This PDF is a selection from an out-of-print volume from ... · a prominent place in business cycle theories — received new emphasis as a result of the Keynesian revolution, so

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

Volume Title: The State of Monetary Economics

Volume Author/Editor: Universities-National Bureau Committee for Economic Research

Volume Publisher: NBER

Volume ISBN: 0-87014-307-7

Volume URL: http://www.nber.org/books/univ65-1

Publication Date: 1965

Chapter Title: Money and Business Cycles

Chapter Author: Milton Friedman, Anna J. Schwartz

Chapter URL: http://www.nber.org/chapters/c5179

Chapter pages in book: (p. 32 - 78)

THE subject assigned for this sessioncovers too broad an area to be given even

a fairly cursory treatment in a single paper.Accordingly, we have chosen to concentrate onthe part of it that relates to monetary factorsin fluctuations. We shall still furthernarrow the scope of the paper by interpreting"monetary factors" to mean the role of thestock of money and of changes in the stock —thereby casting the "credit" market as one ofthe supporting players rather than a star per-former — and by interpreting "economic fluc-tuations" to mean business cycles, or evenmore exactly, the reference cycles studied andchronicled by the National Bureau.

The topic so interpreted has been rather outof fashion for the past few decades. Before theGreat Depression, it was widely accepted thatthe business cycle was a monetary phenom-enon, "a dance of the dollar," as Irving Fishergraphically described it in the title of a famousarticle.' Different versions of monetary the-ories of the business cycle abounded, thoughsome of these were really "credit" theories mis-named, since they gave little role to changesin the money stock except as an incident in thealteration of credit conditions; and there wasnothing like agreement on the details of anyone theory. Yet it is probably true that mosteconomists gave the money stock and changesin it an important, if not a central, role inwhatever particular theory of the cycle theywere inclined to accept. That emphasis wasgreatly strengthened by the course of economicevents in the twenties. The high degree of eco-nomic stability then achieved was widely re-garded as a consequence of the effectiveness ofthe monetary policies followed by the only re-cently created Federal Reserve System andhence as evidence that monetary factors wereindeed a central factor in the cycle.

The Great Depression radically changed eco-nomic attitudes. The failure of the Federal

"The Business Cycle Largely a 'Dance of the Dollar,'"Journal of the American Statistical Association, December1923, pp. 1024—1028.

[32]

Reserve System to stem the depression waswidely interpreted — wrongly as we have else-where argued2 and elaborate below — to meanthat monetary factors were not critical, that"real" factors were the key to economic fluc-tuations. Investment — which had always hada prominent place in business cycle theories —received new emphasis as a result of theKeynesian revolution, so much so that PaulSamuelson, in the best selling textbook in thecountry, could assert confidently, "All moderneconomists are agreed that the important fac-tor in causing income and employment tofluctuate is investment." Investment was themotive force, its effects spread through timeand amplified by the "multiplier," and itselfpartly or largely a result of the "accelerator."Money, if it entered at all, played a purelypassive role.

Recently, a revival of interest in money hasbeen sparked less by concern with businesscycles than with concern about inflation. Easymoney policies were accompanied by inflation;and inflation was nowhere stemmed without amore or less deliberate limitation of growth ofthe money stock. But once interest wasaroused, it naturally extended to the cycle aswell as to inflation. In the United States, in-deed, there has been something of a repetitionof the 192 0'S. A high degree of economic sta-bility has been accompanied by a large meas-ure of talk about an active monetary policy,and the monetary authorities have often beengiven credit for playing an important role inpromoting stability. As the experience of thetwenties suggests, this fair-weather source ofsupport for the importance of money is a weakreed.

Examining the present state of our under-standing about the role of money in the busi-ness cycle, we shall first present some factsthat seem reasonably well established aboutthe cyclical behavior of money and related

'See our forthcoming "A Monetary History of the UnitedStates, 1867—1960" (in press, 1963), Chapter 7.

'Economics, 3rd ed., 1955, p. 224.

MONEY AND BUSINESS CYCLESMilton Friedman and Anna J. Schwartz

University of Chicago and National Bureau of Economic Research

MONEY AND BUSINESS CYCLES 33

CHART I. — MONEY STOCK INCLUDING COMMERCIAL BANK TIME DEposiTs, 1867—1960, AND CURRENCY PLUSDEMAND DEPOSITS ADJUSTED, 1914—1960

— Currency held by the public, plus. demand depositsplus commercial bank time deposits

Currency held by the public, plus demand deposits adjusted

Rilions ot dollars

: :

: : :

.. Ii I II 11111 II 1 1.1 1 1

1867 '68 '69 '70 '7 1'7a'73'74'75'76'77'78'79'eo'el'8z'e3'84'su'ss'ul'88'89'9o'91 '92 '93 '94 '95 '96 '97 'sa.

40

30

20

109

6

—

-'-p-sf

-/-

.........._I....__.._. liii II 111111 j II I II .1 II I

1898 '99 '0l'02'03'04'05'06'07'08'09'10'1l'lZ'13'14'15'16'17'18'19'20'21'2a'23'24'25'26'27'28'29

SOURCE: Friedman and Schwartz, "A Monetary History of the United States, 1867—1960" (in press, 1963), Table A—a, cols. 7 and 8. Theseare seasonally adjusted figures, dated as of end of month, 1867—1946; for 0947—60, currency plus demand deposits adjusted is an averageof daily figures, and commercial bank time deposits, a 2-month moving average of last-Wednesday-of-month figures, for a month centeredat midmonth.

34 MONEY AND BUSINESS CYCLES

magnitudes and then speculate about someplausible interpretations of these facts. Thefacts we present are drawn largely from ourown unpublished work done under the auspicesof the National Bureau of Economic Researchand associated unpublished work by PhillipCagan.

Some Facts About the Cyclical Behavior of Money

Cyclical pattern of the money stock.The outstanding cyclical fact about the stock

of money is that it has tended to rise duringboth cyclical expansions and cyclical contrac-tions. This is clear from Chart i, which plots(i) the stock of money from 1867 to 1960,with money defined as including currency plusadjusted deposits in commercial banks (bothdemand and time) held by the nonbankingpublic (i.e., excluding both balances of thefederal government and of banks); and (2)from 1914 on, a narrower total which excludestime deposits. From 1867 to 1907, our dataare at annual or semiannual dates; from 1907on, monthly. The only major exceptions since

to the tendency of the money stock to riseduring both cyclical expansions and cyclicalcontractions occurred in the years listed in thefollowing tabulation, which gives also the per-centage decline during each exception.

PercentageDecline

4.9

1920—2 I1929—33 35.21937—38 2.4

In addition, there were two minor exceptionssince the end of World War II,

1948—491959—60 I.'

The major exceptions clearly did not fall in arandom subset of years. Each correspondswith an economic contraction that was majoras judged by other indicators; in the periodcovered, there was no other economic contrac-tion more severe than any in the list; and thereappears to be a considerable gap between theseverity of those contractions and of the re-

mainder, with the possible exception of thecontraction of 1882—85 which might be re-garded as a somewhat borderline case.

For mild depression cycles, therefore, thecycle does not show up as a rise and a fall.Chart 2 gives the average reference-cycle pat-

CHART 2.— MONEY STOCK: AVERAGE REFE1tENcE-CY-CLE PATTERNS FOR MILD AND DEEP DEPInssIoN CYCLES,

1867—1961

—1867—1908————1908—1961

terns for mild and deep depression cycles since1867, excluding only war cycles. (Patternsare given separately for the period before andafter 1907, because the availability of monthlydata after 1907 permits the construction of amore detailed pattern — a nine-point insteadof a five-point pattern.) The patterns for milddepression cycles rise almost in a straight line,though there is some indication of a slowerrate of growth from mid-expansion to mid-con-traction than during the rest of the cycle (espe-cially in the nine-point pattern for monthlydata). In its cyclical behavior, the moneystock is like other series with a sharp upward

Referencecycle relatives110— Deep

depression cycles

110—cycles

Years ofException1873—791892—941907—08

—30 —20 —10 0 10 20 30Months from reference peak

3.75.'

Nota: War cycles, not shown, are 1914—19 and 0938—45. Deep depres-sion cycles are 0870—79, 1891—94, 0904—08, 1909—a!, 0927—33 and0933—38. All others are mild depression cycles.

Souaca: For method of deriving reference cycle relatives for the 9-point pattern, see A. F. Burns and W. C. Mitchell, MeasuringBusiness Cycles, National Bureau of Economic Research, 1946,pp. 160-170; we used a variant of the National Bureau's standardtechnique for annual series (pp. 297—202) for the 5.point pattern.

1.4

trend — such as population, the total stockof houses, the number of miles of railroadtrack in operation in the pre-1914 period, theamount of electrical energy produced. In allof these, the cycle shows up not in an absoluterise and fall but in different rates of rise.

For deep depression cycles, the cyclical pat-tern is nearer the stereotype of a rise duringexpansion and a fall during contraction. From

35

these patterns, it would be easy to concludethat the two groups of cycles distinguished aremembers of different species with respect tothe behavior of the stock of money.

Cyclical pattern of the rate of change in themoney Stock.

Because the strong upward trend of thestock of money tends to dominate its cyclical

CHART 3. —MONTH-TO-MONTH RATE OF CHANGE IN U.S. MONEY STOCK, 1867—1960III I I I I I I I

4J_.J_ I I I ....J ...L...

NOTE: Solid vertical lines represent reference cycle troughs; broken lines, peaks. Dots represent peaks and troughs of specific cycles, Thehorizontal broken lines represent high and low steps in the rate of change.

SOURCE: In the annual or semiannual segment, 5867—1907, the change in natural logarithm from one date to the next in the data underlyingChart a was divided by the number of months intervening, and the quotient plotted at the middle of the month halfway between. Intht monthly aegment. 1907—60, the month-to-month change in natural logarithm was plotted in the middle of the second month. Ref-erence dates are from the National Bureau (see Table a).

MONEY AND BUSINESS CYCLES

Per cent

2

—l

—2

867 '68 '69 '70 '71'7a'73'74'75'76'77'78'79'80'81'82'83'84'85'86 87 '88 '89 '9 '91 '92 ''93 '94 '95 '96 '97 '98


behavior, it is desirable to eliminate the effectof the trend in order to reveal the cyclical be-havior more clearly. There are various waysof doing this.4 The method we have used is totake logarithmic first differences of the moneystock, which is equivalent to using the per-centage rate of change from one time unit tothe next. Chart 3 plots the resulting series.It is clear that this device effectively eliminatestrend. It is clear also that, as first differencingusually does, it produces a highly jagged serieswith a sawtooth appearance. The reason isthat independent errors of measurement in theoriginal stock series introduce negative serialcorrelation into first differences. But despitethese short-term irregularities, the series showsclearly marked cyclical fluctuations corre-sponding to reference cycles.

Chart 4 gives the reference cycle patternsfor this series. They show a clear cyclical pat-tern with the mild and deep depression cyclesdistinguished, this time primarily by theiramplitude, so that they now look more likedifferent members of the same species. Thepeak rate of change occurs early in expansionand the trough early in recession. Indeed theseoccur so early as to suggest the possibility ofinterpreting the rate of change series as in-verted, i.e., as generally declining during ref-erence expansion and rising during referencecontraction. We have examined this possibilityelsewhere.5 A full presentation of our tests is

See the discussion of this problem in Milton Friedman,"The Lag in Effect of Monetary Policy," Journal 0/ Pout.ical Economy, October 1961, pp. 453—454.

See our forthcoming "Trends and Cycles in the Stockof Money in the United States, 1867—1960."

The patterns in Chart 4 differ in construction from thereference patterns for the stock of money in Chart 2. Therate of change series, being the percentage change frommonth to month, is already in a form that is independent ofunits of measure. In addition, the rate of change in themoney stock can be zero or negative as well as positive, andhence its average value for a given cycle can hardly serve asa base for computing reference cycle relatives. For thesereasons, the basic data, instead of being expressed as rela-tives to the average for a cycle, are expressed as deviationsfrom the average for a cycle (as in A. F. Burns and W. C.Mitchell, Measuring Business Cycles, New York, NBER,1946, pp. 137—138). This is why the base lines in Chart 4are labeled o instead of zoo as in Chart 2, and the scale isin terms of deviations rather than of relatives.

Because of a discontinuity in the underlying money figuresin early 1933, we have estimated stage IX for the 1927—33cycle and stage I for the 1933—38 cycle from the average

not feasible in this paper; it will suffice to notethat they rather decisively support treating therate of change series as conforming to the ref-erence cycle positively with a long lead, rather

CHART 4. — RATE OF CHANGE IN MONEY STOCK: AVER-AGE REFERENCE-CYCLE PATTERNS FOR MILD AND DEEP

DEPRESSION CYCLES, 1867—1961

— 1867—1908

————1908—1961

NOTE: War cycles, net shown, are 5954—59 and 5938—45. Deep depres..sion cycles are 5870—79, 5895—94, 5904—08, 1919—21, 5927—33,and 1933—38. All others are mild depression cycles.

Sousca: See footnote

than inversely with a somewhat shorter lag.Though we have not analyzed in as much detailthe narrower total of currency plus adjusted de-mand deposits, its cyclical pattern since 1914is very similar in general form to the patternof the broader total.

value for January, April, and May, 1933, instead of forFebruary, March, and April. Restricted deposits before thebanking holiday are counted in full in the recorded moneystock. However, after the holiday both restricted and un-restricted deposits in unlicensed banks are excluded com-pletely from the recorded money stock. That shift intreatment is the major factor behind the sharp decline Inthe recorded figures in March 1933 (see our "A MonetaryHistory of the United States, 1867—1960," Chapter 8,section I).

Deep depression cycles

Per cent+1.0 —

+0.5—

—0.5—

—1.0—

+0.5—

—0.5—

Mild depression cycles

Months from reference peek20 30

Cyclical timing of the rate of change in themoney stock.

Evidence on cyclical timing derived from acomparison of turning points is clearly notavailable from the stock of money series, be-cause it has so few turning points. For therate-of-change series, we have dated turningpoints in two ways: (i) We have sought toapproximate the series by a step function, withsuccessively high and iow steps, because attimes the series gives the impression of drop-

37

a low step ends, the date of a step trough. (2)We have applied the usual National Bureauspecific cycle dating procedure to the rate-of-change series, and have designated specificcycle peaks and troughs. They are marked byblack dots in Chart 3.

Table i gives the step and specific cyclepeaks and troughs we have selected, the datesof the reference cycle turns with which wehave matched them, and the indicated lead(—) or lag (+) at the corresponding turn.6

TABLE I. — TIMING OF SPECIFIC CYCLE AND OF STEP TROUGHS AND PEAKS IN THE RATE OP CHANGEIN THE MONEY STOCK COMPARED WITH TIMING OF BUSINESS CYCLES

TROUGHS PEAKS

Date of:Lead(—) or Lag(-I-) in

Months at ReferenceTrough of:

Step Specific StepTrough Cycle Trough Peak

Date of:Lead(—) or Lag(±) in

Months at ReferencePeak of:

Step SpecificPeak Cycle Peak

Step'Trough

SpecificCycle

Trough

MatchedReferenceTrough

SpecificCyclePeak

MatchedReference

Peak

2/796/856/88

6/9!

6/93

6/g6

6/oo6/04

5/7712/8322/8712/9012/9212/9512/99

12/03

3/795/854/88

5/926/946/97

12/008/04

SEMIANNUAL AND ANNUAL2/72

—I —22 8/8'+2 —27 6/87+2 —4 6/90+1 —5 6/92—12 —i8 6/95

—12 —i8 6/99

—6 —12 6/ol

—2 —8 6/07

DATA7/715/81

12/8522/8912/9112/94

12/9812/0012/04

10/733/823/877/901/93

12/95

6/99

9/025/07

—20 —27—7 —10+3 —15—I —7—7 —13—6 —220 —6

—15 —2!+1 —29

2/088/no7/23

12/145/i87/22

3/2422/264/335/38

i/so4/54i/586/6o

i/o8

4/20

6/13

5/181/216/23

22/2610/3110/3710/41

9/5312/5722/59

6/o8

2/22

12/14

3/19

7/217/24

11/273/336/38

20/4510/49

4/582/61

MONTHLY DATA—4 —5 6/o9

—27 —21 6/22

5/140 —28 7/17

—20 —20 3/200 —6 5/23

—4 —13 9/25—II —II 4/28+1 —17 7/36—i —8

20/45

+3 —9 22/52—4 —II 9/55—3 —4 5/59—8 —14

20/08io/ii22/26

22/184/227/24

11/274/362/416/43

11/512/556/58

1/10

1/13

8/x8

1/205/23

20/268/295/37

2/4521/487/537/575/6o

—7 —'5—7 —15

—23 —20+2 —13

0 —23—23 —27—i6 —21—20 —23

+8 —20

—7 —20—22 —29—12 —23

SOURCE: Chart 3. Step peaks and step troughs are last months of alternate steps shown there.Reference dates through Apr, 1938 are shown in Business Cycle Indicalors, Geoffrey H. Moore, ed., Princeton University Press for NBER,

1961, Vol. p. 670; subsequent dates are from an unpublished National Bureau table. For timing comparisons, both the rate of changeseries and th'e steps made from it are treated as well conforming, because of the nearly i-to-I correspondence between their turning pointsand reference cycle turning points and because the money stock series from which both were derived has moderately high conformity in-dexes (too for expansions, ior contractions, +7I for trough-to-trough full cycles, + co for peak-to-peak full cycles, +6, for full cyclesboth ways). Matching of step and specific cycle turns with reference turns follows Burns anJ Mitchell, Measuring Bussness Cycles, pp. 115—128.Earlier versions of this table were based on data now superseded.

ping suddenly from one level to a decidedly our money series starts in 1867, the first refer-ence turn with which we have matched a specific cycle turnlower level, or of rising from one level to a is the peak in October 2873. Hence we do not match the

decidedly higher level. The horizontal broken reference trough of December 1867, peak of June 2869, and

lines in Chart 3 indicate the steps we have trough of December 1870. The absence of a specific cycleturn to match with the December 2867 trough may simplyused. 'We call the date at which a high step result from the fact that our series does not go far enough

ends, the date of a step peak, the date at which back in time — a possibility suggested by the long average



Clearly, leads predominate, and clearly also,there is much variability.

Table 2 gives the average lead and the stand-ard deviations of the leads for mild depressioncycles, deep depression cycles, all nonwarcycles and all cycles, for both step dates andspecific cycle dates. For step dates, the aver-

age lead for all cycles is 7 months at the peakand 4 months at the trough; for specific cycledates, the average lead is i8 months at thepeak and 12 months at the trough; for stepdates, the standard deviation of the lead is 6months at troughs and 8 months at peaks; forspecific cycle dates, the standard deviation ofthe lead is 6 months at troughs and 7 monthsat peaks.

Estimation of timing relations by a com-parison of turning points seems inefficient,because it uses so little of the information con-tained in the series. Therefore, we have experi-mented extensively with other devices, in par-

lead at troughs. For the other two reference turns, we con-jecture that the annual data for successive Januarys — allwe have for that period — may conceal by their crudenessturns that monthly data would reveal. This conjectureseems especially plausible because of the unusual brevity ofthe expansion phase, only i8 months, followed by a con-traction of equal length.

ticular, cross-correlograms and cross-spectralanalysis. While these devices, particularlycross-spectral analysis, offer great promise forthe future, as yet we have no substantive re-suits worth reporting.

We have tested to determine whether thereis any secular trend in the leads or lags; whether

the pre-1914 timing, before the establishmentof the Federal Reserve System, differs fromthe post-1914 timing; whether timing duringmild depression cycles differs from timing dur-ing deep depression cycles; and whether thereis any relation between the length of the leadand the amplitude of the subsequent or priorcyclical phase. Our results so far are negative:none of these criteria appears to be associatedwith a statistically significant difference in tim-ing.

Amplitude of movements in the rate of changein the money stock.

i. The subdivision between mild and severedepression cycles in Chart 4 corresponds to asharp difference in the amplitude of referencecycles in the rate of change. This result sug-gests that the amplitude of the changes in therate of change in the money stock is related to

TABLE 2. — AVERAGE TIMING OF SPECIFIC CYCLE AND OF STEP PEAKS AND TROUGHS IN THE RATE OF CHANGEIN THE MONEY STOCK AND STANDARD DEVIATION OF LEAD OR LAG, BY PERIOD AND TYPE OF CYCLE

NUMBER OF OBSERVATIONS MEAN LEAD (—) OR LAG (+) IN MONTHSSTANDARD DEVIATION OFLEAD OR LAG IN MONTHS

Step AnalysisSpecific Cycle

AnalysisSpecific Cycle

Step Analysis AnalysisSpecific Cycle

Step Analysis Analysis

Period Trough Peak Trough Peak Trough Peak Trough Peak Trough Peak Trough Peak

ALL CYCLES1870—1908 8 9 8 9 —3.6 —5.8 —13.0 —15.6 5.7 7.7 6.7 8.31908—1960 13 12 13 12 —4.5 —8.1 —11.3 —19.1 5.7 8.3 5.2 5.51870—1960 21 21 21 21 —4.1 —7.1 —12.0 —17.6 5.6 7.9 5.7 6.9

WAR CYCLES1908—1960 I 2 I 2 —20.0 —2.5 —10.0 —20.0 — 14.8 — o

DEEP DEPRESSION CYCLES1870—1908 2 3 2 3 —6.5 —8.7 —20.0 —23.0 7.8 io.6 2.8 8.71908—1960 4 3 4 3 —i.o —8.0 —9.0 —15.7 2.2 9.2 5.5 4.6

1870—1960 6 6 6 6 —2.8 —8.3 —12.7 —19.3 4.8 8.9 7.2 7.4

MILD DEPRESSION CYCLES1870-1908 6 6 6 6 —2.7 —4.3 —10.7 —11.8 5.4 64 6.o 5.61908—1960 8 7 8 7 —5.5 —9.7 —12.6 —20.3 6.3 6.9 5.3 6.3

1870—1960 14 23 14 23 —4.3 —7.2 —ii.8 —16.4 5.9 7.0 5.5 7.2

SouRce: Table s. To avoid duplication, each cycle is represented oniy by its peak and terminaldepression cycles are grouped as in Chart 2.

trough. War, deep depression, and mild

RANK DIFFERENCE CORRELATIONOF AMPLITUDES

NBER ReferenceNBER Reference Full Cycle

Specific Cycles in Rate of Trough- Peak-Change in Money Stock Expan. Contrac- to- to.

Correlated with: sian tion Trough Peak

Annual and semiannualdata 1879—1907 1882—1908

(8 pairs) (7 pairs)Reference cycles

in clearings-debits .36 .64 .43 .68Specific cycles in

Moore index .76 .85 .76 .79Monthly data 1907—1960 1908—1960

('0 pairs) (io pairs)Reference cycles


Moore index .82 .58 .75 .8rWhole-period data 1879—1961 1882—1960

('8 pairs) (17 pairs)Reference cydes


Moore index .70 .78 .77

the severity of cyclical movements in generalbusiness, even though the timing of the changesin the rate of change in the money stock is not.

39

2. One way in which we have investigatedthis relation further is to correlate the rankingof the amplitudes of cyclical movements in therate of change with the ranking of the ampli-tudes of the corresponding cyclical movementsin general business, as measured by two dif-ferent indicators: one, bank clearings to 1919and bank debits thereafter; the other, an indexcomputed by Geoffrey H. Moore. The cor-relations, summarized in Table 3, are through-out positive for expansions alone, for con-tractions alone, and for full cycles, for theperiod before 1908 and for the period since, aswell as for the whole period.

The correlations between the rate of changemeasure and the Moore index are sufficientlyhigh so that, even with the small number ofobservations on which they are based, theycould hardly have arisen from chance. Thereis a less close connection between the clear-ings-debits figures and the rate of change, espe-cially in expansions. The Moore index is ad-justed for trend and reflects primarily changesin physical units. Likewise, the shift from thetotal stock of money to the rate of change is, asnoted earlier, equivalent to adjusting for trend;in addition, it involves a change from a meas-ure expressed in nominal units — dollars — toa measure expressed in relative units — percent — and as a flow — per month. The ampli-tude of clearings-debits, however, is not ad-justed for intracycle trend, and clearings-debits are, in their original form, in dollars.It would be interesting to know whether theadjustment for trend, or the different weightgiven to financial and physical transactions, isprimarily responsible for the closer connectionof the Moore index than of clearings-debits tothe rate of change.

The table as a whole leaves little doubtthat there is a fairly close connection betweenthe magnitude of monetary changes during thecourse of cycles, and the magnitude of theassociated cyclical movement in business. Therelation is by no means perfect for the meas-ures we use. But we have no way of know-ing from this evidence alone to what extent thediscrepancies reflect the inadequacies of ourindexes of economic change, the statistical er-rors in our money series, or a basic lack of con-

MONEY AND BUSINESS CYCLESTABLE 3. — RANK DIFFERENCE CORRELATION BETWEEN

CHANGE IN RATE OF CHANGE IN MONEY STOCKAND CHANGE IN Two INDICATORS OF GENERALBUSINESS, 1879—1961, EXCLUDING WAR CYCLESAND 1945—49

NOTE: In our full study we have used three measures of theamplitude of the change in money, each both in total and as a rateper month, measuring the change in cycle relatives between referencedates, between step dates, and between specific cycle peaks andtroughs in the rate of change. To simplify our presentation here, werestrict the comparison to the total change in amplitude betweenpeaks and troughs in the rate of change.

•War cycles 1914—19 and 1938—45 are omitted because of theirspecial characteristics. The 1945—49 cycle is omitted because the ex-pansion is skipped by the rate of change series (see Table i). Notied ranks correction is used in getting correlation coefficients. 'Ampli.tude" of i-ste of change in money stock is expressed in units of the data asplotted in Chart above. For expansions, it is the change in stagesI—V of the specific cycle; for contractions, the change in stages V—IXof the specific cycle. For clearings-debits the reference cycle ampli-tude (stages I—V—IX), expressed in reference.cycle relatives, wasused. For the Moore index, specific cycle amplitudes only are avail-able, but they have a one-to-one correspondence with reference cycles.For fsll cycles, trough-to-trough the change from V to IX was sub-tracted from the change from 1 to V to obtain the total rise andfall used in the correlations; for full cycles, peak-to-peak thechange from I to V was subtracted from the change from V to IX.

SOURCE: Rate of change in money stock: Figures underlyingChart were analyzed for specific cycles, as in Burns and Mitchell,Measuring Business Cycles, pp. 115—141; matching of peakstroughs with reference turns follows Table x -

Clearings.debits: Bank clearings outside New York City, monthly,1879—1919; bank debits outside New York City, monthly, 1919—61.1879—1942: Seasonally adjusted from Historical Statistics of theUnited States, 1789—1945, Bureau of the Census, 1949, pp. 324—325,337—388. 1943—61: Board of Governors of the Federal Reserve Sys-tem, Division of Bank Operations, mimeographed table, "Bank Debitsand Rates of Turnover" (C. Revised Series, 1943—52), December23, 1953; thereafter Federal Reserve Bulletin, adjusted. for seasonalvariation by NBER. Reference cycle analysis follows Burns andMitchell, op. cit., pp. 560—170.

Moore index: Unpublished memorandum by Geoffrey H. Moore,extending table in ibid., p. 403, and revising and updating table inBusiness Cycle Indicators, G. H. Moore, ed, Vol. I, p. 104. Anaverage of three trend adjusted indexes of business activity — A. T. &T., Persons-Barrono, and Ayres — each of which was analyzed forspecific cycles, suppressing specific cycle turns not corresponding toreference cycle turns.


nection between monetary and economicchanges.

3. To get further evidence, we have investi-gated this relation in a different way using an-nual data. For the period from 1869 to 1960,we have annual estimates of net national prod-uct, and also, of course, annual estimates of thestock of money. For this period, we have com-puted logarithmic first differences (i.e., year-to-year percentage changes) of both series. Wehave then computed moving standard devia-tions (comparable to moving averages) fromthese rates of change involving 3, 4, 5, and 6terms. To illustrate: for the 3-term mov-ing standard deviation, we took the initial threerates of change (1869—70, 1870—71, 1871—72),

computed their standard deviation by the usual

statistical formula,7 and dated the result as of187o—71; then dropped the initial year andadded a year, computed the standard devia-tion for the resulting triplet of rates of change(1870—71, 1871—72, 1872—73), and dated the

results as of 1871—72; and so on.These moving standard deviations are a meas-

ure of the variability of the rates of change —in the one case, of money; in the other case, ofincome. If such a computation were made for

a strictly periodic series, say, a sine wave of

fixed period and fixed amplitude, and if the

length of the moving standard deviation were

the same as the period of the sine wave (or

an integral multiple of it), then the computed

moving standard deviation would be constant

over time, and its value would be equal to

times the amplitude of the sine wave.8 If

the length of the moving standard deviation

were shorter than the period of the sine wave,

the computed moving standard deviation would

fluctuate over time, its value never exceeding

the value just cited. The same proposition holds

if the length of the moving standard deviation is

longer than the period of the sine wave but not

an integral multiple of it, though it is perhaps

— 2

'That is, estimate of s. d. = , where xfl—I

is the observation, the mean, and is the number of itemsin the group, in this example, 3.

8 Let the sine wave be where t is time. Thenin is the period of the wave and A the amplitude, the wavefluctuating from +A to —A.

obvious that, as the moving standard devia-

tion is lengthened, the standard deviation will

approach the constant value noted above, since

the fractional cycle becomes less and less im-

portant compared to the whole cycles includedin the computation of the standard deviation.

It follows from these considerations that, for

our purpose, which is to see how the amplitude

of the cycles in the rate of change in the money

stock is related to the amplitude of business

cycles, we want to use a number of terms equal

to the length of the cycle in which we are inter-

ested. This explains why we have used 3, 4, 5,and 6 terms, for the reference cycle since 1867has averaged four years in length but has oc-casionally been shorter or longer. As it hap-pens, the results are not very different for dif-ferent numbers of terms, so we present a chartfor only the 4-term results, though we givesome numerical data for all.

One more point before turning to the results.

Net national product, which we are using as

an index of general business and whose fluctua-

tions we are interpreting as a measure of the

amplitude of business cycles, has a sharp up-

ward trend, though a less steep one than themoney stock has, so that it typically declines

absolutely during contractions. If we were to

take a moving standard deviation of its ab-

solute values, or their logarithms, the resultwould overestimate cyclical variability becauseof the intracycle trend, and the overestimate

would vary over time as the intracycle trend

did. Accordingly, to eliminate the effect of the

intracycle trend from our measure of vari-

ability, we have used logarithmic first differ-ences for net national product as well. Thisprocedure is of the same class and for the samepurpose as the National Bureau's standard

technique of estimating full cycle amplitudes

by subtracting the change during contraction

from the change during expansion. However,

the use of first differences can also be taken to

mean that what we are calling the amplitude

of business cycles refers to a construct rather

different from the National Bureau's standard

reference cycle; it refers to a cycle in the rate

of change in aggregates rather than in the level

of aggregates. As is well known, for a sine

wave, the rate of change series has the same

amplitude and pattern as the original series but


differs in phase, its peaks and troughs comingone-quarter of a cycle earlier or three-quartersof a cycle later than the peaks and troughs ofthe original series.

Aside from removing the effect of intracycletrend, another advantage of using the first dif-ferences of net national product is that the re-sults would be almost identical for total netnational product and net national product percapita. Since population has grown at a steadyrate over periods of 3 to 6 years, the use of percapita data would affect only the movingaverage of the rates of change but not the mov-ing standard deviation.

Chart 5 plots the 4-term moving standarddeviations for money and net national prod-uct. It should be noted that since we haveused natural logarithms, the vertical scale canbe interpreted directly in terms of percentagepoints. For example, a value of .ioo meansthat the standard deviation is equal to an an-nual rate of growth of io percentage points.9

°Let p(t) be the continuous rate of growth from yearto year t + i, so that

Xs+i = eP(t),

The scale on the chart is logarithmic. Thereason is that, since the standard error of theestimated standard deviation is proportionalto the (true) standard deviation, the standarderror of the logarithm of the standard devia-tion is roughly a constant, regardless of thesize of the (true) standard deviation. Hencethe logarithmic scale makes sampling fluctua-tions appear the same size throughout.

It is clear from the chart that there is aclose relation between the variability of moneyand of net national product: the two curvesparallel one another with a high degree offidelity, especially when it is borne in mindthat standard deviations based on only fourobservations (three degrees of freedom) are

where X, and X,÷1 are successive annual observations.Then log6X,+1 — log6X5 = p(t).Note also that

log6X,+i — = log0 (i +But log0(r + k) is approximately equal to k for small k.Hence the first difference is approximately equal to

xt+I - xS

CHART 5. — MOVING STANDARD DEVIATION OF ANNUAL RATES OP CHANGE IN MONEY, 1869—1958, AND IN INCOME,1871—1958, 4-TERM SERIES

SouRce: Money figures, described in source for Chart x, are annual averages centered on June 30. Income figures are annual estimates ofnet national product, beginning 1869, from worksheets underlying Simon Kuznets, Capital in the American Economy: Its Formation andFinancing, Princeton for NBER, 1961. For computation of moving standard deviation, see subsection of this section, and footnote 7.


subject to a good deal of sampling variation,10that the net national product and money seriesare, so far as we know, wholly independent intheir statistical construction, and that both aresubject to an appreciable margin of error.

At first glance, it appears from Chart 5 thatincome has become more variable relative tomoney over the period covered. Unless weare mistaken, this is a statistical artifact. Acloser look at the chart will show that thechange comes shortly after the turn of thecentury. Before 1900, the standard deviationsfor money and for net national product areroughly equal in magnitude; subsequent tothat date, the standard deviations for net na-tional product are noticeably higher than formoney. The reason, we conjecture, is thechanging statistical character of the net na-tional product estimates, in particular, the roleplayed in them by interpolation between decen-nial census years. The effect of interpolationis to smooth greatly the year-to-year changesand so to reduce the estimated standard devia-tions. For the estimates before 1889, inter-polation played a major role; for those from1919 on, a much smaller role.11 For the inter-mediate decades, the role of interpolation rela-tive to independent data for individual yearsbecame successively smaller. We cannot findany clear indication in the description of thestatistical series that there was a sharp breakaround 1900 in the role of interpolation. How-ever, the data behave as if there were such abreak. For the period before 1900, we con-jecture that the standard deviations appreci-ably understate the variability of income. Forthe subsequent period, it is much harder tomake a comparable judgment. The statisticalerrors of estimation tend to raise the computedstandard deviation; interpolation tends tolower it.

For money, the degree of interpolation inthe annual estimates is small throughout(interpolation plays a much larger role in our

'°A more precise statement for these data is hard toarrive at, since successive first differences are not statisticallyindependent.

See Simon Kuznets, National Product Since 1869, NewYork, NBER, 1946, pp. 90 if. These considerations havethe obvious implication that net national product estimatesare untrustworthy as a source of evidence on secularchanges in the amplitude of business cycles.

monthly estimates). Hence the standard de-viations for money are probably overestimatesof the "true" standard deviations, thanks tothe errors of estimation. However, because ofthe character of the basic data, such errors areprobably appreciably smaller than for net na-tional product.

Aside from the shift in the level of the stand-ard deviations for net national product, themost striking feature of the chart is what ap-pear to be fairly regular cyclical fluctuations,of about 8 to '5 years in length, in the stand-ard deviations of both money and net nationalproduct; these are the counterparts of the longswings that have received much attention.However, a warning is in order about any suchinterpretation of these results. The movingstandard deviations for successive years arehighly correlated because they have three outof four items in common. As is well known,a moving average applied to a series of randomterms will produce a series that seems to movesystematically; and the moving standard de-viation is a moving average and so has the sameeffect. For our purposes, what is important isthe parallelism of the two series plotted inChart 5, not the character of their commonfluctuations.

Table 4 presents numerical evidence for allfour lengths of moving standard deviations wehave computed. Because of the break in thenet national product data, the results are givenseparately for the period before and after1899. We used 1899 as the dividing point be-cause it is a census year. The results for theseparate periods are more meaningful than theresults for the period as a whole.

This table reinforces the visual evidence ofChart 5 and adds to it a number of importantpoints. One is that the correlation is generallyhighest when the standard deviations are com-pared synchronously; it is generally loweredif standard deviations for money are comparedwith either later or earlier standard deviationsfor NNP though, for the earlier period, thecorrelation is highest when money leads oneyear for three of the four lengths of movingstandard deviations. If there be any lead orlag for the later period, it is presumably lessthan a year in length. The slightly higher cor-relations for the later period for NNP leading


by a year than for money leading by a yearmay reflect a lead of NNP by a fraction of ayear. A second point added by the table isthat the standard deviation for net nationalproduct for the period after 1899 is roughlydouble the standard deviation for money.12 Asa first approximation, therefore, the amplitudeof cyclical fluctuations in income is twice thatin money.

The correlations rise steadily as the numberof terms in the moving standard deviations isincreased. The rise presumably reflects thesmoothing of the standard deviations intro-duced by the larger number of degrees of free-dom and hence the reduction in the role ofchance fluctuations. Calculations not sum-marized in the table indicate that the peaksynchronous correlation is reached for seventerms. The fact that the mean standard de-viations rise is less easily explained, sincethese should average the largest for a periodequal to the average length of a cycle. The

explanation is presumably the existence of thelonger waves. We conjecture that the meanstandard deviation would continue to rise asterms are added and reach a maximum at some-thing like io to terms.

To summarize these results: They stronglyreinforce the evidence from the earlier com-parison of reference cycle amplitudes. Thereis unquestionably a close relation between thevariability of the stock of money and the vari-ability of income. This relation has persistedover some nine decades and appears no differ-ent at the end of that period than at the be-ginning, if allowance is made for the changingcharacteristics of the statistical raw materials.

Cyclical behavior of velocity.i. The. ratio of income to the stock of

money, which is to say, the income velocity ofmoney, has been rising in the post-World WarII period. However, over the whole of themore than nine decades our data cover, it hasdeclined sharply, from 4.6 at the outset of theperiod to 1.7 at the end. As a result, velocityhas frequently declined during both expansionsand contractions in general business. Whenthat has not been the case, velocity has con-formed positively to the cycle, rising duringexpansions and falling during contractions.When it has been, the cyclical effect has shownup in a slower rate of decline in expansion thanin contraction. The average cyclical patterns

TABLE 4. — MOVING STANDARD DEVIATIONS OF ANNUAL RATES OF CHANGE IN MONEY AND NET NATIONAL PROD-UCT: MEANS, STANDARD DEVIATIONS, AND CORRELATION COEFFICIENTS FOR DIFFERENT NUMBERS OF TERMS

Period

Number ofTerms inMovingStandard

Deviations

Mean Standard Standard Deviation ofDeviation Standard Deviation

(natural logarithms)

CORRELATION COEFFICIENT BETWEEN STANDARDDEVIATIONS OF MONEY AND NNP

Money Leading NNP by: NNP Leading Money by:(yearn) (years)

Synchro.3 2 a noun I 2 3M NNP M NNP

0869—0898

456

.022 .039.052 .067 .023 .033

.054 .068 .022 .029

.057 .069 .021 .027

293 .535 .6i6 .476 .114 — .011 —.099

.364 .648 .718 .540 .263 — .049 — .163

.378 .672 .707 .657 .430 .044 —.252

.398 .583 .755 .759 .543 .044 — .o6g

1899—1960

456

.039 .o8o .028 .048

.089 .029 .046

.048 .029 .046

.051 .100 .029 .044

.003 .113 .345 .670 .589 .248 .036

.135 .243 .456 .814 .722 .472 .263

.216 .385 .608 .840 .820 .637 435

.272 .480 .672 .870 .840 .707 .5i8

0869—0960 3456

.042 .076 .027 .046

.047 .o8x .027 .044

.o5o .085 .027 .043

.053 .o8g .027 .042

.001 .141 .349 .591 .429 .049 — .026.011 .242 .425 .687 .56i .302 .033

.172 .348 .534 .721 .665 .465 .266

.220 .404 .58i .748 .69o .536 .360

Soueca: Same as for Chart

The same reason that recommends a logarithmic scalefor Chart also suggests an advantage in making compu-tations like those in Table 4 from the logarithms of themoving standard deviations. We have done so for the periodfrom 0899 through 1960. The correlation results are quitesimilar. The synchronous results are .6oo, .797, .83 7, and.88o for 4, and 6 periods respectively.

The ratio of the geometric mean of the standard deviationof NNP to the geometric mean of the standard deviation ofmoney is 2.30, 2.25, 2.29, 2.13, for 3, 4, 5, and 6 periodsrespectively. This method of estimation therefore suggeststhat income is roughly 21/4 times as variable as money.

portant role than it assigns to the discrepancybetween measured and permanent concepts.

3. This interpretation has been criticized asassigning much too small a role to interestrates. Henry A. Latané, in particular, hasargued that the whole of the movement ofvelocity, both over longer periods and over thecycle, can be accounted for by changes in inter-est rates, higher interest rates leading to econ-omy in the use of money and so to highervelocities, and conversely.14 His analysis coversa shorter period than ours does (1909—58).

4. There is no necessary contradiction be-tween these two interpretations, the appearanceof contradiction arising primarily from ourdefinition of money as the sum of currencyplus all adjusted deposits in commercial banks,

Mild depression cycles and Latané's definition of money as the sum ofcurrency plus adjusted demand deposits alone.

a. Time deposits in commercial banks ap-pear to have a substantially higher incomeelasticity of demand than currency or demand

I

deposits have, so that the income elasticity of20 30 money by our use of the term is doubtless

higher than it is by Latané's use of the term.This can explain why we find it necessary tointroduce an income effect to explain the sec-ular decline in velocity, while he does not. Toput this point differently, we find that theelasticity of demand for (real) money balanceswith respect to permanent income is about 1.8when money is defined as we define it. This isconsistent with a corresponding elasticity notmuch different from unity for Latané's nar-rower definition, provided the elasticity for timedeposits is between 2.5 and 3.5.15 Furthermore,since there is a considerable trend element inthe movement of interest rates over the periodLatané's analysis covers — as, of course, there


of velocity, for mild depression and deep de-pression cycles (excluding war cycles), aregiven in Chart 6.

CHART 6. — INCOME VELOCITY: AVERAGE REFERENCECYCLE PATTERNS FOR MILD AND DEEP DEPRESSION

CYCLES, 1870—1958

Referencecycle reletives

depression cycles

90—

110—

100

90 —

—30 —20 —10 0 tOMonths from reference peak

NoTE: War cycles, not shown, are 1914—19 and 5938—46. Deep depres-sion cycles are: 5895—5904, 1904—08, 19153—21, 5927—32,and 1932—38. All others are mild depression cycles. These datesdiffer from those shown in Charts and 4, because they areannual instead of monthly.

2. In an earlier article,13it was demon-strated that this cyclical pattern of velocitycould be largely though not wholly accountedfor by supposing that the amount of moneydemanded in real terms is linked, not to cur-rent measured income and current measuredprices, but to longer-term concepts of perma-nent income and permanent prices. By thisinterpretation, the amount of money demandedrises during the expansion phase of a cycle ingreater proportion than permanent income, assuggested by the secular results. However,measured income rises in still greater propor-tion, so that measured income rises relative tothe stock of money, and conversely during acontraction. While this interpretation does notrule out the possibility that changing interestrates over the cycle play a role in the cyclicalbehavior of velocity, it assigns them a less im-

Friedman, The Demand for Money: Some Theoreticaland Empirical Results, New York, NBER, '959, OccasionalPaper 68, reprinted from Journal of Political Economy,August 5959, pp. 327—355.

"Cash Balances and the Interest Rate — A PragmaticApproach," Review of Economics and Statistics, Nov. 1954,pp. 456—460; also idem, "Income Velocity and InterestRates A Pragmatic Approach," Employment, Growth,and Price Levels, Joint Economic Committee, Hearings,part io, 86th Cong., ist sess., pp. 3435—3443 (reprinted withminor changes in Review of Economics and Statistics, Nov.ig6o, pp. 445—449); and see Allan H. Meltzer, "The Demandfor Money: The Evidence from the Time Series," presentedat the Dec. 1961 meeting of the Econometric Society.

The elasticity of a total is a weighted average of theelasticities of the components, the weights being the ratioof each component to the total. Over the period from 1954to ig6o, commercial bank time deposits have varied from 59to 44 per cent of money as we define it.


is in income for a much longer period — anyexcess of the "correct" elasticity over unitycould readily be confounded in the statisticalanalysis with the effects of interest rates. Ourown readiness to attribute the decline in veloc-ity to income, despite the strong trend in in-come, derives primarily from the consistencyof such an interpretation with a wide range ofother evidence, in particular, cross-section evi-dence for different states in the United Statesand for different countries.

b. It is plausible that the division of cur-rency plus deposits between currency plus de-mand deposits, on the one hand, and time de-posits, on the other, is sensitive to rates ofinterest, since the differential between interestpaid on time deposits and interest paid on de-mand deposits (which can be and for longperiods has been negative) and on currency(typically zero) can be expected to widen asinterest rates rise — and conversely. Hence arise in interest rates might be expected to leadto an increase in commercial time de-posits relative to commercial bank demanddeposits plus currency — and conversely. Itfollows that the interest elasticity of demandcan be expected to be greater in absolute valuefor currency plus demand deposits, than forcurrency plus demand deposits plus time de-posits in commercial banks.

c. The two preceding points have especialimportance for the longer-term movements invelocity. For the cyclical behavior of velocity,the distinction between measured and perma-nent income can be combined with either de-mand function, and will help to explain thecyclical behavior of velocity.

Needless to say, neither definitioncan be said to be "the" correct definition. Justwhere the line is drawn between those tem-porary abodes of purchasing power we chooseto term money and those we term "near-monies," or "liquid assets," or what not, islargely arbitrary. We have found it convenientto draw the line where we do largely becausethat enables us to use a single concept for thewhole of our period, since the distinction be-tween commercial bank demand and time de-posits did not acquire its current significance— or indeed have much significance at all —until after 1914. In the course of using it, we

have found it to have some other advantages.'8In addition, even for the period since 1914, itis by no means clear that demand deposits asrecorded correspond fully with the economicconstruct Latané wishes to measure, namely,deposits subject to check. The lower reservesrequired against time deposits have given banksan incentive to classify as large a fraction ofdeposits as time deposits as possible. There issome evidence that, particularly during the192 0'S, banks managed so to classify somedeposits that were in effect demand deposits.A full understanding of the behavior of moneyin business cycles requires an analysis of thecomponents of the money stock, however de-fined, and of near-monies as well, so, despiteour reservations about the meaning of some ofhis data, we welcome Latané's analysis as avaluable complement to ours.

5. A basically more important question isthe extent to which velocity can be regardedas passively reflecting independent changes inits numerator and denominator. This is thepresumption implicit in the cycle theories, pop-ular these past few decades, that have regardedinvestment as the dominant cycle-producingfactor. These theories implicitly take forgranted that an expansion of investment willproduce an expansion in income regardless ofwhat happens to the money stock. In theirmost extreme form, these theories imply thatthe magnitude of the expansion in income isindependent of the size of any concurrentchange in the money stock. If the money stock

Still another bit of evidence on which of the twodefinitions of money is to be preferred is available. Wecomputed correlations like those in Table 4 for the period1915—60 between the variability of the narrower definitionand the variability of net national product, and also betweenthe variability of our broader definition and the variabilityof net national product. The broader definition has almostalways a somewhat higher correlation coefficient. The syn-chronous results for standard deviations of varying termsare shown in the following tabulation, giving correlationcoefficients between synchronous standard deviations ofannual rates of change in money — defined narrowly andbroadly — and in net national product, for different numberof terms.Definitionof Money 3.Term 4-Term 6.Term 7.Term 8-Term 9.Term

M1 .592 .833 .865 .909 .937 .931 .912M, .596 .785 .842 .883 .907 .899 .874

= Currency held by the public, plus demand depositsadjusted, plus commercial bank time deposits.

M, = Currency held by the public, plus demand depositsadjusted.


does not rise, then velocity will simply rise tofill the gap; if the money stock does rise, velocitywill not rise as much or may even fall. Themost rigorous explicit theoretical formulationof this position is in terms of either a "liquiditytrap" — an infinitely elastic liquidity prefer-ence function at a finite interest rate — or acompletely inelastic demand schedule for in-vestment — a zero response of spending to achange in the rate of interest. Though feweconomists would explicitly maintain thateither the one or the other prevails currently,or has prevailed during most of our past his-tory, many would accept the logically equiv-alent assertions that the rate of cyclical ex-pansion or contraction can be regarded asfairly rigidly determined by the rise or fall ininvestment or autonomous expenditure, thatthe link is far more crucial than any link withthe contemporary behavior of the money stock,and can be reversed, if at all, only by a veryatypical behavior of the money stock. Somerelevant empirical evidence on this issue issummarized in the subsection below on therelative roles of money and investment.

Cyclical behavior of proximate determinantsof the money stock.

i. Changes in the stock of money can,arithmetically, be attributed to changes inthree proximate determinants, each under theimmediate control of a different class of eco-nomic actors:

a. High-powered money, consisting of cur-rency held by the public, plus currency held inbank vaults, plus deposits of banks at FederalReserve Banks. This total is either a conse-quence of international payment flows andassociated gold movements, or of Treasury orFederal Reserve policy.

b. The division of the public's money hold-ings between currency and deposits, which canbe summarized by any one of a number ofratios — of currency to the money stock; ofcurrency to deposits; or of deposits to cur-rency. This division is in the first instancedetermined by the public, the holders of money,though, of course, the public's decision is af-fected by the terms offered by banks for de-posits.

c. The relation between deposits and the

amount of high-powered money held by banks,which can be termed their reserves. This rela-tion can be summarized by either the ratio ofreserves to deposits or its reciprocal, the ratioof deposits to reserves. This ratio is in thefirst instance determined by banks though, ofcourse, their decision is affected by legal re-quirements imposed by the government, bythe terms they must offer to obtain deposits,and by the returns they can receive on thealternative assets they acquire.

Given the two ratios, a rise in high-poweredmoney implies a proportional rise in the stockof money. Given the amount of high-poweredmoney and the deposit-reserve ratio, a rise inthe deposit-currency ratio implies a rise in thestock of money, because it means that lesshigh-powered money is required to meet thecurrency demands of the public and more isavailable for bank reserves to be multipliedby the deposit-reserve ratio. Similarly, giventhe amount of high-powered money and thedeposit-currency ratio, a rise in the deposit-reserve ratio implies a rise in the stock ofmoney, because it means that each dollar ofhigh-powered money held by banks gives riseto a larger number of dollars of deposits.

2. Phillip Cagan has analyzed in detail thecontribution of changes in each of these threeproximate determinants to the cyclical fluctua-tions in the rate of change in the moneystock.17 He finds that the deposit-currencyratio was the most important single contribu-tor. Throughout the period from 1877 to 1954,it accounted on the average for roughly halfthe cyclical fluctuations in the rate of changein the money stock. Though this fractionvaried 'from cycle to cycle, it did not change inany consistent secular fashion and was notmarkedly different for severe and mild move-ments. The main deviation in its contributionoccurred at times of money panics in which itoften played a dominant role.

Changes in high-powered money were aslarge in amplitude as changes in the deposit-reserve ratio but much less regular in timing.Changes in the deposit-reserve ratio were reg-

17 See his forthcoming monograph, "Determinants andEffects of Changes in the U.S. Money Stock, 1875—1955," aNational Bureau study.


ular in timing but relatively small in ampli-tude.

3. Cagan finds that the main impact of theFederal Reserve System has been on the rela-tive importance of changes in high-poweredmoney and in the deposit-reserve ratio. Byproviding banks with an alternative source ofliquidity, the Reserve System intensified atendency for banks to trim any excess of re-serves over legal requirements — a tendencyfostered in earlier decades by the Treasury'sassumption of enlarged money market respon-sibilities. The result was a reduction in theamplitude of cyclical movements in the re-serve ratio after 1914. However, this was morethan offset by an increase in the amplitude ofcyclical movements in high-powered money.

4. The deposit-currency ratio had a risinglong-term trend to 1929, declined substantiallythereafter until the end of World War II, andhas since been rising. Relative to these longer-term movements, the deposit-currency ratiotended to rise during the early part of expan-sions, at first at an increasing rate; to reacha peak near mid-expansion; then to decline tomid-contraction; and then to start rising.Cagan shows that these movements played animportant part in accounting for the tendencyof the rate of change in the money stock toreach its peak around mid-expansion and itstrough around mid-contraction. He attributesthe timing of movements in the deposit-currencyratio to divergent cyclical patterns in the veloc-ity of currency and deposits.

5. The deposit-reserve ratio rose duringmost of the period covered, except for its sharpdecline during the later 1930'S. Relative totrend, it tended to rise during expansions,reaching its peak before the reference peak,and tended to decline during contractions,reaching its trough before the reference trough.

6. These patterns bespeak a rather complexfeedback mechanism whereby changes in busi-ness activity react on the stock of money. Thisfeedback mechanism has not yet been workedout in the detail that would be desirable.

Relative roles of money and investment in thecycle.

In an extensive statistical study using stand-ard correlation techniques rather than the Na-

tional Bureau's cycle analysis, one of us incollaboration with David Meiselman investi-gated the relative stability of monetary velocityand the investment multiplier.'8 Both thestock of money and the level of autonomousexpenditures are positively related to consump-tion and to income over both short and longspans of years. However, it turns out that thecorrelation is generally much higher for moneythan for autonomous expenditures. Moreover,the partial correlation between money and con-sumption, holding autonomous expendituresconstant, is roughly the same as the simplecorrelation, whereas the partial correlation be-tween autonomous expenditures and consump-tion, holding the stock of money constant, ison the average roughly zero, being sometimespositive, sometimes negative. Similar resultswere obtained for year-to-year and quarter-to-quarter changes in the stock of money,autonomous expenditures, and consumption.

Additional evidence is provided by correla-tions between the variability of annual changesin money and in consumption, on the one hand,and between the variability of annual changesin investment and in consumption, on the other.Because there are occasional negative figuresfor net capital formation, we used gross capitalformation as the measure of investment andcomputed first differences of logarithms andmoving standard deviations of the first dif-ferences, as in Table 4, for money, con-sumption, and investment. The synchronouscorrelation coefficients we obtained are con-sistently higher, both for the period as a wholeand for the period since 1899, for money-consumption variability than they are for in-vestment-consumption variability. These areexactly the same results as in the Friedmanand Meiselman study, although derived by awholly different procedure. For the full period,the correlation coefficient for money-invest-ment variability is slightly lower than forinvestment-consumption variability; for theperiod since 1899, slightly higher. In addition,the partial correlation between money-con-sumption variability, holding investment van-

"Milton Friedman and David Meiselman, "The RelativeStability of Monetary Velocity and the Investment Multi-plier in the United States, 1897—1958," in Stabilization Poli-cies (in press, 1963), pp. 165—268, Prentice-Hall for theCommission on Money and Credit.


ability constant, is significantly higher thanthe partial correlation between investment-consumption variability, holding money vari-ability constant; for the period since 1899, thepartial correlation between money-investmentvariability, holding consumption constant, issignificantly higher than the partial correla-tion between investment-consumption variabil-ity, holding money constant, although for thewhole period, the former is lower. Essentiallythe same results were obtained for the simpleand partial correlations with leads and lags.19

These results are striking because theycontradict so sharply the widespread presump-tion among economists that investment (or,more generally, autonomous expenditures) isthe prime mover in cyclical fluctuations, trans-mitting its influence to the rest of income viaa multiplier effect on consumption. So far asthese results go, they suggest that, for a givenstock of money, there is no systematic relationat all between autonomous expenditures andconsumption — in experience, the multipliereffect on consumption is as likely to be negativeas positive.20 These results may of course bemisleading, because some crucial variableshave been neglected, or because the definitionused for autonomous expenditures is inappro-priate, or for some other reason. But theytend to be supported by preliminary resultsfor other countries, and we know of no con-

10The synchronous simple and partial correlation coeffi-cients for the moving 4-term standard deviations of the firstdifferences of logarithms are shown in the following tabula-tion for the full period and the period since

Simple Correlations Partial CorrelationsPeriod TI 'iM TIM 'iM.C

1870—1958 .749 .404 .330 .703 .252 .044

1899—1958 .8ii .6oo .677 .687 .120 .406

C = ConsumptionM = Money stock

I = Gross capital formation

If net capital formation is used as the measure ofinvestment, first differences of absolute values must be ob-

tained. We calculated the standard deviation of those first

differences, and the logarithm of the standard deviation,and then correlated the logarithms as above. There is atrend element in these calculations that it would be desirableto eliminate but, even so, the correlation coefficients aresimilar to those described for the standard deviation of firstdifferences of logarithms.

'° investment multiplier is generally defined as theratio of a change in income rather than in consumption tothe change in autonomous expenditures to which the changein income is attributed. In these terms, the conclusion isthat the multiplier is as likely in practice to be less thanunity as greater than unity.

trary evidence for the United States. Thewidespread presumption to the contrary thatunquestionably does exist, whether it be rightor wrong, does not rest, so far as we can see,on any coherent, organized body of empiricalevidence.2'

Some Plausible Interpretations of theFactual Evidence

The stock of money displays a consistentcyclical behavior which is closely related to thecyclical behavior of the economy at large.This much the factual evidence summarizedabove puts beyond reasonable doubt.

That evidence alone is much less decisiveabout the direction of influence. Is the cyclicalbehavior of money primarily a reflection ofthe cyclical behavior of the economy at large,or does it play an important independent partin accounting for the cyclical behavior of theeconomy? It might be, so far as we know, thatone could marshal a similar body of evidencedemonstrating that the production of dress-makers' pins has displayed over the past ninedecades a regular cyclical pattern; that thepin pattern reaches a peak well before the ref-erence peak and a trough well before the ref-erence trough; that its amplitude is highly

It is well established that (i) investment expenditureshave a wider cyclical amplitude than consumption expendi-tures have relative to their mean value; (2) orders and otherseries reflecting investment decisions, as contrasted withexpenditures, display a consistent tendency •to lead cyclicalturns; there is a high correlation between consumptionand income.

None of these is very strong evidence for the multipliereffect of investment on consumption, which is the point atissue. Item i simply means that investment is a morevariable component of income than consumption is; it saysnothing about whether both fluctuate in response to com-mon influences, investment influencing consumption, or con-sumption influencing investment. Note that a strict multi-plier model has no implications about whether autonomousor induced expenditures should show wider absolute fluctua-tions. Absolute fluctuations in induced expenditures wouldpresumably be wider or narrower as the usual multiplieris greater or less than 2.

Item 2 has more significance and has some suggestivevalue. However, it may simply mean that decisions areaffected early by whatever also affects spending later on(see page 6i, below). Item 3 is entirely irrelevant. Con-sumption is a major component of income, as both aremeasured. For multiplier effects, what is important is theeffect of investment on consumption. See M. Friedman andG. S. Becker, "A Statistical Illusion in Judging KeynesianModels," Journal of Political Economy, Feb. pp. 64—75.


correlated with the amplitude of the move-ments in general business. It might even bedemonstrated that the simple correlation be-tween the production of pins and consumptionis higher than the simple correlation betweenautonomous expenditures and consumption;that the partial corre'ation between pins andconsumption — holding autonomous expendi-tures constant — is as high as the simple cor-relation; and that the correlation between con-sumption and autonomous expenditures —holding the production of pins constant — is onthe average zero. We do not, of course, knowthat these statements are valid for pins and,indeed, rather doubt that they are but, evenif they were demonstrated beyond a shadowof doubt, they would persuade neither us norour readers to adopt a pin theory of businesscycles.

If the only decisive statistical evidence formoney were comparable to the items just citedfor pins, it would correspondingly not justifythe acceptance of a monetary theory of busi-ness cycles. At the same time, it is worth not-ing that, even then, the monetary theory andthe pin theory would by no means be on allfours. Most economists would be willing todismiss out of hand the pin theory even onsuch evidence; most economists would takeseriously the monetary theory even on muchless evidence, which is not by any means thesame as saying that they would be persuadedby the evidence. Whence the difference? Pri-marily, the difference is that we have otherkinds of evidence. We know that while pinsare widely used and occasionally of criticalimportance, taken as a whole, they are aminor, if not trifling, item in the economy. Weexpect the effect to be in rough proportion tothe cause, though this is by no means alwaysthe case — a rock can start a landslide. Wecan readily conceive of an economy operatingwithout pins yet experiencing cycles like thoseof history; we can readily conceive of largeautonomous changes occurring in the produc-tion of pins, but we cannot readily conceive ofany channels through which such autonomouschanges could have wide-reaching effects onthe rest of the economy. Men who havethought about and studied these matters havenever been led to suggest the pin industry as

a prime mover in the cyclical process. In allthese respects, the monetary theory is on awholly different footing. We know that moneyis a pervasive element in the economy; thatthe stock of money is sizable compared withother aggregate economic magnitudes; thatfluctuations of the kind we call business cycleshave apparently occurred only in an economyin which "economic activities are . . . car-ried on mainly by making and spendingmoney." 22 We not only can conceive of themoney stock's being subject to large auton-omous changes, but we can also readily conceiveof channels through which such changes couldhave far-reaching effects on the rest of theeconomy. Men who have thought about andstudied these matters have been led to givemoney a critical role in their theories.

One more preliminary observation. The keyquestion at issue is not whether the directionof influence is wholly from money to businessor wholly from business to money; it iswhether the influence running from money tobusiness is significant, in the sense that it canaccount for a substantial fraction of the fluc-tuations in economic activity, lithe answer isaffirmative, then one can speak of a monetarytheory of business cycles or — more precisely— of the need to assign money an importantrole in a full theory of business cycles. Thereflex influence of business on money, theexistence of which is not in doubt in light ofthe factual evidence summarized above, wouldthen become part of the partly self-generatingmechanism whereby monetary disturbances aretransmitted. On the other hand, if the influ-ence from money to business is minor, onecould speak of a cyclical theory of monetaryfluctuations but not of a monetary theory ofbusiness cycles. To illustrate again with pins:Changes in business conditions doubtless affectthe production of pins, and no doubt there issome feedback effect of changes in the produc-tion of pins on general business. But, whereasthe first effect may well be large relative to thetotal fluctuations in pin production, the feed-back effect is almost certainly trivial relativeto the fluctuations in business. Hence we areready to accept a business cycle theory of pin

22 Wesley C. Mitchell, Business Cycles: the Problem andIts Setting, New York, NBER, 2927, Chapter II, and p. 62.


production but not a pin theory of businesscycles.

The factual evidence summarized above goesbeyond the list of items we conjectured forpins and contains some bits that are relevantto the key question at issue. The most im-portant is the fact that the relation betweenmoney and business has remained largely un-changed over a period that has seen substantialchanges in the arrangements determining thequantity of money. During part of the period,the United States was on an effective goldstandard, during part, on an inconvertible paperstandard with floating exchange rates, duringpart, on a managed paper standard with fixedexchange rates. The commercial banking sys-tem changed its role and scope greatly. Thegovernment arrangements for monetary con-trol altered, the Federal Reserve System re-placing the Treasury as the formal center ofcontrol. And the criteria of control adopted bythe monetary authorities altered. If the pre-dominant direction of influence had been frombusiness to money, these changes might havebeen expected to alter the relation betweenbusiness changes and monetary changes, butthe relation has apparently remained much thesame in both timing and amplitude.23 Yet thisevidence is by no means decisive. As notedabove, Cagan shows that the public's decisionsabout the proportion in which it divides itsmoney balances between currency and depositsis an important link in the feedback mech-anism whereby changes in business affect thestock of money. The changes in monetary ar-rangements have affected greatly the trends inthe deposit-currency ratio but appear not tohave affected its cyclical behavior. Hence thispart of the supply mechanism has been roughlyconstant and has played a roughly constantrole over the whole period.

In our view, the most convincing evidencesupporting the idea that money plays an im-portant independent part is not the evidencesummarized in the first part of this paper butevidence of a rather different kind — thatgarnered from study of the historical circum-stances underlying the changes that occurred

23 See also comments in Friedman, "The Lag in Effect ofMonetary Policy," pp. 449—450.

in the stock of money.24 This evidence is muchmore clear cut for major movements than forminor.

Major economic fluctuations.Major movements in U.S. history include

the deep depressions used here to distinguishdeep from mild depression cycles in our classi-fication of historical reference cycles (seeChart 2 for the classification); the substantialinflations which have occurred primarily dur-ing wartime; and a few long-continued move-ments in one direction, such as the generallyrising level of money income and prices from1896 to 1913. With respect to these events, thehistorical record justifies two important gen-eralizations.

i. There is a one-to-one relation betweenmonetary changes and changes in money in-come and prices. Changes in money incomeand prices have, in every case, been accom-panied by a change in the rate of growth ofthe money stock, in the same direction and ofappreciable magnitude, and there are no com-parable disturbances in the rate of growth ofthe money stock unaccompanied by changes inmoney income and prices.

2. The changes in the stock of money cannotconsistently be explained by the contemporarychanges in money income and prices. Thechanges in the stock of money can generallybe attributed to specific historical circum-stances that are not in turn attributable tocontemporary changes in money income andprices. Hence, if the consistent relation be-tween money and income is not pure coinci-dence, it must reflect an influence running frommoney to business.

INFLATIONARY EPISODES. The second gen-eralization requires little more than its state-ment to be recognized as true for the infla-tionary episodes. During periods of U.S.engagement in wars, the increased rate ofgrowth of the money stock stemmed from useof the printing press, in more or less subtleways, to help finance government military ex-penditures. During our neutrality in WorldWar I from 1914 to early 1917, it had its origin

"For the United States, since the end of the Civil War,see our forthcoming volume, "A Monetary History of theUnited States, 1867—1960."


in use by the Allies of their gold reserves tofinance war purchases here. During those waryears, the reflex influence of the rising tide ofbusiness on the stock of money was in theopposite direction to the actual movement inthe money stock, since business expansion ofitself tended to produce a worsening in thebalance of payments and hence an outflow ofgold or a decreased inflow.

The situation is equally clear from 1896 to1913. The rise in the stock of money reflectedpredominantly an increase in the U.S. goldstock, which was part of a worldwide growthof the gold stock emanating from the discoveryof new mines and improvements in techniquesof extracting gold from low-grade ore. Thedomestic expansion alone would have made forgold outflows. The feedback was thereforecounter to the main current.25

For the wartime episodes, the evidence isequally consistent with a different theory, thatthe independent force was a major shift ingovernment spending propensities; that theshift in spending propensities would have hadthe same effect on income and prices if it hadbeen financed wholly by borrowing from thepublic at large with an unchanged moneystock, rather than being financed in part bythe use of monetary reserves (as it was in theearly years of World War I) or by governmentcreation of money (as in the other war years);that it was not financed wholly by borrowingbecause resort in part to use of monetary re-serves and the printing press was politicallyeasier and perhaps financially cheaper.

Evidence from the study by Friedman andMeiselman (discussed in the subsection on therelative roles of money and investment, page.47, above) rather decisively contradictsthis alternative explanation. In any event, thealternative explanation will not hold for the1896—1913 inflation, since there was no obviousindependent shift Of major magnitude in spend-ing propensities. The only immediate factorproducing such a shift that comes to mind isthe income earned from gold production. How-ever, although the increase in the stock of goldover that period was large compared to the

This point is discussed in more detail in Cagan's forth-coming "Determinants and Effects of Changes in the U.S.Money Stock, 1875—1955."

gold stock at the start and was capable ofproducing large increases in the stock of moneyvia a multiplicative effect on other kinds ofmoney, the gold stock itself was a small frac-tion of the total money stock, and the increasein the money stock only a fraction of the in-crease in money income. Hence, the value ofgold production was a small fraction indeedof the increase in income.26 The increased goldproduction could hardly have produced theobserved increase of money income throughany spending multiplier effect. But any effectit might have had must have been through itseffect on the stock of money.

DEEP DEPRESSIONS. For deep depressions,the historical evidence justifying our secondgeneralization is as clear as for the inflationaryepisodes, though less well known and henceless self-evident. A summary statement of theproximate source of the change in the moneystock will in most instances enable the readerto judge for himself the extent to which the de-cline in the stock of money can be explainedby the contemporary change in money, income,and prices.1875—78: Political pressure for resumption. led

to a decline in high-powered money,and the banking crisis in 1873 andsubsequent bank failures to a shiftby the public from deposits to cur-rency and to a I all in the deposit-reserve ratio.

1892—94: Agitation for silver and destabilizingmovements in Treasury cash pro-duced fears of imminent abandon-ment of the gold standard by theUnited States and thereby an out-flow of capital which trenched ongold stocks. Those effects were in-tensified by the banking panic of1893, which produced a sharp de-cline, first in the deposit-currencyratio and then in the deposit-reserveratio.

1907—08: The banking panic of i9o7 led to26 For the United States from 1896 to 1913, the value of

the gold stock increased by roughly $1.4 billion or by about$80 million a year; net national product increased fromabout $ii billion in 1896 to $34 billion in 1g13 or at therate of about $2,300 million a year.


a sharp decline in the deposit-cur-rency ratio and a protective attemptby banks to raise their own reservebalances, and so to a subsequentfall in the deposit-reserve ratio.

1920—21: Sharp rises in Federal Reserve dis-count rates in January 1920 andagain in June 1920 produced, withsome lag, a sharp contraction inFederal Reserve credit outstanding,and thereby in high-powered moneyand the money stock.

1929—33: An initial mild decline in the moneystock from 1929 to 1930, accom-panying a decline in Federal Re-serve credit outstanding, was con-verted into a sharp decline by awave of bank failures beginning inlate 1930. Those failures produced(i) widespread attempts by thepublic to convert deposits into cur-rency and hence a decline in thedeposit-currency ratio, and (2) ascramble for liquidity by the banksand hence a decline in the deposit-reserve ratio. The decline in themoney stock was intensified afterSeptember 1931 by deflationary ac-tions on the part of the Federal Re-serve System, in response to Eng-land's departure from gold, whichled to still further bank failures andeven sharper declines in the depositratios. Yet the Federal Reserve atall times had power to prevent thedecline in the money stock or toincrease it to any desired degree,by providing enough high-poweredmoney to satisfy the banks' desirefor liquidity, and almost surelywithout any serious threat to thegold standard.

1937—38: The doubling of legal reserve re-quirements in a series of steps, ef-fective in 1936 and early 1937, ac-companied by Treasury sterilizationof gold purchases led to a halt in thegrowth of high-powered money andattempts by banks to restore their

reserves in excess of requirements.The decline in the money stock re-flected largely the resultant declinein the deposit-reserve ratio.

A shift in the deposit-currency ratio and theaccompanying bank crises played an importantrole in four of these six episodes. This ratio,as we have seen, has a systematic cyclical pat-tern which can be regarded as a feedback effectof business on money. However, in each ofthose episodes, the shift in the deposit-currencyratio represented a sharp departure from thetypical cyclical response and, in at least two(1875—78 and 1892—94), represented a sub-sequent reaction to an initial monetary dis-turbance that had no such close link withcontemporary changes in money income andprices. Moreover, in two episodes (1920—2 Iand 1937—38), neither a shift in the deposit-currency ratio nor bank failures played anyrole. And such a shift has played no importantrole in any of the large expansions in the stockof money. A fractional reserve banking struc-ture susceptible to runs is an institutional fea-ture that renders the stock of money sensitiveto autonomous deflationary changes; henceruns may frequently play an important rolein sharp declines. This feature, however, isclearly not essential for a large economicchange to be accompanied by a large monetarychange in the same direction.

The 1907—08 episode is a particularly niceexample of the intermixture of autonomousmonetary disturbances and a feedback. Thefailure of the Knickerbocker Trust Companyin the fall of 1907 converted what had beena mild decline in the money stock as a resultof gold exports and a consequent decline inhigh-powered money into a severe decline asa result of bank runs and a consequent declinein the deposit-currency ratio. The accompany-ing sharp rise in short-term interest rates anda premium on currency produced a large goldinflow. The accompanying sharp intensifica-tion in the business decline worked in the samedirection by its effect on the balance of inter-national payments. Since the runs were pre-vented from producing widespread bank fail-ures through the concerted suspension bybanks of convertibility of deposits into cur-


rency, these feedback effects fairly promptlyreversed the money decline and, along with thereversal, the business decline came to an end.

CONCLUSIONS FOR MAJOR MOVEMENTS. Thefactors that produced the changes in the stockof money are autonomous only in the senseof not being directly attributable to the con-temporary cyclical changes in money incomeand prices. In a broader context, each ofcourse has its origins and its explanation, andsome are connected fairly clearly with longer-term economic developments. There can beno doubt, for example, that the silver agitationwas intensified by prior declining agriculturalprices, or that the financial boom in the early1900's encouraged financial activities whichlaid the basis for Knickerbocker Trust's fail-ure, or that the worldwide declining pricetrend of the 1870's and i88o's encouragedexploration for gold and improvement of re-fining techniques.

The narrower sense is, however, importantfor our purpose. The question at issue iswhether the one-to-one relation between mone-tary change and major economic change canbe explained by a relation running from eco-nomic change to money, as a one-to-one rela-tion between changes in pin production andin economic activity could be explained if itexisted. Such an explanation would requirethat the changes in money be connected ratherrigidly with either the contemporary changesin economic conditions or more basic factorsthat could account alike for the course ofeconomic events and for the changes in thestock of money. The demonstration that themajor changes in the stock of money have beenattributable to a variety of sources, many ofwhich are connected directly neither with con-temporary business developments nor withearlier business developments — which them-selves can be regarded as determining the con-temporary course of business — thereforecontradicts any such explanation of the one-to-one relation between economic change andmonetary change.

There seems to us, accordingly, to be anextraordinarily strong case for the proposi-tions that (i) appreciable changes in the rateof growth of the "stock of money are a neces-sary and sufficient condition for appreciable

changes in the rate of growth of money in-come; and that (2) this is true both for longsecular changes and also for changes overperiods roughly the length of business cycles.To go beyond the evidence and discussion thusfar presented: our survey of experience leadsus to conjecture that the longer-period changesin money income produced by a changed sec-ular rate of growth of the money stock arereflected mainly in different price behaviorrather than in different rates of growth of out-put; whereas the shorter-period changes inthe rate of growth of the money stock arecapable of exerting a sizable influence on therate of growth of output as well.

These propositions offer a single, straight-forward interpretation of all the historicalepisodes involving appreciable changes in therate of monetary growth that we know about inany detail.27 We know of no other single sug-gested interpretation that is at all satisfactoryand have been able to construct none for our-selves. The character of the U.S. bankingsystem — in particular, for most of its history,the vulnerability of the system to runs onbanks — can come close to explaining whysizable declines in money income, howeverproduced, should generally be accompanied bysizable declines in the stock of money; butthis explanation does not hold even for alldeclines, and it is largely irrelevant for therises. Autonomous increases in governmentspending propensities plus the irresistible po-litical attraction of the printing press couldcome close to providing a single explanation forwartime infiations, accounting for the coin-

27Though we have summarized here and have, ourselves,investigated in detail only the U.S. experience since 1867,this statement is deliberately worded so as to cover a widerrange of experience. For example, it is consistent with thehyperinflations studied by Cagan ("The Monetary Dy-namics of Hyperinflation," Studies in the Quantity Theoryof Money, M. Friedman, ed., University of Chicago Press,1956, pp. 25—217) ; with U.S. experience during the 2830'sand 2840's, studied by George Macesich ("Monetary Dis-turbances in the United States, 2834—45," unpublished Ph.D.thesis, University of Chicago, June with U.S. expe-rience during the Revolutionary War, the War of 1812, andthe Civil War; with Chilean experience, as studied by JohnDeaver ("The Chilean Inflation and the Demand forMoney," unpublished Ph.D. thesis, University of Chicago,1961); with the price revolution in the sixteenth century, asstudied by Earl J. Hamilton (American Treasure and thePrice Revolution in Spain, 1501—1650, Harvard UniversityPress, 1934).


cidence of rising incomes and rising stock ofmoney without any necessary influence run-fling from money to income; but this explana-tion cannot account for peacetime inflations,in which the growth of the money stock hasreflected a rise in specie rather than in govern-ment-issued money; and it is not even a satis-factory explanation for the wartime episodes,since price rises in different wartime episodesseem more closely related to the concurrentchanges in the stock of money than to thechanges in government expenditure.28

It is perhaps worth emphasizing and repeat-ing that any alternative interpretation mustmeet two tests: it must explain why the majormovements in income occurred when they did,and also it must explain why such major move-ments should have been uniformly accom-panied by corresponding movements in therate of growth of the money stock. The mone-tary interpretation explains both at the sametime. It leaves open the reasons for the changein the rate of growth of the money stock and,indeed, at this point is highly eclectic, takingaccount of the fact that historically there havebeen many different reasons.

We have emphasized the difficulty of meet-ing the second test. But even the first alone ishard to meet except by an explanation whichasserts that different factors may from timeto time produce large movements in income,and that these factors may operate throughdiverse channels — which is essentially toplead utter ignorance. We have cited severaltimes the apparently widespread belief in in-vestment as the prime mover. The alternativeexplanation for times of war, suggested above,is a special application of this theory, withinvestment broadened to mean "autonomousexpenditures" and government spending in-cluded in the same category. But even for thefirst test alone, we find it hard to accept thistheory as a valid general explanation: can adrastic collapse in autonomous investmentexplain equally 1873—79, 1892—94, 1920—2 i,1929—33, 1937—38? Capital formation at theend of the seventies was apparently one andone-half times its level at the beginning and

See Friedman, "Price, Income, and Monetary Changesin Three Wartime Periods," American Economic Review,May 1952, pp. 622—625.

seems not to have slumped seriously at any timeduring the decade, judging by the rough indica-tions given by Kuznets' figures.2° The 1890'ssaw some decline, but the following decade wasmarked by a vigorous and sustained rise. The1920—21 episode was destined to be followed bya construction and investment boom. If theexperience of 1920—21 is to be interpreted as aresult of an investment collapse, that declinemust have been a consequence of the declinein government expenditures and the subsequentcollapse of inventory speculation before fixedcapital expenditures had developed to take theirplace. But why, then, did the sharp decline ingovernment expenditures after World War IInot produce a subsequent economic collapse?Emphasis on inventory speculation involves ahighly episodic interpretation, since it charac-terizes few of the other episodes. Surely, onecannot adduce that in World War I, slow usingup of investment opportunities — often implic-itly or explicitly called on to explain why, fromtime to time, there is allegedly a collapse ofinvestment or a position of stagnation — wasresponsible for the 1920—2 I recession. This isan equally implausible explanation for 1937—38and, as already implied, for earlier episodes aswell.

Of course, in most or all of these contrac-tions, the incentive to invest and the actualamount spent on investment declined. Thequestion at issue, however, is whether the de-cline was a consequence of the contemporaryeconomic collapse — triggered, we would say,by monetary changes — or the ultimate work-ing out of autonomous elements of weaknessin the demand for investment that themselvestriggered the contraction.

Even if all these episodes of contraction cansomehow be interpreted as reflecting an autono-mous decline in investment, is a sharp increasein investment opportunities a satisfactory ex-planation for the worldwide 1897—1913 rise inmoney income? If money is not a critical linkbut only a passive accompaniment of change,how is it that China escaped the early years ofthe Great Depression? We would say thanks tobeing on a silver standard and hence having afloating exchange rate vis-à-vis gold currencies,

n Kuznets, Capital in the American Economy: ItsFormation and Financing, Princeton for NBER, 1961, p. 572.


whereas all countries linked to gold were en-meshed in the depression. And how is it thatChina had the most severe contraction of all inthe years from 1933 to 1936, when our silverpurchase program drained silver from Chinaand caused a sharp decline in its money stock,whereas the rest of the world was in a periodof business expansion? And we could extendthis list of embarrassing questions withoutdifficulty.

We feel as if we are belaboring the obviousand we apologize to any reader who shares thatfeeling. Yet repeated experience has led us tobelieve that it is necessary to do so in order tomake clear how strong is the case for the mone-tary explanation of major movements in moneyincome.

Of course, it is one thing to assert that mone-tary changes are the key to major movementsin money income; it is quite a different thing toknow in any detail what is the mechanism thatlinks monetary change to economic change;how the influence of the one is transmitted tothe other; what sectors of the economy will beaffected first; what the time pattern of the im-pacts will be, and so on. We have great confi-dence in the first assertion. We have littleconfidence in our knowledge of the transmissionmechanism, except in such broad and vagueterms as to constitute little more than an im-pressionistic representation rather than an en-gineering blueprint. Indeed, this is the chal-lenge our evidence poses: to pin down thetransmission mechanism in specific enough de-tail so that we can hope to make reasonablyaccurate predictions of the course of a widevariety of economic variables on the basis ofinformation about monetary disturbances. Inthe section below on the relation between varia-tions in income and money, we outline one partof the transmission mechanism which can ac-count for the greater amplitude of variation inincome than in money and on which we havesome empirical evidence; in the last section, wesketch in a much more tentative way the majorchannels through which monetary fluctuationsmight be able to account for economic fluctua-tions, both the major movements we have sofar been considering, and the minor movementsto which we now turn.

Minor economic fluctuations.The case for a monetary explanation is not

nearly so strong for the minor U.S. economicfluctuations that we have classified as mild de-pression cycles as the case is for the majoreconomic fluctuations. Clearly, the view thatmonetary change is important does not precludethe existence of other factors that affect thecourse of business or that account for the quasi-rhythmical character of business fluctuations.We have no doubt that other factors play a role.Indeed, if the evidence we had were solely forthe minor movements, it seems to us most un-likely that we could rule out — or even assigna probability much lower than 50 per cent to —the possibility that the close relation betweenmoney and business reflected primarily the in-fluence of business on money.

If we are inclined to assign a probabilitymuch lower than 50 per cent, it is primarilybecause the evidence for minor movements doesnot stand alone. If money plays an independ-ent role in major movements, is it likely to bealmost passive in minor movements? The minormovements can be interpreted as less virulentmembers of the same species as the majormovements. Is not a common explanation forboth more appealing than separate explana-tions, especially when there is no well-testedalternative separate explanation?

A fully satisfactory explanation of the minormovements would require an explicit and rigor-ously stated theory, which could take the formof a series of simultaneous differential equationsdescribing the reaction mechanism of the econ-omy, together with a specification of the jointdistribution function of the random disturb-ances impinging on it, and a specification of thesystematic disturbances that could be intro-duced into it. Our belief that 'money plays animportant role in minor movements is equiva-lent to asserting that some of these differentialequations would contain the stock of money asa variable; that disturbances in the stock ofmoney are among the random or systematicdisturbances impinging on the system; and thatthese disturbances alone would be capable ofgenerating a path for such major economicvariables as money income, prices, output, andthe like, comparable to the path they actuallyfollow during mild depression cycles.


One factor that has doubtless contributed toskepticism about a monetary theory is the fact,documented above, that fluctuations in incomeare wider in relative amplitude than fluctua-tions in the stock of money. We have seenthat income velocity varies positively over thecycle, which means that income varies morewidely than money. We have seen also that thestandard deviation of year-to-year percentagechanges in income tends to be roughly doublethe standard deviation of year-to-year changesin the stock of money. How is it that suchsmall changes in money can produce so muchlarger changes in income? Why should mar-ginal velocity be systematically higher thanaverage velocity?

While we are far from having a rigorous andcomprehensive theory to answer this and re-lated questions, in the next section we outlineone element of such a theory which can, in ourview, explain the difference in amplitude; andin pages 59—63, we outline even more broadly atentative transmission mechanism.

Relation between amplitude of cyclical varia-tions in income and money.

One of us has elsewhere suggested that hold-ers of money can be regarded as adjustingthe nominal amount they demand to theirviews of their long-run income status — itselfa measure of their wealth — of the long-runlevel of prices, and of the returns on alterna-tive assets.8° Let us neglect for the time beingthe effect of returns on other assets, as well asstill other possible variables, so that we canwrite the relationship for the community as(I) = Pp f(yp),where Md is nominal amount of money percapita, is permanent prices, and is per-manent aggregate real income per capita.8' Thecapital letters here and later refer to magni-tudes in nominal terms or current prices, the

The Demand for Money.We call to the reader's attention the difference in this

notation from that in The Demand for Money. and y,,here refer to per capita money and income, whereas in theearlier paper they were used to refer to aggregate money andincome. The shift was prompted by the desire to simplifythe expressions that follow. The same shift is made for allvariables referring to money and income. The remainingsymbols all have the same meaning here as in ibid.

lower-case letters, to magnitudes in real termsor constant prices.

Let us suppose further that estimates of percapita permanent income and permanent pricesare compounded of two elements: (i) an ex-pected average annual rate of change to allowfor secular trend at a rate of, say, a8 for incomeand ap for prices; (2) a weighted arithmeticor geometric average of past per capita in-comes and prices adjusted for such a trend.

For the present, we shall assume that andap are both zero, or alternatively that theactual past record is replaced by the past recordadjusted for trends of a9 and ap in magnitude.At the present level of discussion, this assump-tion involves no loss of generality, since theonly effect of nonzero values of a9 and ap is toadd secular trends without affecting cyclicalfluctuations. On a more sophisticated level, itwould make a difference, since both and apmight be variables in the demand function formoney, the former since future prospects mightmodify present demand for money, the lattersince it would affect the returns on some alter-native assets.

We can then write:(2) P9(T)=F[P(t); t<T](3) y,,(T) = G[y(t); t < T},where P(t) and y(t) are measured prices andmeasured real income per capita at time t, andthe functions are to be interpreted as sayingthat permanent prices and income are func-tions of the past history of measured incomeor prices. If we consider discrete data, say,annual data, we can approximate equations 2and 3 by either

cc(2a) P5(T) P(T — 1) = wo' P(T)i=o

+ (' — we') P9(T — I),

(3a)=

— i) = wo' y(t)

where

or by

+ (' — w0') y9(T —

I;

00(2b) logP,,(T) = — i) = w0log

1=0P(T) + (i — log —

MONEY AND BUSINESS CYCLES00

(3b) log y3(T) = log y(T — 1) =1=0

log y(T) + (' — w0) log i)where

z =

Note that, in both cases, we have assumedthat the same weights are used for income andprices.

Suppose the community is regarded as al-ways being on its demand curve for money.Then an increase in the stock of money will re-quire an increase in permanent income or pricesor both sufficient to make the community satis-fied with the new stock of money, and theseincreases can be brought about only by in-creases in measured income or prices or both.To illustrate: Suppose, for simplicity, realmeasured income and real permanent incomeremain unchanged. Then from equation i, aone per cent change in M will require a one percent change in P,,. But from equationa one per cent change in will require thatP(T) rise by more than one per cent, or by4- per cent for equation 2 a and per cent forwo woequation 2b. But w0' and w0 are less than unity.Hence, the percentage rise in measured pricesand income will be larger than the percentagerise in money.

To be more specific and to allow for changesin both prices and income, let us replace equa-tion i by a special form we have found to workrather well empirically:

M(4) _=7(yp)8,P p

where y and 8 are numerical constants (or,more generally, functions of omitted variables,such as returns on other assets), all the vari-ables are at time T, and we have dropped thesubscript d from M because of our assumptionthat the amount demanded is always equal tothe amount supplied. In logarithmic form,is(4a) log M(T) log y + log

+ 8 log y9(T).Substitute (2b) and (3b) into (4a), giving

log M(T) = log y + w0(i — 8) log P(T)+ 8 wo log Y(T) + (' — w0)

[log — ') + S log —

wherelog Y(T) = log y(T) + log P(T),

i.e., Y(T) = measured income per capita. Re-place the final bracket in by its equivalentfrom (4a) for T — i, namely, [log M(T —— log 'y], and then solve (5) for log Y(T).This gives

(6) log V(T) = —f.-— {log M(T) — logSw0

—w0(i — 8)logP(T)— (I —we)[log M(T — i) — log y]}.

Differentiate equation 6 with respect toM(T), allowing for the fact that P(T)change along with Y(T). This gives

d log Y(T) i=—[i—w0 (z—S)d log M(T) Sw0d log P(T) d log Y(T)d log Y(T) X d log M(T)

Solve for d log V(T)/d log M(T) to getd log Y(T) — i

" d log M(T) — w0[8 + (' — 8)

where is the elasticity of the measured pricelevel with respect to measured income, and canbe expected to be between zero and unity forcyclical fluctuations (i.e., both prices and out-put can be expected to move in the same direc-tion as money income). We may designated log Y(T)/d log M(T) the money multiplier,analogous to the investment multiplier, thoughit should be noted that the analogy is somewhatincomplete. The money multiplier gives theratio of the percentage change in income to thepercentage change in the money stock.32 Toget the number of dollars of income change perdollar change in the stock of money, it is neces-sary to multiply the money multiplier by theincome velocity of money.

It so happens that our earlier work furnishesempirical estimates for the United States of allthe quantities entering into the right-hand sideof equation 8. Hence, we can construct anestimate of the elasticity of money income withrespect to the money stock. These estimatesare as follows: 88

Because of the assumption that and ap are zero, oralternatively that the actual past record is replaced by thepast record adjusted for trend, what is here called a changein the money stock is logically equivalent to a change in themoney stock relative to its trend, or to a change in therate of change.

Friedman, The Demand for Money. (x) A

logwill


= 0.338 = 1.81

= 0.20Inserting these figures in equation 8 gives

d log Y(T)Estimate d log M(T) = 1.84

This estimate is certainly remarkably closeto the estimate, based on Table 4, of the ratioof the variability of income to the variabilityof money. It will be recalled that we therefound this ratio to be almost exactly 2.0. So

esti-Yet,

value of = 0.4 implies a weight for the first year of 0.33;the value of 8 is from equation 9 of ibid.; the value

of ij is derived from Table x of ibid. by dividing the entryfor "implicit price deflator" in column (3) by the entry for"money income" in the same column.

With respect to Ci), it should be noted that permanentincome and prices were computed in ibid. by equationsand rather than 2b and 3b. We have nonetheless takenthe resulting value of W'o in our present notation as anestimate of w,. This is correct as a first approximation,but in further work it would probably be better to workdirectly with equations 2b and 3b.

With respect to (s), the number used is for aggregatemoney income, not per capita. However, since the numberis the difference between the per month rates of rise duringexpansion and contraction, and since population shows littleresponse to cycles, the per capita figures would be lower byroughly the same amount for expansion and for contraction,and hence the difference would be unaffected.

3' One way to see this is to consider the problem ofestimating directly the magnitude of the money multiplierfrom data on actual year-to-year changes in the logarithmsof income and money. The first step would be to expressthe first differences as deviations from some mean values,designed to be the empirical counterparts of our theoreticalconstructs: e3 + ii,, = the expected rate of change in money(permanent) income; and + = the rate of change inthe stock of money that would be consistent with a rate ofchange of ev in real income and a,, in prices. That is, ifmoney income, prices, and the stock of money all changedat exactly these rates, all expectations would be realized andthere would be no disturbances to set the money multiplier,as we have defined it, to work. This first step is accom-plished in our moving standard deviation analysis by com-puting, first, moving averages, and then expressing theobserved first differences as deviations from the relevantaverage. Call these deviations from means, A' log V and

log M.The second step would be to estimate the mean ratio of

log V to A' log M. But it would be undesirable to dothis by averaging the ratio of the one to the other, sinceeither might on occasion be close to zero (i.e., the varianceof the ratio is in principle infinite). It would be better toestimate a value of, say, K in

A' log V = K A' log M.But as a statistical matter, there is no particular reason toprefer the estimate obtained by regressing A' log V onA' log M to the estimate obtained by regressing A' log Mon A' log V. In its rigid form the money multiplier analysis

In such further exploration it would be de-sirable to generalize this analysis in a numberof respects. (i) should not be treated as anumerical constant. One would expect it to bedifferent at different stages of the cycle andunder different circumstances. Under condi-tions of full employment and inflation, it wouldbe unity or close to it, which — given that 6is greater than unity — would make the moneymultiplier a maximum of i/w0, or with ourwould imply perfect correlation, so the two regressionswould be the same except for statistical errors of estimate.The "correct" regression then depends on the magnitude oferrors in A' log V and A' log M. As is well known, thetwo simple regression coefficients give upper and lowerbounds to any estimates obtained by treating both variablesas subject to error. The geometric mean of these twobounding estimates is precisely the ratio of the standarddeviation of A' log V to the standard deviation of A' logM.

have used the estimates of Wo, 8, and abovebecause they are available in published form. We have beenexperimenting further with estimating demand equationsusing annual data instead of cycle bases, and estimating Wointernally from the money correlations themselves, ratherthan externally. This work is still tentative but one set ofresults may be cited, because they are at the moment themost divergent from those given above.

For the years 1885—1960, one estimate of w0 15 0,22 andof & is 2.27. Inserting these along with = 0.20 into equa-tion 8 gives an estimate of the money multiplier of 2.25, oron the other side of the estimate of 2.0 from Table 4. In-terestingly enough, this estimate is very close to the ratio,formed from the geometric means of the computed standarddeviations, which ranges from 2.13 to 2.32 for differentnumbers of terms (see footnote 22).

far as we can see, these two numbers aremates of the same theoretical construct.84

statistically, they are almost completely inde-pendent. The estimate in equation 9 comesfrom the following sources: w0 is based on astudy of the consumption function which usedno data on money whatsoever; 6 is based on acorrelation between average cycle bases ofmoney and estimated permanent income; and

is based on the ratio of per month cyclicalamplitudes computed from average cycle pat-terns of money income and prices. Hence, sofar as we can see, no one of these items uses inany way the intracyclical movements of money.Yet the estimate of 2.0 based on Table 4 has inits denominator the average standard devia-tion of sets, containing 3, 4, 5, or 6 years, ofyear-to-year percentage changes in the stockof money. The close agreement of two esti-mates, statistically so independent, certainlystrongly suggests that the theoretical structurewhich produced them deserves further explora-tion


estimate of w0, 3. At the other extreme, if therewere extensive unemployment, might be closeto zero (though it is by no means clear that thishas been true in experience), which would makethe money multiplier a minimum of orwith our estimates, 1.67. More generally,plays an important role not only in any theoryalong the general lines we have been sketchingbut also in income-expenditure theories.36 Itdeserves much more systematic study than ithas received. (2) The demand equation 4should be expanded to include interest ratesand perhaps the rate of change in prices.Though our studies suggest that these are farless important than income in affecting thedemand for money, interest rates do have astatistically significant effect and, since theyhave a fairly regular cyclical pattern, shouldbe included in a cyclical analysis. Theeffect of expected trends in prices and incomeshould be allowed for explicitly and not simplyneglected, as we have done. For cycleanalysis, the demand equations should beestimated on a quarterly rather than annualbasis. In generalizing to a quarterly basis,it will no longer be satisfactory to suppose thatactual and desired money balances are alwaysequal. It will be desirable to allow instead fora discrepancy between these two totals, whichthe holders of balances seek to eliminate at arate depending on the size of the discrepancy.This will introduce past money balances into theestimated demand equation not only as a proxyfor prior permanent incomes but also as adeterminant of the discrepancies in the processof being corrected. In addition, it will permitlag patterns other than the simple exponentialkind we have used.

A Tentative Sketch of the MechanismTransmitting Monetary Changes

However consistent 'may be the relation be-tween monetary change and economic change,and however strong the evidence for the auton-omy of the monetary changes, we shall not bepersuaded that the monetary changes are thesource of the economic changes unless we canspecify in some detail the mechanism that con-

See Friedman and Meiselman, "The Relative Stabilityof Monetary Velocity."

nects the one with the other. Though ourknowledge is at the moment too meager toenable us to do this at all precisely, it may beworth sketching very broadly some of the pos-sible lines of connection, first, in order to pro-vide a plausible rationalization of our empiricalfindings; second, to show that a monetarytheory of cyclical fluctuations can accommo-date a wide variety of other empirical findingsabout cyclical regularities; and third, to stimu-late others to elaborate the theory and render itmore specific.

Let us start by defining an Elysian state ofmoving equilibrium in which real income percapita, the stock of money, and the price levelare all changing at constant annual rates. Therelation between these rates depends on whetherreal income is rising or falling, whether wealthis remaining constant as a ratio to income or isrising or falling relative to income, on the be-havior of relative rates of return on differentforms of wealth, and on the wealth elasticity ofdemand for money. To simplify, let us supposethat all interest rates in real terms (i.e., ad-justed for the rate of change in prices) and alsothe ratio of wealth to income are constant, sothat the wealth elasticity of demand for moneycan be approximated by the elasticity of de-mand for money with respect to permanentincome. If real income is rising at the rate ofa6 per year, the stock of money demanded willthen be rising at the rate of Sa,, per year, whereS is the income elasticity of demand for money,and prices will be rising at the rate of ap = aM —Sa,,, where aM is the rate of rise in the nominalstock of money per capita. For example, ifincome per capita is rising at 2 per cent peryear, the stock of money at 4 per cent a year,and S is 3/2, then prices would be rising at iper cent a year.37 If S and were to be thesame, and the stock of money were to rise at,say, io per cent a year, prices would be risingat the rate of 7 per cent a year; if the stock ofmoney were to be declining at io per cent ayear, prices would be falling at the rate of 13per cent a year.38

"These are roughly the actual values of a,,, ap, and aMover the go years 1870—1960 in the U.S. They yield a rathersmaller value of 8 than we estimate by multiple re-gression techniques (roughly i.8).

"It may seem strange that a i percentage point differencein the rate of change of the stock of money produces pre-

6o MONEY AND BUSINESS CYCLES

Let us now suppose that an unexpected riseto a new level occurs in the rate of change inthe money stock, and it remains there indefi-nitely — a single shock, as it were, displacingthe time path of the money stock. In tracingthe hypothetical effects of the higher rate ofgrowth of the money stock, there will be somedifference in detail depending on the source ofthe increase — whether from gold discoveries,or central bank open-market purchases, orgovernment expenditures financed by fiatmoney, or a rise in the deposit-currency ratio,or a rise in the deposit-reserve ratio. To bedefinite, therefore, let us suppose it comes froman increased rate of open-market purchases bya central bank.

Although the initial sellers of the securitiespurchased by the central bank were willingsellers, this does not mean that they want tohold the proceeds in money indefinitely. Thebank offered them a good price, so they sold;they added to their money balances as a tempo-rary step in rearranging their portfolios. Ifthe seller was a commercial bank, it now haslarger reserves than it has regarded before assufficient and will seek to expand its invest-ments and its loans at a greater rate than be-fore. If the seller was not a commercial bank,be is not likely even temporarily to want tohold the proceeds in currency but will depositthem in a commercial bank, thereby, in ourfractional reserve system, adding to the bank'sreserves relative to its deposits. In either case,therefore, in our system, commercial banks be-come more liquid. In the second case, in addi-tion, the nonbank seller has a higher ratio ofmoney in his portfolio than he has had hitherto.

Both the nonbank seller and commercialbanks will therefore seek to readjust their port-cisely a a percentage point difference in the rate of change ofprices regardless of •the magnitude of the rate of change ofmoney. Will there not, it is tempting to say, be a flight frommoney as the rate of change in prices and hence the costof holding money rises? The answer is that we are com-paring states of equilibrium, not the transition from onestate to another. In a world in which prices are rising at 7per cent a year, the stock of money will be smaller relativeto income (i.e., velocity will be higher) than it would in aworld in which prices are falling at 13 per cent a year. But,in both, velocity will be changing only in response to thechange in real income, which is by assumption the same inthe two worlds. Of course, it is possible that 6 is differentat different levels of cost of holding money; but that wouldbe an effect of a rather subtler kind.

folios, the only difference being that the com-mercial banks will in the process create moremoney, thereby transmitting the increase inhigh-powered money to the total money stock.The interposition of the commercial bank in theprocess means that the increase in the rate ofgrowth of the money stock, which initially wasless than in high-powered money, will for atime be greater. So we have here already amechanism working for some overshooting.

It seems plausible that both nonbank andbank holders of redundant balances will turnfirst to securities comparable to those they havesold, say, fixed-interest coupon, low-risk obli-gations. But as they seek to purchase thesethey will tend to bid up the prices of thoseissues. Hence they, and also other holders notinvolved in the initial central bank open-markettransactions, will look farther afield: the banks,to their loans; the nonbank holders, to othercategories of securities — higher-risk fixed-coupon obligations, equities, real property, andso forth.

As the process continues, the initial impactsare diffused in several respects: first, the rangeof assets affected widens; second, potentialcreators of assets now more in demand are in-duced to react to the better terms on which theycan be sold, including business enterprises wish-ing to engage in capital expansion, house build-ers or prospective homeowners, consumers whoare potential purchasers of durable consumergoods — and so on and on; third, the initiallyredundant money balances concentrated in thehands of those first affected by the open-marketpurchases become spread throughout theeconomy.

As the prices of financial assets are bid up,they become expensive relative to nonfinancialassets, so there is an incentive for individualsand enterprises to seek to bring their actualportfolios into accord with desired portfolios byacquiring nonfinancial assets. This, in turn,tends to make existing nonfinancial assets ex-pensive relative to newly constructed nonfinan-cial assets. At the same time, the general risein the price level of nonfinancial assets tends toraise wealth relative to income, and to make thedirect acquisition of current services cheaperrelative to the purchase of sources of services.These effects raise demand curves for current

MONEY AND BUSINESS CYCLES 6i

productive services, both for producing newcapital goods and for purchasing current serv-ices. The monetary stimulus is, in this way,spread from the financial markets to the mar-kets for goods and services.

Two points need emphasis at this stage. Thefirst is that the terms "financial markets,""assets," "investment," "rates of interest" and"portfolio" must, in order to be consistent withthe existing empirical evidence, be interpretedmuch more broadly than they often are. It hasbeen common to restrict attention to a smallclass of marketable financial securities and thereal capital it finances, to regard "the" rate ofinterest as the market yield on such securities,and the "investment" which is affected bychanges in the rate of interest as solely ormainly the items classified as "capital forma-tion" in national income accounts. Some of theempirical results summarized earlier are incon-sistent with this view.39 To rationalize theresults, it is necessary to take a much broaderview, to regard the relevant portfolios as con-taining a much wider range of assets, includingnot only government and private fixed-interestand equity securities traded on major financialmarkets, but also a host of other assets, evengoing so far as to include consumer durablegoods, consumer inventories of clothing andthe like and, maybe also, such human capitalas skills acquired through training, and the like.Similarly, it is necessary to make "rate of in-terest" an equally broad construct, coveringexplicit or implicit rates on the whole spectrumof assets.4°

The second point is to note how readily thesetentative lines on our sketch accommodate someof the documented regularities of businesscycles. The cyclical counterpart to our assumedinitial shock is the rise in the rate of growth ofthe mOney stock that generally occurs earlyin contraction. On the basis of the sketch sofar, we should expect it to have its first impacton the financial markets, and there, first onbonds, and only later on equities, and only stilllater on actual flows of payments for real re-sources. This is of course the actual pattern.The financial markets tend to revive well before

"In particular, those in Friedman and Meiselman, "TheRelative Stability of Monetary Velocity."

40 See ibid. for a fufler discussion of these points.

the trough. Historically, railroad bond priceshave risen very early in the process. Equitymarkets start to recover later but still generallybefore the business trough. Actual expenditureson purchases of goods and services rise still later.The consistent tendency for orders to lead actualpurchases would of course be expected on thistheory, but it would follow simply from themechanics of the production process. Hence itgives no definite support to this or any othertheory. It is simply a stage in the way any im-pulse, however generated, will be transmitted.The tendency for the prices of financial assetsto rise early in the pattern is quite a differentmatter. If the initial impulse were generatedby an autonomous increase in spending on finalgoods and services, it would be plausible toexpect the timing to be the reverse of what itactually is. Of course, on the theory beingsketched, the precise timing will depend on thesource of the initial monetary impulse. How-ever, under the banking structure of the UnitedStates and other financially developed coun-tries, whatever the initial impulse, commercialbanks will play a key role in transforming itinto an increased rate of growth in the moneystock, and this will impose a large measure ofuniformity on the outcome.

One other feature of cyclical experience thatour sketch may be able to rationalize and thatis worthy of special note is the behavior of thedeposit-currency ratio. The initial monetaryimpulse is concentrated among holders of fi-nancial assets and is then diffused to the rest ofthe community. But this means, as we havenoted, that the redundant balances are initiallyin the hands of asset holders with a high ratioof deposits to currency. As the redundant bal-ances are diffused, they spread to more nearlya representative group in the population. Con-sistently with this sequence, the ratio of depositsto currency starts to rise early in contraction,not very far removed in time from the troughin the rate of rise in the money stock; thedeposit-currency ratio continues to rise duringthe rest of contraction and early expansion butthen reaches a peak around mid-expansion, andfalls. The turning point, on this sketch, reflectsthe point at which the net tide of redundantbalances has shifted from the financial com-munity to the rest of the community.


To return to our sketch, we had reached thestage at which the demand for the services offactors of production was rising, which means,of course, a rise in money incomes. This willtend to be partly reflected in a rise of the pricesof resources and of final goods; at the sametime, the prices of nonfinancial assets will al-ready have been rising as demand shifted tothem from financial assets. These price risesthemselves tend to correct portfolios by makingthe real value of monetary assets less than theyotherwise would be. The result is to reduce therelative redundancy of monetary assets, whichsets the stage for a rise in the structure of in-terest rates in place of the prior decline. Theexact sequence of rises in prices, whether itaffects first prices of final products, and onlylater prices of factors and so shifts profit mar-gins — and so on — depends on the structureof the product and factor markets. Like therelation between new orders and production,this is part of the transmission mechanismcommon to all theories and tells little or nothingabout the generating impulse. This does notmean it is unimportant. On the contrary, itmay well determine the sequence of events oncethe stage is reached at which income is rising,as well as the time duration of subsequentreactions.

However, the important point for our pur-poses is very different. It is that the process wehave described will tend to overshoot the mark;it will not simply produce a smooth movementto the new path consistent with the new rate ofgrowth of the money stock assumed to prevail.There are two classes of reasons embodied inour analysis that explain why the process willovershoot. One, and in our view the more basictheoretically, has to do with the demand formoney. At the higher rate of price rise that isthe new ultimate equilibrium, the amount ofmoney demanded will be less in real terms thanit was initially, relative to wealth and henceincome. But this means that, in the process ofgoing from the initial to the new equilibrium,prices must rise at a faster rate than their ulti-mate rate. Hence the rate of price rise mustovershoot. This effect is reinforced by that em-bodied in the subsection, "Conclusions for Ma-jor Movements." In the initial stages of theprocess, money holders overestimate the extent

of monetary, redundancy, since they evaluatemoney stocks at unduly low levels of prices;they are slow, that is, to revise their estimatesof permanent prices upward, hence they initial-ly seek more radical readjustments in theirportfolios than will ultimately turn out to berequired. (If this analysis is applied to acyclical process rather than to our special caseof a shift from one moving equilibrium to an-other, a second element from that part of thesection would also enter to produce overshooting— a slow revision of estimates of permanentreal income.) The second class of reasons forovershooting has to do with feedback effectsthrough the monetary mechanism. Two ofthese have already been mentioned. First, theeffect of the initial assumed shock is to cause agreater rate of rise in high-powered money thanin the money stock as a whole. But since thereis nothing about the shock that will permanentlyalter the ratio of money to high-powered money,it follows that the money stock must for a timegrow faster than ultimately in order to catchup. Second, there is reason for the deposit-currency ratio to rise in the initial stages of theprocess above its long-run equilibrium level.In addition to these two classes of reasons forovershooting, which derive from the specificallymonetary elements in our sketch, there may ofcourse be those arising from the other elementsof the transmission mechanism common toalmost any theory.

The tendency to overshoot means that thedynamic process of transition from one equi-librium path to another involves a cyclical ad-justment process. Presumably, these cyclicaladjustments will be damped, though no merelyverbal exposition can suffice to assure that theparticular mechanism described will have thatproperty. Presumably also, the extent of over-shooting will not be negligible relative to thedistuthance, though again no merely verbalexposition can suffice to assure that the mech-anism described will have that property.

The passage from this analysis of a singledisplacement of the rate of growth of money toa monetary theory of partly self-generatingcyclical fluctuations is direct and has in largepart been embodied in the preceding statement.It may be worth noting, however, that itwould be rather more plausible to suppose a


shock to take the form of an unusually highor low rate of growth of the stock of money forsome time, with a reversion to a previous levelrather than a shift to a permanently new level.Such a shock is equivalent to two shocks ofthe kind we have been considering — butshocks in opposite directions. Hence the shockitself gives rise to a cyclical movement in addi-tion to the cyclical adjustment to each shockseparately. The fact that in the cycle thereis never that complete adjustment to the exist-ing state of affairs that is present in the as-sumed initial Elysian state of moving equilib-rium is of no decisive importance. It merelymeans that one state of incomplete adjustmentsucceeds another and that successive widen-ings and narrowings of discrepancies betweenactual and desired portfolios replace the intro-duction of a discrepancy and the correction ofit. As noted parenthetically earlier, of some-what more moment are the fluctuations in realincome and employment over the cycle, whichintroduce an important reason for overshoot-ing.

The central element in the transmissionmechanism, as we have outlined it, is the con-cept of cyclical fluctuations as the outcomeof balance sheet adjustments, as the effects onflows of adjustments between desired andactual stocks. It is this interconnection ofstocks and flows that stretches the effect ofshocks out in time, produces a diffusion overdifferent economic categories, and gives riseto cyclical reaction mechanisms. The stocksserve as buffers or shock absorbers of initialchanges in rates of flow, by expanding or con-tracting from their "normal" or "natural" or"desired" state, and then slowly alter otherflows as holders try to regain that state.

In this stock-flow view, money is a stockin a portfolio of assets, like the stocks of fi-nancial assets, or houses, or buildings, or in-ventories, or people, or skills. It yields a flowof services as these other assets do; it is alsosubject to increase or decrease through inflowsand outflows, also as the other assets are. Itis because our thinking has increasingly movedin this direction that it has become natural tous to regard the rate of change in the stock ofmoney as comparable to income flows and toregard changes in the rate of change as a gen-

erating force in producing cyclical fluctuationsin economic activity.

Summary

The statistical evidence on the role of moneyin business cycles assembled in the first sectiondemonstrates beyond any reasonable doubt thatthe stock of money displays a systematiccyclical behavior. The rate of change in themoney stock regularly reaches a peak beforethe reference peak and a trough before the ref-erence trough, though the lead is rather vari-able. The amplitude of the cyclical movementin money is closely correlated with the ampli-tude of the cyclical movement in general busi-ness and is about half as large as the amplitudeof cyclical movements in money income. Themost important single determinant, from thesupply side, of the cyclical pattern of moneyis the cyclical pattern in the division of thepublic's money holdings between currency anddeposits. The stock of money is much moreclosely and systematically related to incomeover business cycles than is investment orautonomous expenditures.

In the second section we suggested plausibleinterpretations of these facts, pointing out thatthe close relation tells nothing directly aboutwhether the cyclical changes in money aresimply a consequence of the changes in incomeor are in large measure the source of thosechanges. For major movements in income, weconcluded that there is an extremely strongcase for the proposition that sizable changes inthe rate of change in the money stock are anecessary and sufficient condition for sizablechanges in the rate of change in money income.For minor movements, we concluded that,while the evidence was far less strong, it isplausible to suppose that changes in the stockof money played an important independentrole, though certainly the evidence for theseminor movements does not rule out other inter-pretations. In the subsection, "Conclusions forMajor Movements," we formalized one ele-ment of a theory designed to account for theobserved tendency of cyclical fluctuations inincome to be wider in amplitude than cyclicalfluctuations in money are. The theory, plusearlier empirical work, yielded an independent


statistical estimate of what we call the moneymultiplier, or the ratio of the percentage changein income to the associated percentage changein the stock of money. The independent esti-mate was 1.84; the directly observed ratio2.0. This agreement does not reflect any com-mon statistical origin of the two estimates. Ittherefore suggests that further elaboration ofthe theory might be well worthwhile.

Finally, in the last section, we sketched inbroad strokes the kind of transmission mechan-ism that could explain how monetary changescan produce cyclical fluctuations in income, andthat is consistent with our knowledge ofeconomic interrelationships. The final plc-

ture that might ultimately develop out ofthis sketch could be of a partly self-generat-ing cyclical mechanism. Disturbances in therate of change in the money stock set in traina cyclical adjustment mechanism including afeedback in the rate of change in money itself.Additional disturbances from time to timewould prevent the fluctuations from dying out.The mechanism emphasizes the reciprocal ad-justment of stocks to flows, with money play-ing a key role as a component of the stock ofassets. We emphasize that this sketch is ex-ceedingly tentative and, of course, not pre-clusive. The mechanism outlined can be com-bined with other adjustment mechanisms.

COMMENTHYMAN P. MINSKY,

University of California, BerkeleyFriedman and Schwartz examine monetary

factors in economic fluctuations. Monetaryfactors are narrowly defined as ". . . the roleof the stock of money and changes in thestock." The authors cast "the 'credit' marketas one of the supporting players rather than astar performer." Apparently the plot waschanged after the show was cast for, with thepossible exception of the last section of thepaper ("A Tentative Sketch of the MechanismTransmitting Monetary Changes"), whereportfolio adjustments are introduced in a notvery precise way into the discussion of thechannels by which monetary changes lead tooutput changes, credit has a walk-on ratherthan a supporting role. Since the authors citeFisher as authority for the "dance of the do!-lar" view of business cycles, we can supposethey are familiar with and chose to ignoreFisher's "debt deflation" view of businesscycles in which the behavior of the capital andcredit markets is a star performer in generat-ing deep depressions.'

The proposition Friedman and Schwartz as-sert to be true is that the business cycles ofexperience — chronicled by the National Bu-reau of Economic Research reference cycles —

11. Fisher: Booms and Depressions, New York, 1932;idem, "The Debt-Deflation Theory of Great Depressions,"Econometrica, Oct. 2933, pp. 337—357.

can be explained in the sense of being causedby the behavior of the money supply. Accord-ing to their view, however, the monetary eventthat calls the tune for economic activity isneither the supply of money nor a change inthe supply of money; rather, it is a changein the rate of change in the supply of moneythat is the critical variable. The acceleratorview of the way in which monetary changesoperate to affect economic activity can be linkedto an accelerator formulation of business cyclebehavior quite easily by way of the need tofinance externally part of the induced invest-ment during periods of expansion.2 As the au-thors show no interest in integrating their ex-plorations of the supply of money with anincome-expenditure view of business cycles,this possibility is ignored.

The authors make their point of view clearwhen, in commenting upon the recent revivalof interest in monetary policy, they assert"this fair-weather source of support for the im-portance of money is a weak reed." Belief ina narrowly defined monetary explanation ofbusiness cycles is implicitly a matter of ideo-logical importance. Hence we can expect theauthors to engage in a hard sell of a point ofview. In the process of selling their viewpointthe authors invent "plausible interpretationsof the factual evidence" and "a plausible ra-

2 H. P. Minsky: "Monetary Systems and AcceleratorModels," American Economic Review, Dec. 2957, pp. 85g-.883.


tionalization of our empirical findings," theycite as supporting evidence both work in prog-ress and material not generally available, andthey ignore evidence that does not supporttheir views, such as the fiscal relations in the1937—38 recession and the success of short-runforecasting models which place little or noemphasis upon monetary phenomena.

The Friedman and Schwartz paper containsa number of ingenious studies and construc-tions, some of which I feel sure will prove tobe of lasting value. However, the argumentdoes not depend solely upon these studies; itis also based upon their reading of history andthe relations they derive between changes inthe rate of change in the stock of money asgiven by their time series (money to them al-most always includes time deposits in com-mercial banks) and the National Bureau's ref-erence cycle chronology.

Two classes of business cycles, mild anddeep depression cycles, are identified. Twenty-one business cycles are recorded as occurringbetween 1870 and 1960, of which six were deepdepression cycles. The mild depression cyclesare associated by the authors with increases anddecreases in the positive rate of growth of themoney supply, whereas the deep depressioncycles are associated with significant decreasesin the money supply.3 Two of the decreasesin the stock of money, those of 1919—20 and1937—38, can be identified as the result ofspecific actions by the Federal Reserve System.Neither of the administered decreases in themoney stock was associated with a financialcrisis or panic. In the other four deep depres-sion cycles a monetary-financial crisis waspart of the process by which the money supplywas decreased. An explanation which makesfinancial crises endogenous rather than exog-enous events seems to be required to completethe monetary explanation of business cycles.

To one who remembers Schumpeter'sKitchen, Juglar, and Kondratief cycles, thechronology of major and minor cycles in theFriedman and Schwartz paper recalls thephrase "3 Kitchens and then a Juglar," forthree mild depression cycles separated thedeep depression cycles following 1873, 1893,1907 and 1920, and two mild recession cyclesoccurred between the deep depression cyclesthat began in 1920 and 1929. The 1929 and1937 deep depression cycles were not separatedby any mild depression cycles. As the declinesin the money supply associated with the 1920and the 1937 contractions can be related toadministrative actions, the deviation from theSchumpeterian sequence in the post—WorldWar I period can be laid to the effects of inter-vention. The chronology of major and minorcycles lends some weight to the propositionthat the necessary conditions for a financialcrisis emerge as a result of the financial rela-tions that develop during an extended period inwhich growth is interrupted only by mild de-pressions. The events since 1920 indicate thatthe sequence of events leading to severe businesscycles can be modified by central bank action:the hope is that, with increased knowledge,lender-of-last-resort acts by the central bankcan prevent serious financial crises.

I have organized the body of my comments onthe Friedman and Schwartz paper around twotopics, the importance of m9ney and the trans-mission mechanism.

The Importance of MoneyFriedman and Schwartz argue that "money is

important." This slogan can be interpreted ina number of ways. The belief that money isimportant is not inconsistent with acceptanceof the basic validity of the modern income-expenditure approach to business cycles. Toone holding such a view, the nonexistent orprimitive monetary and financial system incor-porated in income and expenditure models suchas those of Duesenberry, Eckstein, and Fromm,of Klein, and of Suits is a defect that should

4J. S. Duesenberry, 0. Eckstein, and G. Fromm, "ASimulation of the United States Economy in Recession,"Econometrica, Oct. ig6o, pp. 749—809. There are manyKlein models, one of the most recent is L. Klein, "A Post-war Quarterly Model: Description and Applications," pre-

8 Friedman and Schwartz associate two "mild" declinesin the money stock (1948—49 and 1959—6o) with mild busi-ness contractions. Those mild declines were about one-halfas large as the smallest of the declines they consider to besignificant. On the other hand, the largest of the decreasesthey consider to be significant was about 15 times as largeas the smallest "significant" one (all decreases are measuredas a percentage decline in the money stock). The authorsdo not explain why the break between mild and deep de-pressions has been associated with the difference between1.3 per cent and 2.4 per cent declines in the money supply.


be corrected. Holders of this position with re-spect to the importance of money would arguethat serious work on monetary and financialinterrelations is needed in order to improveupon what we now know. Even though thecurrent models are adequate predictors of shortand relatively mild movements of income andso forth, they are deficient, because in theirpresent form they are not able to generate themore extreme movements characteristic of deepdepressions and inflations. The conjectureguiding research would be that the completionof these models, so that they could encompasssuch broader possibilities, would entail the useof a more sophisticated model of money andfinance: in particular relations which state howmoney and financial institutions behave, andhow portfolio adjustments and financial liabili-ties affect the various sectors would have to beincorporated into the more complete model.

Another interpretation of the phrase "moneyis important" would reject the basic validity ofthe income-expenditure approach. The ob-served path of money income and prices is in-terpreted as the result of shocks imposed uponan otherwise inherently stable growth processby random or systematic changes in either thenominal quantity of money or the rate of changein the money supply. In this view, the introduc-tion of a money supply that behaves in the cor-rect manner would eliminate either all or atleast a large part of those disturbances thatconstitute the major malfunctioning of an en-terprise system. The effects of changes in thestock of money and changes in the rate ofchanges in the stock of money are independentof the financial interrelations underlying themonetary system, as well as of the other finan-cial usages and institutions which exist. Basi-cally, the nominal supply of money calls thetune and money income and prices follow, per-haps with a lag.

Almost always Friedman and Schwartz seemto adopt the second interpretation of the im-portance of money and they thereby reject thevalidity of the income-expenditure approach tobusiness cycles. However, at times they do

pared for an NBER conference on Research in Income andWealth, 1962. D. B. Suits, "Forecasting with an Econo-metric Model," American Economic Review, Mar. 1962, pp.104—132.

come within hailing distance of the first inter-pretation, particularly when they assert:"clearly the view that monetary change is im-portant does not preclude the existence of otherfactors that affect the course of business or thataccount for the quasi-rhythmical character ofbusiness fluctuations. We have no doubt thatother factors play a role."

They go on to state:A fully satisfactory explanation of the minor move-ments would require an explicit and rigorously statedtheory, which could take the form of a series ofsimultaneous differential equations describing thereaction mechanism of the economy, together witha specification of the joint distribution function ofthe random disturbances impinging on it, and aspecification of the systematic disturbances that couldbe introduced into it. Our belief that money playsan important role in minor movements is equivalentto asserting that some of the differential equationswould contain the stock of money as a variable, thatdisturbances in the stock of money are among therandom or systematic disturbances impinging uponthe system, and that these disturbances alone wouldbe capable of generating a path for such major eco-nomic variables as money income, prices, output andthe like, comparable to the path they actually followduring mild depression cycles.

The first sentence of this quotation is of coursea statement of the content of the income-expend-iture prediction models. The conjecture that

- . these [monetary] disturbances alonewould be capable of generating a path .

comparable . . ." to the observed path couldbe interpreted as a hypothesis to be testedwithin a specific income-expenditure model. Ihappen to believe that the introduction ofmonetary and the related financial phenomenaonly by way of the stock of money as a variablein the equation system would turn out to benot tenable; that any serious work on monetaryphenomena in relation to business cycles willresult in the inclusion in the system of variablesand equations that reflect not only the moneystock but also the asset structure of themonetary authorities and the financial liabil-ities of other units. Nevertheless, the viewexpressed in the above quotation of the roleof money in determining system behavior iswithin working-hypothesis distance of the in-come-expenditure approach.

However, the dominant theme of Friedmanand Schwartz is that a complex simultaneous


equation system is not needed in order to under-stand the essential business-cycle properties ofthe economy.

There seems to us, accordingly, to be an extraordi-narily strong case for the propositions that (i) ap-preciable changes in the rate of growth of the stockof money are a necessary and sufficient condition forappreciable changes in the rate of growth of moneyincome; and (2) that this is true both for long secularchanges and also for changes over periods roughly thelength of business cycles. To go beyond the evidenceand discussion thus far presented: our survey of ex-perience leads us to conjecture that the longer-period changes in money income produced by a changedsecular rate of growth of the money stock are reflectedmainly in different price behavior rather than in dif-ferent rates of growth of output; whereas the shorter-period changes in the rate of growth of money stockare capable of exerting a sizeable influence on therate of growth of output as well.

The more modest view of the part played bythe stock of money and changes in the stockof money is taken when Friedman and Schwartzdiscuss minor movements; the stronger view istaken when they discuss major movements.However, they argue: "If money plays an in-dependent role in major movements, is it likelyto be almost passive in minor movements? Theminor movements can be interpreted as lessvirulent members of the same species as themajor movements. Is not a common explana-tion for both more appealing than separate ex-planations, especially when there is no well-tested alternative separate explanation?" Ifboth major and minor movements are to beexplained by the same process, then either bothare generated by a complex interdependentsystem in which nonmonetary factors are ca-pable of generating the observed movements, orboth are generated by random or systematicshocks from the supply of money. Another oftheir statements requires comment: "The casefor a monetary explanation is not nearly sostrong for the minor U.S. economic fluctuations

." and for minor movements they cannotrule out ". . . the possibility that the close re-lation between money and business reflectedprimarily the influence of business on money."It seems to me that Friedman and Schwartzreally have to choose between accepting a dualtheory, in which major fluctuations are some-thing different from minor fluctuations or, if

they insist upon a theory that covers both majorand minor cycles, they can accept a view thatfactors other than the narrowly monetary phe-nomena are important in generating both typesof business cycles.

I believe the adoption of the middle groundbetween the Friedman and Schwartz positionand the models that operate with a skeletalmonetary and financial system would be a mostfruitful research strategy. The working hypoth-esis should take the form of a model thatintegrates a more complete monetary and finan-cial system into an income-expenditure frame-work, in particular, financial commitmentsalong with financial assets should be integratedinto the various behavior equations. One aimof such an integrated model would be to explainwhat is admittedly an open aspect of the Fried-man and Schwartz narrow monetary interpre-tation of business cycles, which is that "Itleaves open the reasons for the change in therate of growth of the money stock and, indeed,at this point is highly eclectic, taking accountof the fact that historically there have beenmany different reasons." That is, an objectiveof such a more complete model would be toexplain not only money income but also howmonetary and financial crises are generated.

Elsewhere I have worked with a model inwhich the need to deficit finance private invest-ment during a period of sustained growth (aperiod in which economic growth is interruptedby only mild recessions) results in an ever-increasing portion of income flows being neededto service financial commitments.5 Such chang-ing relations between income receipts and pay-ments due to financial commitments generatean unstable situation in which a not abnormalinitial decline in income can explode, by way offinancial repercussions, into a major decline inincome. The not abnormal initial decline inincome, which triggers the large movement offinancial variables, is explained by a dynamicincome-flow model. In such a model the modestincome-induced changes in the rate of changein money, that give pause to Friedman andSchwartz when they assess the role of money in

'H. P. Minsky, "Financial Crises, Financial Systems andthe Performance of the Economy," to appear in PrivateCapital Markets, Vol. XVI, The Supporting Papers, Corn-mission on Money and Credit.


minor cycles, can be one of a number ofelements that may generate the not abnormalinitiating decline in income.

Once the financial repercussions of the needto deficit finance private investment are known,we are in a position to investigate whether thefinancial impact necessary to generate a severedepression can occur in the absence of a markeddecline in the money supply. Can a seriousdebt deflation occur even though the nominalsupply of money remains constant or expands?An implication of the Friedman and Schwartzexplanation of business cycles is that, even ifsharp declines in asset prices and net worthoccur owing to a financial crisis centeringaround the nonmonetary part of the financialsystem, no serious depression will take place,for they assert that without a large decline inthe nominal money supply no deep depressioncan occur.

The Transmission Mechanism

Friedman and Schwartz say, ". . . we shallnot be persuaded that the monetary changes arethe source of economic changes unless we canspecify in some detail the mechanism that con-nects the one with the other." The marketprocesses they specify as connecting the twoare consistent with the liquidity-preferencedoctrine as to how money affects income. Tothis reader, the connections they specify de-crease the difference between a monetary andan income-expenditure explanation of businesscycles, almost to the point where it is a matterof the specification of the interest elasticity ofvarious classes of expenditures.

An empirical finding of Friedman andSchwartz is that the changes in the rate ofchange in the money supply have a long andrather variable lead over the business cyclepeaks and troughs. In order to generate suchlags, the authors believe that the connectionbetween money and demand must be complex;the mechanism they sketch requires that variousfinancial and real markets be affected by reper-cussions of the initiating monetary change be-fore income responds. Any process that resultsin such long and variable leads can also resultin considerable slippage between monetarychanges and the presumably resultant changes

in income. As Keynes put it, "If however weare tempted to assert that money is the drinkwhich stimulates the system to activity, wemust remind ourselves that there may be sev-eral slips between the cup and the lip" (TheGeneral Theory of Employment, Interest andMoney, New York, 1963, p. 173).

The relevant question in regard to the sig-nificance of monetary changes is whether thelag (and slippage) of income changes followingchanges in the money stock (or its rate ofchange) is so large and variable that it is notfeasible, under all circumstances, to operate onthe money stock in order to achieve desiredchanges in income. But the feasibility of usingchanges in the money supply to affect incomedepends upon the channels through which amonetary change affects income. (In what fol-lows I shall be examining an increase in themoney supply, or in its rate of change: theargument for a decrease is quite similar asidefrom the possibility of a liquidity crisis.)

Two models of how bank operations changethe supply of money can be distinguished: thesecan be labeled the open market and the com-mercial loan 6 models. Open market operationsby banks and monetary authorities result inchanges in the money supply due to a substitu-tion of money for income-earning assets inportfolios. Aside from the effects of interestrate changes, no change in the total value ofthe assets and the financial liabilities of theother sectors result. Commercial loan opera-tions result in changes in the money supply dueto acquisition by the banks and authorities ofnewly created financial liabilities of other sec-tors. The consolidated financial assets andliabilities of these other sectors change withthe money supply.7

In addition to the open market and com-mercial loan routes to a change in the moneystock, the money stock can be changed by theproduction of the monetary commodity (eitherby mining or international trade) and the crea-tion of fiat money. Aside from the creation of

6 am not happy with this label.'Commercial loan operations include, not only business

lending by banks and rediscounting by the monetaryauthorities of business paper, but also acquisition by banksand authorities of corporate and government bonds, mort-gages, etc., either directly from the issuer or from a financialintermediary which functions as an underwriter or broker.


fiat money to pay off government debt — whichas far as I know has not taken place in theUnited States since — the link betweenchanges in the money supply by these tech-niques and changes in demand is direct.8 Hencethese ways of changing the money supply arein terms of their direct effect upon income quiteanalagous to commercial loan operations; theanalogy breaks down, in that no increase inprivate debt accompanies these changes.

Throughout most of the period under con-sideration the United States was on a goldstandard—commercial loan monetary system;open market operations were utilized by com-mercial banks and the central bank for onlyrelatively short periods, when the demand fornew financing was not sufficiently vigorous toutilize the commercial banks' ability to acquireassets. Since World War II the United Stateshas been on a mixed open market operations—commercial loan monetary system, in which theFederal Reserve System has affected reservesof commercial banks by purchasing govern-ment debt from the market and, on the whole,the commercial banks have acquired newlycreated private debt. (Commercial banks haveengaged in open market purchases duringrecessions and open market sales and com-mercial loan operations during expansions. Theresulting trend has shown a net change in com-mercial bank liabilities, offset by assets whichwere acquired as the result of commercial loanoperations.) The only recent period in whichboth monetary institutions operated mainly byway of open market purchases was followingthe 1929 crash (although the increasing nation-al debt and the various silver purchasing actsgave the operations some of the characteristicsof commercial loan operations).

Friedman and Schwartz state: "The mostimportant [evidence in support of the mone-tary explanation of business cycles] is thefact that the relation between money and busi-ness has remained largely unchanged over aperiod that has seen substantial changes in thearrangements determining the quantity ofmoney. During part of the period, the United

8An increase in the money supply by means of a goldflow that results in portfolio investment within the recipientcountry by the gold importer — such as occurred during thehot money epoch of the x93o's—is analogous to an openmarket rather than to a commercial loan operation.

States was on an effective gold standard, duringpart, on an inconvertible paper standard withfloating exchange rates, during part, on a man-aged paper standard with fixed exchange rates.The commercial banking system changed itsrole and scope greatly. The government ar-rangements for monetary control altered, theFederal Reserve System replacing the Treasuryas the formal center of control. And the cri-teria of control adopted by the monetaryauthorities altered." The above changes aremostly irrelevant to the relation between mon-etary changes and aggregate demand; they areprimarily legal and organizational. (I do notknow what Friedman and Schwartz mean by"The commercial banking system changed itsrole and scope greatly.") Throughout thatperiod, the largest part of the increases in themoney stock was associated with the acquisitionby the banks of newly created financial assets;almost always, the largest portion of thesefinancial assets were the liabilities of privateunits; and almost always these financial assets,acquired in exchange for net changes in themoney supply, were linked with investment ex-penditures or the holding of capital goods orfinancial assets. No matter how substantial theinstitutional and organizational changes havebeen, they have not seriously affected the eco-nomic relations underlying a change in themoney supply: the largest part of the changein the money stock throughout the period wasthe result of commercial loan operations bycommercial banks.

Following open market operations, the chan-nel connecting the monetary change and achange in income involves portfolio adjustmentsand changes in relative prices. In order togenerate an increase in private-sector demand,either the price of services must fall relative tothe price of assets, which may lead to an in-crease in consumption demand; or the price ofsecondhand nonfinancial assets must rise rela-tive to the price of new assets, which may leadto an increase in investment demand. Whathappens to the demand for services depends notonly upon what is happening to the interest ratebut also upon what is happening to consumers'disposable income or, if you wish, to consumers'views as to what is happening to their "perma-nent" income. The rise in the price of second-


hand nonfinancial assets relative to new non-financial assets will generate an increase ininvestment only if, after the increase, the priceof secondhand assets is not appreciably lowerthan the price of new assets. Given that therate of decline in asset prices during the initialstages of a serious depression is greater than therate of decline in money wages, a sharp fall inincome will result in a price level of secondhandassets that is low relative to the production costof new assets. Hence, a significant rise in in-vestment demand cannot be expected to followclosely upon a modest rise in the relative priceof used as compared with new assets such aswould result from a fall in interest rates. Therecovery of investment in such circumstanceswill follow upon the more marked rise in therelative price of used as against new assets,owing to the effects of the elimination of excesscapacity by way of depreciation and obsoles-cence, all of which operate quite independentlyof interest rate changes. That is, the relevantinvestment demand functions will not be inter-est elastic but will be responsive to factors thatfit naturally into an income-expenditureframework °

Following commercial loan operations by themonetary authorities and commercial banks,the connection between a change in the moneysupply and a change in income is direct. Theunits which generate the newly created liabili-ties acquired by the monetary system will doso because they wish to purchase either finan-cial or real assets. The portfolio adjustmentportion of the connection between a monetarychange and a change in income is not operativein the initial stages of the adjustment followinga commercial loan operation. The channel be-tween an increase in money and a net increase

° Friedman and Schwartz trace the effects of aninitial open market operation, their starting point is "anElysian State of moving equilibrium." From such an initialcondition the change in relative prices associated with in-terest rate changes will have an effect upon demand. If theinitial conditions include large price differences between themarket price of used and the production costs of new capitalgoods, such as can be expected to exist after a sharp fall inincome, then as has been argued in the text, the relativeprice changes following an open market operation increasein the money supply, will not have an appreciable effectupon demand. We therefore have a world in which thereaction to a monetary change depends upon the initialconditions; which is one of the major attributes of theKeynesian view.

in the money value of financial and real assetsheld by the borrowers is direct. As we are con-sidering a net increase in assets, unless themonetary change is wholely absorbed by a risein asset prices, additional real or financial assetsmust be created in exchange for the moneyspent by the initial borrowing unit. The crea-tion of new real assets is of course investment.Typically, a monetary change resulting from acommercial loan operation enables the borrow-er to make a real or financial market "demand"effective. The route connecting monetary andincome changes is direct; the lag between themonetary change and the initial change indemand will be short and the slippage slight.Of course the full impact of the monetarychange will be lagged behind the initial in-crease in the money supply, but this is fullyconsistent with the multiplier process.

Given that a strong demand for funds tofinance investment exists and that a portion ofthe supply of such funds has been the resultof monetary changes, a decrease in the rate ofincrease of the money supply will result in anincrease in the pressure to finance investmentfrom other sources. In addition to the net flowof savings and the net debt acquisition by themonetary system, the substitution of income-earning assets for money is a source of financ-ing for investment. Velocity increasing—liquidity decreasing portfolio changes will fol-low a decrease in the rate of increase in themoney supply. This is consistent with theFriedman and Schwartz findings that velocityconforms to the reference cycles (Chart 6),whereas the rate of change in the money supplyleads (Chart 4) the reference cycles.

Of course the commercial loan model of howthe money supply changes begs the question ofwhat determines the strong demand for fundsto finance investment. Friedman and Schwartzimplicitly assume that a strong investment de-mand exists when they start from a position ofmoving equilibrium, and they also assume thatthis investment demand is elastic with respectto relative interest rates. To an income-expend-iture theorist, the existence of a strong demandfor funds to finance investment to which a com-mercial loan monetary system responds requiresexplanation. On the basis of the observed over-all success of the economy, it is easy to assume


that the flow of technical progress and the feed-back from past demands has been on the wholesufficient to generate a buoyant investment de-mand. However, a corollary of the view thatnonmonetary factors have generated a stronginvestment demand throughout most of theperiod under consideration is that the incomeand financial past of an economy can lead to aweak investment demand, and a temporary"saturation" of capital can exist.1°

Friedman and Schwartz, in their example,specify that the initial change in the rate ofchange in the money supply is the result ofcentral bank open market operations. Eventhough they recognize that". . . whatever theinitial impulse, commercial banks will play akey role in transforming it into an increasedrate of growth in the money stock . . .," theirtransmission mechanism operates mainlythrough portfolio changes which are associatedwith changes "covering explicit or implicit(interest) rates on the whole spectrum ofassets." They pay no attenton to the nature ofthe assets acquired by the commercial banks.A transmission mechanism which operates byway of financial markets and portfolio changesis fully consistent with the income-expenditureapproach, but the effect upon income thatFriedman and Schwartz state exists dependsupon a strong reaction to small differences inrelative interest rates.

Even though the commercial loan model ofhow the money supply changes yields a moredirect linkage between monetary changes andincome changes than the open market opera-tions model, it has properties that are incon-sistent with the narrow monetary view of busi-ness cycles that Friedman and Schwartz es-pouse. The link between business debt and themoney supply is direct in the commercial loanview of the monetary system: hence a feedbackfrom debt to investment demand is an en-dogenous repercussion in the money creationprocess. In addition, the combination of astrong investment demand, a money supplylinked with the financing of private investment,the additional possibility of financing invest-

'° Saturation is really not an accurate label for a situationduring a severe depression, for the excess supply of capitalgoods is the result of the fall of incomes substantially belowpast incomes. Once demand recovers, such "saturation"vanishes.

ment by velocity increasing—liquidity decreas-ing portfolio changes, and the effect uponequity and real asset prices of a long run with-out a major recession can easily lead to a viewthat deep depressions are the result of thedevelopment of an unstable debt structure dur-ing periods in which the economy is a success.Instead of autonomous monetary changes lead-ing to deep depressions, the monetary changesare a result of the debt deflation process ofwhich Fisher wrote, and the debt deflation isthe result of an unstable debt structure gener-ated in the process of financing investmentduring a period in which sustained growth isinterrupted by only mild recessions.1'

Incidentally, Friedman and Schwartz do nottry to explain the variability of the length ofthe lag between a monetary change and thepresumably related income change. I conjec-ture that an explanation of this phenomenawould center around the actual mix of opera-tions by which the money supply is changed.This conjecture is inconsistent with the Fried-man and Schwartz view that money is moneyfor, if it is true, the effect of a change in thequantity of money or its rate of change doesdepend upon both the process by which themoney supply is changed and the assets ac-quired by the money system.'2

Thus, whether we take an open market or acommercial loan view of how the money supplyis changed, the market processes which followare not always satisfactory from the authors'point of view. With an open market monetarysystem the required investment and consump-tion reaction will not always take place; with acommercial loan monetary system the develop-

"There may be equilibrium which, though stable, isso delicately poised that, after departure from it beyondcertain limits, instability ensues, just as, at first, a stickmay bend under strain, ready all the time to bend back,until a certain point is reached, when it breaks. Thissimile probably applies when a debtor goes "broke," orwhen the breaking of many debtors constitutes a "crash,"after which there is no coming back to the original equilib-rium." I. Fisher, "The Debt-Deflation Theory of GreatDepressions," Econometrica, Oct. 1933, p. 339.

12 The authors recognize that the effects of a higher rateof growth of the money supply will differ in detail "de-pending on the source of the increase," but the sourcesthey mention are sources of commercial bank reserves orare determinants of the efficacy of a given amount of re-serves. They never examine the effects of the assets ac-quired by commercial banks and monetary authorities onfuture demand.


ing private debt structure can feed back uponfuture demand for financing. In addition, acommercial loan monetary system is consistentwith a debt deflation view of how major reces-sions are generated: a view in which the his-torically observed changes in the money supply,particularly those associated with deep depres-sion, are a result of business behavior.

Conclusion

I have not commended Friedman andSchwartz for the many fine and nice points intheir paper. I fully expect that much of theirwork will become a part of the body of knowl-edge that all workers in money will accept asvalid. Nevertheless, I feel they have failed tomake a convincing case for the strong view thatmonetary changes fully explain observed busi-ness cycles. If all they wish to assert is that afuller integration of monetary phenomena intothe basically income-expenditure models wouldimprove these models, then I am sure that thereis wide agreement with Friedman and Schwartz.But if this more modest view is their positionthen I wonder what all the shouting is about.

ARTHUR M. OKUN,Yale University

"Money and Business Cycles" is an impor-tant addition to previous studies by Friedmanand his associates on the relationship betweenthe stock of money and the level of economicactivity. The empirical findings offered byFriedman and Schwartz are impressive. No-body can justly quarrel with their claim thatthe paper "demonstrates beyond any reason-able doubt that the stock of money displays asystematic cyclical behavior." But, as the au-thors point out with care and clarity, theseempirical results are just as consistent withthe view that changes in aggregate activityinfluence the money supply as they are with theposition that changes in the money supply af-fect economic activity. While we can all agreethat the lines of causation run in both direc-tions, the key conclusion of their paper is thatthe causality runs substantially from money toincome, substantially enough to warrant amonetary theory of business cycles. Indeed,

the monetary view of economic fluctuationsstands as the principal characteristic of Fried-man's extensive research in this area.

To state my own principal conclusion di-rectly, I do not agree with the Friedman andSchwartz appraisal of the importance of money.I find their view of the world fascinating andstimulating, but I am not converted. So far,this rejection stands merely as a testament ofmy faith. I shall use these comments to explainmy misgivings about their theoretical position.

The Money Multiplier

In effect, Friedman and Schwartz ask me tobelieve that a $i billion open market purchaseof government securities will raise annualGNP by more than $20 billion. The $i billionrise in high-powered money should increasetotal money by more than $5 billion, or morethan 2 per cent. (That is my estimate — notthat of the authors — but I would not expectserious argument.) In turn, according to theFriedman and Schwartz estimate of 2 for theelasticity of net national product with respectto money, this expansion of the money supplyshould raise national product by more than 4per cent — or more than $20 billion.

This money multiplier estimate has ex-tremely significant implications for economicpolicy. It brings monetary policy to the foreand pushes fiscal policy into the background.To see these implications concretely, considera combined fiscal-monetary action whereby areduction in personal taxes of $x billion isfinanced entirely by the creation of new high-powered money. This action may be dividedinto two steps. The first step consists of apurely fiscal action: taxes are cut by $i billionand the reduction is financed by an issue ofinterest-bearing marketable government se-curities. The second step is then a purelymonetary action: the central bank buys the$i billion of government securities for cash onthe open market. The combined multiplier-accelerator effects on GNP of the fiscal actionmight be estimated anywhere from $1.5 to $5billion, but nobody's estimate of that stepwould approach the $20 billion attributed tothe monetary action, in line with Friedman andSchwartz. To follow them, one must believe


that the monetary step generates 8o to 90 percent of the total stimulus in the combinedaction.

Under typical conditions in the Americaneconomy, I would be inclined to reverse theallocation, putting 8o to 90 per cent of theweight on the fiscal step. I should expect atax rebate that came to its recipients in theform of readily marketable government secur-ities (the equivalent of the purely fiscal action)to be considerably more than half as expan-sionary as an equal rebate in the form of cash.In the world we inhabit, money substitutes areplentiful and fill the spectrum between moneyand capital goods. Government securitiesshare many of the properties of cash and fewof the characteristics of capital. The form ofthe increase in the public's net worth (as be-tween money and Treasury securities) seemsless vital to aggregate demand than the fact ofthis increase generated by the fiscal action.

If Friedman and Schwartz are right, the na-tion is going far astray in current discussionsof tax reduction, when we can get all thestimulus we need from moderate shifts inmonetary policy toward greater ease. Theirverdict on the power of money would not beaccepted by the makers of monetary policy,their most ardent critics, or the majority ofmonetary economists. Even the original archi-tects of quantity theory expected only half asmuch bang from a new buck when they tookproportionality of money and income as a firstapproximation. The distance of the Friedmanand Schwartz position from most professionalthinking on stabilization issues does not provethey are wrong. Collective professional judg-ment has been in error before, but I amorthodox enough to believe that it deservessome credit.

Money and Permanent IncomeThe quantitative estimates of large monetary

effects on income are vital to the Friedmanand Schwartz qualitative view that money ac-counts substantially for fluctuations in income.Percentage fluctuations in the path of moneyover cycles are considerably smaller than thepercentage variations in income. The rela-tively stable money variable can be responsible

for these wide movements in income, only ifmoney operates on income with great leverage.Friedman and Schwartz account for theirlarge estimates of leverage through the perma-nent income formulation of the demand formoney. The short-run income elasticity ofdemand for money is found to be low because(i) the demand depends on permanent income,and (2) permanent income varies from yearto year by only an estimated one-third asmuch as aggregate measured income. Then,on the assumption that any imbalance betweensupply and demand for cash is equilibratedby changes in measured income, a change inthe money supply must induce magnified in-come changes.

If I translate from permanent income towealth — just because it is a more familiarlanguage to me for some purposes — I cansee some point to Friedman's formulation ofthe demand for money. While I should expectthe transactions demand for money to contractalong with measured income in a recession, theasset demand for money will not decline solong as wealth is growing. Except for the mostsevere depressions, wealth in real terms haskept growing through all phases of the cycle.But the market valuation of wealth in port-folios is presumably most important to theasset demand for money, and I doubt that themarket value of wealth has fared any betterthan incomes have in recessions. If my guesson wealth values is correct, I see no reason fora particularly low short-run income elasticityof demand for money.

In Friedman's view, the short-run elasticityis low because transitory income is not usedto increase money holdings. Previously, wewere told that transitory income does not gointo current consumption outlays. I will feelmore comfortable about both those proposi-tions when I am told how and where transitoryincome does get allocated.

In sharp contrast to the low short-run in-come elasticity of demand for money, theirestimate of the long-run elasticity is r.8. Isee no logical reasons for money to be such aluxury — with an income elasticity rivallingthat of steak — and I am not prepared to ac-cept the finding as a structural characteristicof demand. Since time deposits of commercial


banks are the component of Friedman andSchwartz money that contributes most to thelarge coefficient, I wonder whether the resultmay stem from institutional developments af-fecting the supply conditions for time depositsby commercial banks.

The Transmission Mechanism

How does a rise in the money supply in-crease the level of income? Obviously, peoplewho have been induced to part with an earningasset and to take on cash cannot simply decideto get themselves more income and more wealthin order to restore balance to their portfolios.Some adjustment and transmission mechanismmust be involved.

I was surprised to find that the transmissionmechanism described by Friedman andSchwartz is so similar to the one I visualize.In their view and mine, money is one ingredientof a general equilibrium theory of asset hold-ings. Changes in the stock of money willproduce waves and ripples that influence thedemand for all assets, extending to newlyproduced goods and services. When the mone-tary authorities induce people to part withbonds, these people are not really demandingextra cash to hold. They absorb it as a tem-porary matter before turning to other earningassets. When the resulting expansion in the de-mand for nonmonetary assets gets reflected inan increased demand for reproducible "capital"and its services, it stimulates either productionor prices or both. But there are many linksin this chain of asset adjustments runningfrom cash to capital, and I would expect someof them to be rather weak links much of thetime.

If monetary factors have the leverage on in-come that Friedman and Schwartz expect, theymust have a better fulcrum than the generalequilibrium asset model implies. If I acceptedtheir empirical conclusions, I should be look-ing carefully at the mechanics of "creditavailability." At any time, some firms andhouseholds are prevented from acquiring allthe goods and services they would like becauseof their limited ability to borrow. Their port-folios are not in balance and they remainhungry for more capital (or more dissaving).

If an expansion of the money supply channelsmore funds into the hands of these eagerspenders, it can have a large and direct effecton aggregate activity. We do not know thatmonetary expansion is accompanied by greaterease of credit in this manner, but this disequilib-rium. mechanism deserves exploration.

The Role of Interest Rates

The strength of the links in the money-capital chain depends, in large measure, on theimportance of interest rates as equilibratorsof portfolios. In relying on the general equilib-rium fulcrum, Friedman and Schwartz contendthat changes in the rates of return on earningassets do not affect substantially the desiredratio of money to total wealth (or to permanentincome). In that case, output and prices arelinked rigidly and inexorably to the moneysupply. The money-income relationship isthen the key determinant of the time path ofthe system, playing the same dominant roleas the capital-output relationship does in Har-rod's growth model.

If, however, changes in the opportunity costof holding money can make people hold sig-nificantly more cash in their portfolios, thechain is weakened and money is no longer anirresistible force. Over the long run, the econ-omy can move along any one of a large num-ber of possible money-income paths so long asthere are compensating changes in rates ofreturn. In the short run, too, changes in yieldsgive flexibility to the economy, preventing thefull force of a change in the money supplyfrom impinging on the demand for output. Ifvariations in the prices of earning assets putportfolios back in balance, then the output ofcapital goods will be less affected by a chang-ing money supply.

The Friedman-Schwartz conclusion on thesmall role of interest does not rest firmly oneither theoretical or empirical evidence. I amnot satisfied with their discussion of Latané'sresults. Latané excludes time deposits of com-mercial banks from his measure of money;Friedman and Schwartz offer the opinion thatdifferent definitions of the money supply maygive totally different results on the role ofinterest rates. They cannot afford to stop

there. We need to know which set of resultsoffers the more appropriate guide for estimat-ing the economic impact of conventional ac-tions of monetary policy. Furthermore, Fried-man and Schwartz seem obliged to explainhow time deposits in commercial banks can bea close substitute for demand deposits and yetnot be closely substitutable for savings bankdeposits, savings and loan shares, and others.This is one example of a repeated puzzle intheir analysis: asset prices and returns playa key and oft-cited role in encouraging sub-stitution among nonmonetary earnings assets,but somehow do not influence the demand forthose assets which are counted as money.

Nor can I accept the procedures in Fried-man's quantitative work on the demand func-tion for money. There, other variables get thefirst chance to explain demand while interestrates wait in line and are given only the op-portunity to eat the leftovers, the residuals ofthe basic equations. Friedman and Schwartztell us they are treating money as they wouldany other stock of assets, such as houses. Ifeel confident they would give much moreprominent treatment to the prices of housesand rental services in estimating demand forthat stock than Friedman has given to theopportunity cost of money.

The Direct Link

The Friedman and Schwartz monetary viewof fluctuations is sharply distinguished fromthe investment approach. Suppose monetaryimpulses were the principal cause of changesin investment demand and that these invest-ment changes in turn induced fluctuations inconsumption and additional responses in in-vestment through a multiplier-accelerator proc-ess. Then, economic fluctuation could beequally well described either as a monetary oran investment phenomenon, and presumablywould be best described as a money-through-investment matter. But they argue that thedirect route from money to income offers amore satisfactory explanation than the paththat travels from money to interest to invest-ment to income.

In espousing the direct linkage of money toincome, Friedman and Schwartz are armed

with the powerful empirical results of theFriedman and Meiselman study. There, it wasshown that aggregate consumer expenditurehas historically been much more closely relatedto the money supply than to the level of"autonomous expenditures," defined as netprivate investment plus the government def-icit. I have some reservations about the tests,particularly the treatment of the governmentdeficit as autonomous. I also find some of theresults puzzling: for example, (i) the "moneymultiplier" is much smaller than Friedman'sother work implies; (2) surprisingly, moneyexplains consumption better than it explainsinvestment. Furthermore, the results maysimply mean that changes in the money supplycan be rather accurately described as "meet-ing the needs of trade." With all these lines ofdefense, I would be less than candid if I dis-missed the findings. Had I known what vari-ables were going to be correlated, I wouldhave been willing to bet my nickel on the sidethat turned out to be the decisive loser. Untiland unless there is a satisfactory explanationof these results, the unconverted cannot resteasy.

On the other hand, I doubt that Friedmanand Schwartz can feel comfortable with theirexplanation of the direct linkage. They pointout that the direct influence of money extendsbeyond the items classified as investment inthe national accounts, impinging on consumerdurable goods, consumer inventories of itemsclassified as nondurable, on the rental servicesof capital goods, and perhaps on investmentin human capital. If it were possible to adjustthe national accounts to get a better measureof capital items, this strategy would be indi-cated. But the current-capital distinction is ablurred one and many items may have bothcapital and current aspects that defy classifica-tion. Hence, in their view, precision may belost rather than gained by tracing from moneyto any concept of capital items before movingalong to the impact on income.

Certainly, any line separating current fromcapital items must be a fuzzy one. I shareFriedman's view that major consumer durablegoods belong in the capital account and do notfind them a serious conceptual problem. Else-where, the range of fuzziness seems small



relative to the magnitude of total investmentexpenditures. The doubtful items listed byFriedman and Meiselman and by Friedmanand Schwartz look rather trivial to me. Thebattle between velocity and the multiplier ismost unlikely to turn upon the impact ofmonetary changes on pantry and boudoirstocks held by households or on private educa-tional expenditures, and even less likely tohinge on the substitution of auto rentals fornew car purchases. In general, the difficultiesof drawing a line between capital and currentoutlays seem less serious than those associatedwith dividing liquid assets between money andnonmoney.

The Cyclical MechanismThe monetary cycle outlined by Friedman

and Schwartz is monetary both in its impulseand its overshoot mechanism. In their ex-ample, which is explicitly not meant to be pre-clusive, an increase in the money supply startsthe fluctuation. The cyclical character of thefluctuation comes from an overshooting mone-tary cobweb: as prices and incomes adjust tothe expansionary impact of the monetarystimulus, people have to scramble back forsome of the cash they unloaded initially. Pre-sumably, the authors prefer the double mone-tary version of the cycle, but it seems worthnoting that they could be "half right." Dis-turbances of a nonmonetary origin could pro-duce cyclical movements through the Fried-man and Schwartz monetary cobweb; alter-natively, disturbances of a monetary charactercould be the usual shocks in a world where theovershoot mechanism was not monetary.

At first glance, a monetary explanation ofbusiness cycles seems impossible because themoney supply does not decline in many cyclicalrecessions of economic activity. However,Friedman and Schwartz meet this challenge.They advance a monetary accelerator hypoth-esis whereby a slowdown in growth of themoney stock can produce an absolute down-turn in income. The workings of the monetaryaccelerator are not explained lucidly by Fried-man and Schwartz. It would have been help-ful if they had included a verbal description ofa cycle that was initiated by a reduced (butcontinued positive) growth rate of money.

The following model has some of the char-acteristics of a monetary accelerator, and I offerit because it may resemble the Friedman andSchwartz view:

i. The stock of money determines the de-sired stock of capital, in line with some rigiddesired portfolio balance.

2. Then, the growth rate of money fixes thegrowth rate of desired capital.

3. Investment depends on the excess of de-sired capital over actual capital.

In a moving state of exponential equilibrium,the percentage excess of desired over actualcapital must be constant. Now, if the growthrate of money declines, the growth rate of de-sired capital falls, reducing the excess of de-sired over actual capital. As a result, invest-ment declines absolutely. This is potentiallya cycle model; but, as I see it, the overshootmechanism lies in the way investment adjustsactual to desired stocks. A slowdown in mone-tary growth can be the initiator of recession inthis world, but it does not look like a doublemonetary version of the cycle to me. I lookforward to a more explicit formulation of theFriedman and Schwartz cycle model in theirfuture work.

ConclusionI consider Friedman's works on money a

major challenge to the unconverted. They arefull of stimulating ideas and significant factsand parameters, and they offer some explana-tion for all the empirical findings. While Iprotest against a few of the empirical proce-dures, the research techniques generally com-mand praise and admiration. I find many ofthe results and their explanation paradoxicaland implausible, yet I cannot really account forthem on other grounds. So I feel very uneasy.I wonder whether I can appropriately holdstrong intuitive views about relationships andparameters I have never estimated. I wonder,on the other hand, how Friedman can disregardall intuitive evidence.

In reviewing the Friedman and Schwartzpaper, I feel as though I am commenting ona highly competent and comprehensive de-mand study with some astonishing results:coffee and tea are found to be independentcommodities while coffee and champagne are


close substitutes; bus rides have an estimatedincome elasticity of i .8; the knowledgewhether a family owns an auto offers no as-sistance in explaining its outlays for gasoline.But the values of parameters in aggregativerelationships have far more consequence forthe direction of economic policy and economicresearch than any coffee-tea examples. Thedisciples and the critics of this school of mon-etary analysis have a social and a scientificresponsibility to pursue the issues with inten-sive empirical and theoretical research.

CLARK WARBURTON,Federal Deposit Insurance Corporation

The conclusions of Milton Friedman andAnna Schwartz regarding the role of moneyin business cycles should be no surprise toanyone who has surveyed the history of busi-ness-fluctuation theory. The basic cause ofthe phenomena described by economic writersof the past as financial panics, commercialcrises, revulsions, and depressions has been asubject of intense controversy and widely di-verse opinions. However, the most continuousthread running through the controversy hasbeen the idea that misbehavior of the monetaryand banking system, in the sense of monetarydevelopments or central bank policy inducingcontractions in the circulating medium, is thedominant causal element in the generation ofbusiness downswings of any substantial size.Also, operations of the banking system in thedirection of substantially expanding the stockof money have almost universally been re-garded as the basic element in inflationarybusiness booms, and particularly the great in-flations of history, recognizing, of course, thatexigencies of government finance have beena fundamental force beneath an expansion ofthe circulating medium in time of war.

The economic literature focusing on changesin the stock of money as the dominant causalforce in the generation of major businessfluctuations goes back in time throughout andfar beyond the ninety years covered by theproject on which Friedman and Schwartz areengaged, though no factual study has pre-viously been made covering so long a span oftime with such thoroughness and competent

analysis. However, the previous literaturecontains many descriptions by contemporaryobservers of the sequence of events and ofthe character of causal relationships. Amongthese observations is an emphasis on channelsof influence which are treated lightly in theauthors' sketch of the mechanism transmittingmonetary changes. The theory of the impactof changes in the stock of money on businessactivity, developed over a period of two cen-turies before publication of the GeneralTheory, recognized the role of changes ininterest rates (in the writings of Irving Fisherand Knut Wicksell, for example) but, in gen-eral, gave more attention to the process bywhich successive uses of new units of moneycoming into the economy had an impact onprices and on business activity, and to thecorresponding process when the circulatingmedium was contracted. In that literaturethere was also much emphasis upon businessanticipations and expectations, related to ob-served factors influencing or expected to in-fluence the stock of money, such as actual andpending international specie movements, largechanges in trade balances likely to have aneffect upon such movements, legislation affect-ing the creation or operations of banks or is-sue of other forms of circulating medium, andcentral bank policy announcements. It wasrecognized that these anticipations and ex-pectations might accelerate or retard — or ac-centuate — the usual impact of changes in thestock of money on business and consumer de-cisions, and on some occasions might evengive some of these decisions a priority in timeover the actual changes in the stock of money.

I would suggest that Friedman andSchwartz's findings regarding the variabilityof the length of lag between rates of change inthe stock of money and in business conditionsare likely to be found, upon further examina-tion, to be related to the relative impact, indifferent cycles, through these various chan-nels. His description of the process by whichthe initial impact of central bank action, ifconducted through open market operations, isconcentrated among holders of financial assetsand then diffused to the rest of the communityis essentially an analysis of lags, under modernmonetary and financial arrangements, between


central bank action designed to stimulate crea-tion of additional units of money, the actualcreation of the increments, and receipt ofthose increments by persons and enterpriseswho spend them on consumer or capital goods.It is this part of the monetary theory of theorigin of business fluctuations, as was broughtout in discussions from the floor, that bearssome resemblance to what is called Keynesiantheory. But when increments to and decre-ments from the money stock were producedby the banking system without prior actionby a central bank, as was the case in theUnited States during half of the period coveredby the Friedman and Schwartz study, the char-acter of the stimulus to the banks and the lagsin the process were necessarily somewhat dif-ferent. This, together with recognition of therole of anticipations and the timing of theirimpact, suggests that further analysis, with ascrutiny of the circumstances surrounding thevarious business cycle turning points, mightthrow much light on the variability in thelength of the lag which these authors havefound between monetary policy actions andbusiness cycle turning points.

Friedman and Schwartz's estimate of the

size of what they call the money multiplier —the ratio of the cyclical percentage change inincome associated with the percentage changein the stock of money — indicates that thecyclical variation in the income velocity ofmoney is of approximately the same mag-nitude as the cyclical variation in the stock ofmoney. This conclusion, together with theobserved leadership of the change in the stockof money, implies a lag in cyclical changes invelocity behind business cycle peaks andtroughs. But surely velocity can hardly lagas far behind the cyclical turning points asthe stock of money leads, at least when thelatter is measured by the change in the rateof change. Let us hope that Friedman andSchwartz, in their forthcoming volume ontrends and cycles in the money stock, will in-clude an analysis of the trends and cycles,particularly the cycles, in income velocity andthe timing relationship between such cyclesand those in the stock of money. Pursuit ofthis question should provide a considerabledegree of enlightenment on the sequential proc-ess and the direction in which causal factorsoperate throughout the course of businesscycles.

Documents

This PDF is a selection from an out-of-print volume from ... · a prominent place in business cycle theories — received new emphasis as a result of the Keynesian revolution, so