were emphasized in these papers, but when the time cameto make empirical estimates they frequently returned to either accelera-tion or capacity-output theories. This emphasis was not due to a lackof trying to bring in other factors but was largely forced upon them bythe lack of data as well as by variables that failed to vary. The authorsconsistently tried to incorporate more information into the explanations,but the endeavors were generally not too successful.

More recently, empirical studies have been made which significantlyextend the evidence supporting alternative theories that improve ourexplanation of investment. They accomplish this by complementing analready accepted theory—not competing with it.

The papers presented here both emphasize the cost and availabilityof funds and its effect on investment. A number of possible positionscan be taken concerning the cost of funds.

One can say that the cost of funds is fixed for all firms in the econ-omy by the appropriate rate of interest. Jorgenson takes this positionin his empirical work1 and places emphasis on it in his theoreticalwork as, for example, that presented in this conference. He does find,of course, that investment is a decreasing function of the rate of interestbut the rate is not dependent on the internal financial structure ofthe firm.

Miller and Modigliani have made an important contribution in thisarea. Earlier in this Conference they attempted to achieve a real mea-

Dale W. Jorgenson, "Capital Theory and Investment Behavior," AmericanEconomic Review, May 1963, pp. 247—259.


sure of the cost of funds. In their view, this cost may differ from firmto firm. It does however remain constant for any given firm, regardlessof financial structure (apart from tax complications).

In contrast to these views, Lintner in his paper indicates the beliefthat cost of funds is related directly to the internal financial structureof a firm and may change according to the volume of investment beingmade by the firm.

The two papers presented here are consistent with this latter view.Anderson explicitly introduces an expression for the marginal cost offunds to the firm. His expression relates this Cost to the cost of equityand bond financing and to the internal financial position of the firms inthe industry. His model is estimated at a very high level of aggregationso the results may be obscured by aggregate relations. However, his lagstructure is likely to minimize this problem.

Dhrymes and Kurz construct a model which emphasizes the totalvolume of funds available and allocation of these funds by the firm.This model does not put specific emphasis on other phases of the costelement but the approach is closely related to Anderson's.

Dhrymes and Kurz build a three-equation model based on the alter-native sources and uses of funds for a firm. They depend on theaccounting identity:

Ii+12=EF1+EF2+P—D+Dep. (2)

This equation states that fixed investment plus inventory andother short-term investment (12) are the uses of funds. These mustequal the sources of funds. These are external finance from bonds(EF1), external stock finance (EF2), and the internally generatedfunds-profits (P) minus dividends (D) plus depreciation (Dep). Theyassume short-term investment to be predetermined for the model, aswere profits and depreciation. Stock financing is assumed to be resid-ually determined. Thus only three equations must be estimated. Theseare for fixed investment, dividends, and bond external financing.

Their particular specification of the model was:2D (IEFPN

I (DEF/P\ NS—S(t—3)S(t— 3)

EF I PDep R LTD

K LTDK—LTD2 Time subscripts are added only when they are not t.

Comment 473

The variables are defined in Dhrymes and Kurz (hereafter referredto as D-K). Several peculiarities in this specification seem immediatelyapparent. First, the normalization of variables is half related to salesand half to capital stock. One cannot help but wonder what the out-come would have been if capital stock had been used throughout.Since sales are likely to fluctuate more than capital stock, the resultsmay be sizably affected by the sales. For example, if the investmentequation is solved to the left,

I N S(t)J=h2<D,EF,S•I—) ,S.—,[S—S(t— 3)]K S(t—3)

Both profits and the net current position of the firm are multipliedby sales. D-K find that both of these variables are frequently significant.Because of their specification, they may be finding the effect of salesrather than profits. The same type of situation exists in each equationof the model. Moreover, the constant term may have a coefficient thatis attributable to sales and that explains the variation which otherwisewould be explained by the accelerator variable, [S — S (1 — 3)].

Secondly, one can question the lag structure used in the model.Considerable evidence has been accumulated indicating that there is asizable lag between decisions to invest and the specific expenditure. Theinterest and belief in this lag is evidenced by this Conference's specificpapers on anticipations and appropriations. Well-known papers byAlmon3 and de Leeuw4 have made specific estimates of the size of thelag. One conclusion that might be reached from this point of view isthat the investment decision is reached first and that the method offinancing the investment is subsequently and separately determined. Thedecision concerning financing would take place when the money wasactually spent. This might tend to indicate that a recursive model iscalled for. D-K indicate they attempted such a model unsuccessfullybut since the lags employed in that attempt are unknown, the questionremains unsettled.

An alternative approach (which I prefer) would be that, at the timeinvestment decisions are made, tentative plans for financing are alsoconsidered. Certainly the variable cost of funds theory requires thisview since the type of financing affects the cost of funds. This viewsays the D-K model is inappropriate because nearly all of the variablesare used with no lag whatsoever while they should be lagged.

Shirley Almon, "The Distributed Lag Between Capital Appropriations andExpenditures," Economelrica, January 1965, pp. 178—196.

Frank de Leeuw, "The Demand for Capital Goods by Manufacturers: A Studyof Quarterly Time Series," Econometrica, January 1962, pp. 407—423.


This question about the lags leads to a related problem concerningthe identification of the model. P/K appears unlagged in the dividendmodel but lagged one year in the investment equation. This differenceis not explained. Thus profits affect future investment but do not, atthat same instant in time, affect the other future uses.of funds, e.g., divi-dends. If the lag structure in the model is inappropriate, in particular ifprofits appropriately should appear with the same lag in both of thefirst two equations, then the investment equation is not identified andone will estimate only a linear combination of the first two equations.This problem is particularly troubling because this part of the specifica-tion is clearly uncertain, and as a result the investment equation esti-mates are unreliable.

Finally, the specific choice of variables can be questioned. In justi-fying the dividend equation, mention is made of the work of Lintner.His principal finding is that present dividends are largely dependent onpast dividends. Dhrymes and Kurz did estimate this equation with thisvariable but they report that, ". . . multicollinearity was so severe as topreclude any meaningful results. Hence this was abandoned." Thistype of problem is acute whenever it arises, but it does not seem to beappropriate to exclude the single most important explanatory variablewhen the multicoilinearity arises. The resulting is sosevere as to preclude any meaningful results.

This particular problem arises since the lagged endogenous variableis the explanatory variable. Frequently this type of variable is employedto represent the lag structure, as suggested by Koyck and Jorgenson.In this type of usage the lagged endogenous variable serves as a proxyfor most of the lagged dependent variables. Instead of such a proxy,it may be argued that it is preferable to avoid using the endogenousvariable and to represent the lag in some other fashion. This argumentdoes not apply to the present situation as the lagged dividends are,according to the theory, in and of themselves an important determinantof dividends.

Investment in the D-K model depends on the change in sales, profits,and variables indicating the financial position of the firm. Recent modelshave, in general, found that capacity adjustment mechanisms were atleast as effective as those representing the acceleration principle throughchange in sales. A model employing sales and capital stock as variablesis therefore a strong alternative to the use of change in sales.

The rate of profit is used 'in this equation as a variable, "which maybe taken to lead to changes in the expected profitability of new invest-ment." Generally profit theories depend on the cash flow arising from

Comment 475

profits which lead to a lower cost of funds as well as future profits.This is indirectly taken into consideration in this model through theeffects of profits on dividends and external finance and the resultingeffect on investment. It is generally felt that profits and sales are highlycorrelated variables. Since sales enters as a first difference [S — S(t — 3)]and profits directly, one cannot help but wonder whether the coefficientsmeasure the relative importance of original variable versus first differ-ences rather than the profits versus sales.

One should also question the information on internal funds of thefirm contained in the investment equation. The financial position of thefirm is indicated through dividends, external finance, and the net cur-rent position of the firm. The quantity of money available for invest-ment is presumably contained in these. The cost of funds seems onlyto operate in affecting the external funds which in turn affect invest-ment. The implication seems to be that the cost of funds schedule isperfectly elastic at a very low rate up to some critical value where itbecomes perfectly inelastic. It seems much more realistic to say thatthe interest rate affects investment directly through the cost of funds aswell, as indirectly through the external financing. Even a firm with noexternal funds would be affected by the interest rate although the modeldoes not allow for it.

Finally, the use of N/K by the authors should be considered. Theyindicate that this variable, ". . . enters the model as a consequence ofthe use of the budget constraint to eliminate one of the variables in thesystem." The nature of this relation and the sign this would imply forthe coefficient are somewhat unclear. Anderson employs a similarvariable lagged one period and gets a positive relation with investment,indicating that since current funds are available, investment will behigher. Dhrymes and Kurz find a negative value for the unlagged vari-able. This is somewhat harder to explain, but apparently means thatfunds have been used for current purposes and therefore investmentcould not take place during this period.

Up to this point we have commented on four major features of thespecification of the D-K model. The normalization provides some ques-tion concerning what is being measured; the lag structure was not care-fully formulated; the second equation is nearly underidentified; andfinally the variables used in the equations certainly are not the onlypossible candidates for inclusion. These comments are nothing new foreconometric work as every author must choose among mutually exclu-sive alternatives, but in this model specification problems seem par-ticularly severe.


Three alternative sets of results are computed by D-K. These areone-stage, two-stage, and three-stage least squares. They reject ordinaryleast squares as inappropriate early in the paper. There can be no ques-tion but that they are correct in doing this and are providing a stepforward, given their particular model. Their use of three-stage leastsquares is much more to be questioned. They justify it on the groundsthat the correlation coefficients between the two-stage residuals arehigh. This is correct and it is true that this is the type of informationthat three-stage least squares use effectively.

On the other hand, the model employed may be greatly in error dueto specification errors as discussed above. No estimating techniqueunder consideration will provide consistent estimates of parametersunder this type of situation. Two-stage estimates may be less affectedby the specification errors than three-stage estimates, although truesampling properties are simply not known under this type of error. Ifthis is true, their two-stage estimates may be more reliable than thethree-stage values emphasized by the authors.

Let us turn now to the paper by Anderson. He sets up a two equa-tion model explaining the marginal rate of return and the marginalcost of funds for firms. Since these must be equated by a rational inves-tor, he then solves for investment and estimates the reduced formequation.

The marginal rate of return here is dependent on output and capitalstock as well as the level of investment. This is reasonable and may beobtained from a capacity adjustment model. Marginal cost of fundsdepends on the internal financial position and the external cost offunds. The former here is noncapital assets minus liabilities plusretained earnings. This variable then represents approximately what thefinancial situation would be at the end of the period if all internallygenerated funds were used to improve this position. This formulationseems to be an excellent way to handle the generally difficult fundsproblem. In particular, Anderson seems to avoid the problem of col-linearity that plagues alternative formulations of this variable.

The external cost of funds is measured by the equity yield and bondyield. However, it is clear that the yield on equities reflects more thanthe cost of equity funds. It is also greatly affected by the growth expec-tations of the firm. A firm with great prospects for growth would havea much lower yield than one with low growth prospects. Thus thisstock variable reflects both expectations for output and cost of funds.The former, which I consider to be more impOrtant, would appropri-ately be considered as a determinant of the marginal rate of return of

Comment 477

the firm. This in no way affects the reduced form model to be estimatedbut does significantly affect the interpretation to be put on the resultsobtained.

Anderson's results strongly support the hypothesis being tested andstrongly support the cost of funds theories of investment behavior.Clearly, as he points Out, the implications for governmental policy maybe sizable but are likely to be complex.

Anderson uses very aggregate data, which may lead to problemsrelated to the contrast between macrodata and use of what is essen-tially a microtheory. Anderson justified his model by saying that,

policy formulation and forecasting often require quick and dirtyestimates of economic parameters." Clearly the proof of this modelwill be in the forecasting ability of the equation obtained.

Together, these two papers provide a great deal of light on theinternal financial behavior of finns and the effect of this on investment.

ON DHRYMES-KURZ AND ANDERSON

BY WILLIAM VICKREY, COLUMBIA UNIVERSITY

Dhrymes and Kurz examined the concept that in a large firm deci-sions on investment policy, dividend policy, and external finance arenot the components of a single-decision vector arrived at by a unitarymanagement on the basis of exogenous influencing factors; rather theyare somewhat independent, though mutually interrelated, decisionsarrived at and modified at different times and by different groups. Thisconcept has considerable appeal as an attempt to improve understand-ing of the large organism. And it is reasonably plausible that this inter-related but nonmonolithic decision process can, to some extent, berepresented by a structural model in which these interactions amongthe dependent variables can be exhibited. However, the results of thisexperiment do not seem to me to be as conclusive as the authorsappear to believe.

The statistical results of their model are summarized in Table 1,in which the entries represent the range of the t-coefficients for the tensingle-year cross sections, with some comments on their behavior. It isclear that the equations of the model are formally overidentifled, whichwould provide an opportunity for the use of some test, such as thelikelihood ratio, as an indication of whether the model fits the datasufficiently well compared with a less constrained one to maintain thehypothesis that the excluded coefficients are in fact zero. As it is, we


are asked to take the stipulated constraints of the model more or lesson faith and to suppress any suspicion we may have that knowledgeof the leverage position of the firm or the interest rates that it paysmight directly influence the decisions of those concerned with dividendpolicy, or of those concerned with investment.

If the derivation of a specific structure has any usefulness, over andabove what can be done with direct least-squares estimation of theendogenous variables in terms of the exogenous ones, it lies in thecoefficients in the first three lines, which express the internal structureof the decision mechanism concerning which we are seeking informa-tion. Unfortunately the results seem meager. The only coefficient toshow a reasonable amount of consistency is the inhibiting influence ofcommitments as to dividends on investment decisions. Even this issomewhat suspect, inasmuch as current sales used to deflate investmentand dividends also occurs as a component of the ëhange in sales vari-able, which appears as an exogenous explanatory variable. As for tak-ing the coefficients for the individual year structures at face value, it isreally hard to believe that the composite internal decision structure offirms changes that much from one year to another. I cannot escape theimpression that the model at hand is much too procrustean to provideus with any reliable insights into the internal structure of the firm.

As for the possible uses of the model for forecasting, it is my under-standing that, as long as one does not anticipate any exogenouslydeterminable change in the effective structure of the organism whosebehavior is being predicted, straightforward least-squares regression ofthe single endogenous variables on the exogenous variables provides asgood predictions as more elaborate methods. It is only when one expectsto be able to predict independently some change in the structure thatdiffers in character from changes occurring in the period of observationthat knowledge of the specific structure becomes important, as whenit is desired to predict the effect of a change in a tax rate that has beenstable during the observation period. In the case at hand, I see littlelikelihood of there being available any such independent informationof a general change in the internal organization of large businesses.Accordingly, the usefulness of the results for forecasting purposeswould seem to me to reside in the single-equation least-squares results.Unfortunately, even these are not quite of the nature required, sincethe other endogenous variables are included in each of the single equa-tions as explanatory variables, which makes a consistent forecastdifficult.

TAB

LE 1

Sum

mar

y of

Sta

tistic

al R

esul

ts o

f Dhr

ymes

and

Kur

z

Var

iabl

e Ex

plai

ned

by th

eEq

uatio

n

Div

iden

ds P

erIn

vest

men

t Per

Long

-Ter

m E

xter

nal

Var

iabl

eU

nit o

f Sal

esU

nit o

f Sal

esFi

nanc

e Pe

r Uni

t of S

ales

Endo

geno

us V

aria

bles

Div

iden

ds p

er—

l—

l.O:—

4.9

unit

of sa

les

smal

l at p

eaks

erra

tic

Inve

stm

ent p

er—

0.6:

—2.

7—

1—

0.9:

+4.8

unit

of sa

les

erra

ticer

ratic

; mos

tly +

Long

-ter

mex

tern

al fi

nanc

e—

0.8:

+2.9

—1.

1—

1pe

r uni

t of s

ales

erra

ticer

ratic

Exog

enou

s Var

iabl

esIn

dust

ry d

umm

y va

riabl

es4,

5, 7

, 8, 9

4, 5

, 7, 8

, 9ex

clud

edge

nera

lly si

gnifi

cant

Prof

its p

er u

nit

+2.3

:+6.

3ex

clud

ed—

1.8:

+2.1

5of

cap

ital

fairl

y st

able

erra

tic; +

atpe

aks

Net

shor

t-ter

m a

sset

s per

—1.

2:—

3.O

—l.5

:—3.

1ex

clud

edun

it of

cap

ital

fairl

y st

able

fairl

y st

able

Prof

its th

is y

ear p

er u

nit

excl

uded

—0.

28 +

2.5

excl

uded

of c

apita

l las

t yea

rer

ratic

Rel

ativ

e ch

ange

in sa

les

excl

uded

—.1

3 : ÷

3.6

excl

uded

over

the

past

3 y

ears

mos

tly +

Deb

t-to

-equ

ityra

tioex

clud

edex

clud

ed—

3.0

: +1.

1m

ostly

—

Inte

rest

rate

on

debt

excl

uded

excl

uded

—.0

8 :-.

6.7

Rat

e of

dep

reci

atio

nex

clud

edex

clud

ed—

2.8

: +1.

9al

low

ance

sm

ostly

—


By way of contrast, the statistical techniques employed in theAnderson paper are relatively simple-minded, and the results straight-forward and convincing. My only comment is that it seems to me that,in attempting to reconcile the timing of variables relating to a wholeyear and those relating to a point in time, it would be preferable torelate the flow over a given year to an average of the stocks at thebeginning and end of the year, rather than, as Anderson does, relatethe stock at a point in time to an average of the flows for the precedingand subsequent periods. Not only do the data in the former case spanonly one year in time rather than two, but the former treatment seemsto conform more closely to the data that would be uppermost in themind of a decision maker at a given time. Even if he does not at a giventime have before him the stock figures for the end of the current year,but only the flow data for, say, the first eleven months of the year, itseems plausible to say that he will be more strongly influenced by hisrough knowledge of what the probable stock situation will be at theclose of the year than he will by what the flow was for the precedingfull year. Thus for an output-capital ratio it would seem preferable touse 2Qt/(Kt + rather than + Qji)/2Kti, and similarly

+ rather than + as the retained earn-ings element in the financial position. Otherwise the results seem tospeak for themselves.

REPLY TO RESEK, AND VICKREY

BY DHRYMES AND KURZ

We are indeed very grateful to the discussants of our paper for theirdetailed comments. We are, however, dismayed that a number of pointsraised are either adequately covered in our paper or involve an incom-plete understanding of our arguments. We hope that the followingparagraphs will aid in dispelling much of the misunderstanding andelucidate the issues. We shall examine the points raised by each dis-cussant individually and conclude with some general comments.

It is our understanding that Henry Latané raises two points. First,he questions the omission of lagged dividends in the specification

of our model; this was also raised by other dis-cussants and we shall deal with it below. Second, he maintains thatour result relating to the reversal of sign in the coefficient of the divi-dend variable in the investment equation when one uses single- andsimultaneous-equation techniques is the product of our normalization

Comment 481

scheme. He goes on to say "The meat packer pays a dividend equiva-lent to 5 per cent of book and invests 12 per cent of book, the publicutility pays 6 per cent of book and invests 10 per cent of book. Thedividend ratio is negatively correlated with the investment ratio. But themeat packer has $10 of sales for each $1 of book, while the publicutility has only 25 cents of sales for each $1 of book, so the dividend-sales ratio for the meat packer is .5 per cent and the investment-salesratio is 1.2 per cent, while the corresponding ratios for the publicutility are 24 and 40 per cent. The dividend-sales ratios are positivelycorrelated with the investment-sales ratios." While no one will takeexception to the arithmetic of the example, Latané appears to haveentirely misunderstood the issue. First, it is well known that the factthat the (simple) correlation coefficient between the dependent andone of the explanatory variables in a multivariate regression is of agiven sign does not imply that the corresponding multivariate regressioncoefficient must be of the same sign. Second, this example is com-pletely irrelevant to the results we cite. What we find is the following.Noting that there is a body of thought which views the dividend policyof a firm as independent, at least in the short run, of its investmentpolicy (and presumably independent of the error term in the relevantequation), we introduced dividends as an explanatory variable in ourinvestment equation. The corresponding regression coefficient turns outto be significant and positive; when the same relation is estimated aspart of the larger structure by simultaneous-equation techniques, thenthis same coefficient turns out to be significant and negative. This is avery important empirical result, which must not be confused with theproblems, if any, induced by our normalization.

William Vickrey's two comments are that we have conducted no testfor identifiability of the structure we have estimated, and that the resultsare too unstable to permit us an insight into the complex of the inter-dependence we seek to estimate. In regard to the first point, Vickreysuggests that we should have employed a likelihood ratio test. Whilethere is no objection to performing such a test, it simply involves theextraction of characteristics roots of a certain matrix, we do notthink that this is a particularly useful exercise. There is not much evi-dence to convince us that the error terms of the equations of ourmodel have a joint normal distribution; if this is not so, then theexecution of the likelihood ratio test is an empty exercise, the outcomeof which is difficult to interpret.

Concerning the second point, Vickrey apparently bases his conclu-sion on the first three lines of his Table 1. This criticism, however,


seems largely without foundation. A look at our Table 14 will indicatethat, in the dividend equation, the investment variable appears uni-formly with a negative sign, while the external finance variable, whensignificant, appears uniformly with a positive sign. Moreover, the con-tracting influence of investment on dividends is most apparent in peaksand upswings (1952, 1955, 1957, and 1960); the same is generallytrue of the external finance variable, whose positive influence on divi-dends is significant in 1952, 1953, 1955, and 1960.

Important conclusions can also be derived from the investment equa-tion, in which the dividend variable is nearly invariably significant witha negative sign. Our results also indicate a sharp dichotomy in the roleof the external finance variable; thus in the pre-Accord period thisvariable is quite significant and positive in its effect, while in the greatinvestment boom of the mid-1950's it appears to have had no influencesince its coefficient is uniformly insignificant from 1955 through 1960.

Thus, while not claiming that our paper constitutes the definitivework on this topic, we sharply disagree with Vickrey on his secondpoint.

Our results give us a great deal of insight into the complex of thedecision-making process we wish to study and, furthermore, hint atsome very interesting cyclical patterns that bear more careful andsystematic investigation—obviously the next step in this type of research.

Finally, to the best of our understanding, Resek makes the followingpoints: (1) the "lag structure" in our model is not "properly speci-fied"; (2) the second equation is "nearly underidentified"—whateverthat may mean; (3) there is some ambiguity in the interpretation ofcoefficients in the investment equation; (4) several questions of mis-specification are raised, in particular the exclusion of lagged dividendsas an explanatory variable in the dividend equation.

In view of the fact that, as they are presented, the first two pointsare closely related we shall discuss them concurrently. In the firstpoint we are taken to task for ignoring the work of Almon and deLeeuw relating to the lag between appropriations and expenditures incapital investment. On the other hand, in the second point we are takento task for "underidentification" in that we enter the profit rate in thedividend equation and the lagged (one period) profit rate in the invest-ment equation as explanatory variables without adequate justification.

Now if it is true, and surely one might wish to test it, that both elementsof a rate of profit and elements of an accelerator theory govern invest-ment behavior, and if Almon and de Leeuw are correct in pinpointingthe specific lag structure between appropriations for investment and

Comment 483

expenditures for investment, then it follows that our specification of thelagged rate of profit in the investment equation is a perfectly sensiblespecification. In particular, Almon's work shows that within a year afterappropriation the bulk of it has actually been spent, while within six toeight quarters nearly all appropriations have been expended. Since hersample comprises two-digit manufacturing industries and certain manu-facturing aggregates and since our sample consists mostly of firms whichcan be classed as manufacturing, it would appear that we acted properlyin specifying the lagged rate of profit as an explanatory variable in theinvestment equation. Indeed, this "near underidentifiability" of whichour model is accused is the logical consequence of taking into accountthe very works he urges us to take into account in an earlier passage.In this connection, we should mention that more complicated lag struc-tures were attempted in specifying the role of the profit rate in theinvestment as well as the dividend equations, but these generally didnot perform as well as the ones finally reported. Given the choicebetween the simple and the complex, other things being equal, wewould of course choose the simplest specification possible. One possibleexplanation for Resek's views is that he may have overlooked the factthat investment in this context means investment expenditures, notappropriations.

We take it that there is no dispute in using the current profit rateas an appropriate explanatory variable in the dividend equation.

In his third point Resek raises a question of interpretation of thecoefficients of the investment equation in particular. There appears tobe some confusion in his mind about the meaning of the constantterm and the coefficient of the rate of profit variable; he claims that"Because of their specification, they may be finding the effect of salesrather than profits." We do not think there is any ambiguity and shouldbe glad to elucidate the matter. If we "solve" the investment equationby multiplying through by sales (S), then the coefficient of the linearterm containing sales is, in fact, the decomposition of the constant termin our equation, a decomposition incidentally that reveals highly sig-nificant industrial classification contrasts.

The accelerator variable in our version is St — There cannot beany confusion in the manner Resek alleges "Moreover the constantterm may have a coefficient that is attributable to sales and that explainsthe variation which otherwise would be explained by the acceleratorvariable, [S — S(t — 3)]." Now, in its usual empirical formulation theaccelerator version of investment theory does not attribute investmentto "sales" but rather to changes in sales, so it is difficult for us to


understand Resek's criticism. The meaning of these two estimates isquite simple. The constant term relays the following information: con-sider any two firms, which are identical in every other respect except fortheir volume of sales. Then to the extent that there is a significant constantterm, or a significant industrial classification contrast, the two firmswould behave differently with respect to investment. Thus, the constantterm of the equation captures a purely "size" effect; while the contrastscapture both a size and an effect that is to be attributed to the type ofactivity in which the firm in question engages.

On the other hand, consider any two firms which are identical inevery other respect except that they differ in the variable St — i.e.,one has grown more rapidly than the other. Then, to the extent thatthe coefficient of this variable is signfficant, our model would tell usthat investment would differ between the two firms. This is what onecustomarily takes as the accelerator effect. Thus, far from confoundingthe issue, our formulation serves to differentiate clearly between effectsthat are to be attributed to size or industrial classification, and thosethat are to be attributed to the rapidity with which pressures areregistered on the firm's productive capacity-accelerator effects.

There is only one valid criticism that can be made against our formu-lation, viz., that our accelerator variable is in effect —

This induces a "bias" on the accelerator coefficient because of thepresence of the second term in parentheses, a point not made by Resek.This is, however, innocuous in the present context since we are notparticularly interested in the magnitude of that coefficient, nor are weinterested in testing any hypothesis concerning its magnitude—otherthan whether the coefficient is significantly different from zero or not.A similar interpretation is to be placed on the rate of profit coefficient.

Finally, the question of specification raised by Resek concerns chieflyour omission of lagged D/S from the explanatory variables in thedividend equation. In our work with the dividend behavior of electricutilities, cited in our current paper, we showed that in our sample(which was a cross-sectional one) the simple Lintner hypothesis yields•an optimal dividend pay-out ratio of 1.22. Since we refused to enter-tain seriously the notion that firms optimally desire to pay out as divi-dends $1.22 for every dollar of profits, we proceeded to show that analternative specification of the dividend behavior of (electric utility)firms relying on essentially the same structure we use in the currentstudy has as much explanatory power as the Lintner formulation.Moreover, in our specification we gain some additional insight into thestructure of the dividend decision-making process. It is for this reason

Comment 485

that we were not particularly perturbed when multicollinearity forcedus to abandon the version Resek refers to, i.e., a specification in which(D/S) appears as an explanatory variable in the dividend equation.

Perhaps it should be noted that the problem of multicollinearity isgenerally unrelated to the explanatory power of the omitted variablewith respect to the explained variable. Here multicollinearity arises,and severely so, because (D/S) is very highly correlated with

and not because is highly correlated withwhich is what one might infer that Resek had in mind.

It seems fitting to conclude with a summary of the conceptual frame-work of this study. We view the firm as making the three decisionsunder consideration interdependently. The dividend behavior of thefirm depends on the profitability of its operation, the type of activityit engages in, and the decisions it takes or reviews concurrently withrespect to investment and external finance.

Investment depends on elements of the accelerator and rate of profittheories, perhaps with a lag between the appropriation and expenditureof funds on capital projects. Investment also depends on the decisionstaken with respect to dividends and external finance.

Finally, external finance may depend on the interest rate, the leverageposition of the firm, and the decisions with respect to dividends andinvestment.

Our results tend to substantiate this view of interdependence andincidentally show that investment and dividends are relatively littleaffected by external finance, while the latter is chiefly affected by thedecisions of the firm regarding its dividend and investment policies.

Documents

In vestment, Dividend, and External Finance Behavior … · significant dependence of the dividend policy of electric utility firms on ... On the other hand, the rather marked reliance