Monetary Explanations of the Great Depression: A Selective ...€¦ · (1964) refers to a Schumpeterian perspective— that innovation and the movement toward mass production increased

1Federal Reserve Bank of Atlanta E C O N O M I C R E V I E W Third Quarter 2004

The Great Depression (1929–33) was themost severe economic contraction inthe United States during the twentiethcentury (see Figures 1–3). During thecontraction industrial production fellnearly 50 percent from its prior peak.

The unemployment rate reached 24 percent in1933: About one in four people in the workforce waswithout a job.1 By 1933, the price level was morethan 25 percent below its 1929 level. During theGreat Depression a few contemporary analysts andeconomists were critical of Federal Reserve policies(see Currie 1934; Warburton 1945; Burgess 1946).But the prevailing view at that time held that therewas little that Federal Reserve monetary policy couldhave done to moderate the severe contraction (seeBox 1 on page 6).

Economists continue to investigate the hypothet-ical causes of the Great Depression in empiricalresearch, and the culpability of Federal Reservepolicies remains an unsettled issue. Two leadingaggregate explanations for the Great Depression aredistinguished by whether or not they point to theFederal Reserve System and its monetary policies asmainly responsible for the propagation and magnifi-cation of the initial contraction into a depression.

This article is a selective survey of recent macro-economic modeling efforts and empirical workthat examine aggregate explanations for the GreatDepression. The discussion maintains a sharp dis-

tinction between the evidence for each side—thecase for a monetary cause of the Great Depressionversus an alternative view in which monetary policyinstead appears to respond to (rather than cause) asevere contraction. From the voluminous literatureusing vector autoregression (VAR) techniques, theempirical results offer no consensus that monetarypolicy was the main cause of the Great Depression.Empirical results from the more recent, sparser liter-ature using dynamic stochastic general equilibrium(DSGE) modeling also reveal no consensus thatmonetary policy caused the Great Depression.2

Each empirical literature requires sufficientstructure of the economic model to isolate mone-tary policy disturbances, and structure requirescertain assumptions. Monetary policies, in general,are actions by the monetary authority associatedwith the adjustment of money supply as distinct frommoney demand responses (changes in monetarymeasures that take place in response to economicactivity). A monetary policy disturbance arises froman action by the monetary authority that deviatesfrom the model-specified monetary policy reactionfunction, which identifies and distinguishes betweenmoney supply and money demand. The empiricalresearch differs on whether the identified measureof monetary policy initiates the economic contractionor, if the measure does not initiate the contraction,whether monetary policy measures play a crucialrole in explaining the magnitude of the real output

Monetary Explanations of theGreat Depression: A SelectiveSurvey of Empirical Evidence

PAUL EVANS, IFTEKHAR HASAN, AND ELLIS W. TALLMANEvans is a professor of economics at Ohio State University and the editor of the Journal

of Money, Credit, and Banking . Hasan is a professor and acting dean at the Lally School of

Management at Rensselaer Polytechnic Institute in Troy, New York. Tallman is a vice

president and head of the macropolicy team in the Atlanta Fed’s research department.

They thank Jim Nason, Eric Leeper, and Chris Sims for helpful comments.

2 Federal Reserve Bank of Atlanta E C O N O M I C R E V I E W Third Quarter 2004

19

77

= 1

00

1926

18

16

14

6

1927 19321928 1929 1934 1935

4

1931

10

8

19331930

12

F I G U R E 1

Industrial Production

Source: Board of Governors of the Federal Reserve System

1929

30

25

20

5

19331930 1931 1934 1935

0

1932

10

15

F I G U R E 2

Unemployment Rate

Source: National Industrial Conference Board, Business Cycle Indicators, vol. 2 (April 1929–June 1942, pp. 35, 123), NBER series no.08292, NBER Macrohistory Database

1. In contrast, during the post–World War II period, the highest unemployment rate observed was about 11 percent (December1982), and the average “peak” unemployment rate is 6.2 percent (one in sixteen people in the labor force out of work). Notethat unemployment rates may continue to rise after the declared end of a recession.

2. VAR models are statistical time-series models that emphasize the correlations in the data to describe economic dynamics.DSGE models are theory-based models that rely on the behavior of households, firms, intermediaries, and the government,which restricts the dynamics of the economy. Robertson and Tallman (1999) and Del Negro and Schorfheide (2003) provideaccessible introductions and reviews of these tools.

3. A bank intermediates between its depositors, who may leave relatively small balances on deposit at the bank, and its borrowers,to whom the bank issues loans from the proceeds of accepting deposits.


contraction. In each literature, identifications ofmonetary policy vary in their ability to captureimportant institutional features of the banking andfinancial system and how monetary policy, as prac-ticed at that time, affected them. Further refiningthe propagation and transmission mechanism ofmonetary policy to the real economy and isolatingthe key initial shocks will deepen our understandingof how the Great Depression happened and howsuch catastrophes can be avoided.

A Motivating Example: Data Speaking in Tongues

Difficulties in evaluating competing explanationsof the Great Depression arise because data

alone are insufficient to distinguish the importanceof monetary policy during that period. Figure 4 dis-plays the M2 monetary aggregate, the combination

of currency and demand deposits, measuredmonthly over the 1926–35 period. The contractionin M2 occurs more than a year later than the down-turn in either the industrial production index or inthe price level (aside from the brief sharp vacilla-tions in September and October 1929). By late 1930,depositors were making widespread withdrawalsfrom bank demand deposit accounts—that is,demand deposits were converted into currency ona large scale. Figure 5 shows the ratio of M2 to themonetary base (currency plus bank reserves)—a ratiocalled the money multiplier—measured monthlyover the 1926–35 period. The money multiplier indi-cates the level of bank intermediation activity.3 Asan indicator of bank intermediation between bor-rowers and savers (depositors), the multiplier felldramatically during the Great Depression and fell ataround the same time as the M2 aggregate.

Ind

ex

(1

95

8 =

10

0)

1925

60

55

50

35

1926 19311927 1928 1933 1935

30

1930

40

19321929

45

1934

F I G U R E 3

The Price Level Measured by the Implicit GNP Deflator

Source: Barger and Klein (1954), NBER series no. 8260, NBER Macrohistory Database


Bil

lio

ns

of

U.S

. d

oll

ars

1925

50

45

40

25

1926 19311927 1928 1933 1935

20

1930

30

19321929

35

1934

F I G U R E 4

M2 Money Supply

Source: Friedman and Schwartz (1970)

Ra

tio

of

M2

to

th

e m

on

eta

ry b

as

e

1926

7

6

4

1927 19311928 1929 1933 1935

3

1930 1932

5

1934

F I G U R E 5

The Money Multiplier


4. Bernanke (2002) discusses identification generally and in the context of the Great Depression.5. The shocks included the residual financial effects of World War 1 on sovereign nations, the restrictions of the gold standard,

misapplication of the real bills doctrine at the Federal Reserve, agricultural problems in the United States, reparations indebt-edness of the war’s losing forces, and others.


From a cursory examination of Figures 4 and 5,one might conclude that the monetary authority isnot responsible for the Great Depression becausethe contraction in the real economy (industrial pro-duction, the increase in unemployment) precededany notable contraction in monetary aggregates. Butthe vindication of the Fed in this simplistic argumenthas flaws. Figure 6 compares two nominal interestrate series—the call money interest rate and theinterest rate on discount window loans from theFederal Reserve Banks. If these interest rates indi-cate monetary policy actions, then discount rateincreases between January 1928 and August 1929suggest that Federal Reserve policies restricted con-ditions in the financial markets and the real economyprior to the October 1929 stock market crash. Theinterest rate on money borrowed on call at the NewYork Stock Exchange (the call money interest rate)was above the discount rate from January 1928 untilOctober 1929. The Federal Reserve implementedpolicies in 1928 aimed at limiting loans for “specula-tive” purposes, of which call loans were the chief

form (see Wicker 1966). These policies may haveinfluenced the increase in the call rate, therebyhelping to precipitate the real and financial contrac-tion at the start of the Great Depression. The vindi-cation is therefore not clear-cut.

Examinations of the data series alone are unableto distinguish unambiguously between competingmodels or theories. The graphical display of mone-tary and real economic data does not characterizemonetary policy—that is, what the Fed had beendoing systematically through the 1920s in responseto its key operational objectives. These ambiguitieshighlight the problems surrounding the identifica-tion of monetary policy and the identification prob-lem more generally (see Box 2 on page 8).4

It is perhaps too much to ask that a single sourceby itself explains the economic contraction. Recentwork by Eichengreen (2004) and Meltzer (2004)offers several sources, both real and nominal, of neg-ative economic shocks and institutional rigiditiescontributing to the economic collapse.5 Bernanke(in Rolnick 2004) offers a similar perspective on the

1926

12

10

8

4

1927 19311928 1929 1933 1935

0

1930 1932

6

1934

2

Call money interest rate

Discount rate

F I G U R E 6

Short-Term Nominal Interest Rates

Source: Board of Governors of the Federal Reserve System, Banking and Monetary Statistics, 1914–1941 (1943). The call money interestrate is the average rate on stock exchange call loans, new (Table 120, 450–51); the discount rate is the rate at the Federal Reserve Bankof New York (Table 115, 440–42).


In the 1920s and 1930s, Federal Reserve Systemmonetary policies reflected then-popular mon-

etary theories—mainly, a misapplication of thereal bills doctrine to monetary policy. The realbills doctrine implied that the nominal quantityof credit outstanding (namely, bank loans or theliability counterpart, bank deposits) respondedendogenously to the “needs of business and indus-try.”1 Prevailing wisdom in the 1930s argued thatthe Great Depression reflected mainly real eco-nomic phenomena and that rapid liquidation offailed businesses would speed recovery. Textbookdescriptions of that economic consensus wereless concerned with monetary factors. Robertson(1964) refers to a Schumpeterian perspective—that innovation and the movement toward massproduction increased the productivity of the econ-omy so rapidly that the economy could not absorbthe additional output. Separately, banking issueswere offered as a secondary effect of—an endoge-nous response to—the weakened economy. Inaddition, Robertson suggests that Fed policy hadlittle to do with inducing the contraction althoughthe Fed exacerbated the Great Depression through“idiotic” monetary policies.

Wheelock (1991) examines Fed monetary pol-icy during the Great Depression and suggests thatit was more restrictive than policy in the 1920s. Atthe same time, he shows that the 1930s’ monetarypolicy was consistent with an inadequate operatingpolicy—that is, the Fed was inferring the degreeof tightness in monetary conditions from the levelof reserve borrowings at the discount windowwithout sufficient regard to the level of the dis-count rate relative to other short-term interestrates. When demand for discount window loansfell dramatically during 1930, the alarms shouldhave gone off at the Fed. Instead, the existingpolicy framework led to an improper inference—given no demand for discount window loans, thenthe reserves market must have been flush—with-out regard to the price (the discount rate).

Some researchers have argued that the quan-tity theory of money was not developed suffi-

ciently to inform Fed policymakers at the time ofthe Great Depression—that analysis as found inFriedman and Schwartz (1963) would have beentoo revolutionary (see Steindl 1995 and Wicker1999, countered by Humphrey 2001). Although inthe minority at the time, Currie (1934) arguesthat the lack of explicit attention to the behaviorof the aggregate money supply was a key failurein Fed deliberations at that time. Although thereasons behind their respective policy recom-mendations may differ, Wicker (1996), Wheelock(1992), Bordo, Choudhri, and Schwartz (2002),and many others join in support of earlier critics(Currie 1934; Warburton 1945; Friedman andSchwartz 1963) that Fed monetary policy shouldhave aggressively expanded the monetary baseto support economic stability. Whether or notthat policy could have affected the money sup-ply (implicitly, demand deposit) growth rate toa significant degree remains uncertain. Empiricalresearch often imposed strong assumptions abouthow base money growth would affect the moneymultiplier, implicitly assuming that base moneygrowth would stem widespread withdrawals ofdeposits from banks and the observed bank fail-ures. Still, the money supply and its apparent lackof growth during the Great Depression shouldhave alarmed Federal Reserve officials andprompted more drastic monetary base expansion,and it is unlikely that Fed provision of additionalreserves would have worsened bank runs. Thereare limits, however, to the Federal ReserveSystem balance sheet, especially at that timegiven the short existence of the institution andthe lack of explicitly legislated procedures torecapitalize the Fed—the system or individualdistrict banks—if the system became insolvent.Sims (1998) (discussed on page 16) may be refer-ring to the fiscal implications of Federal Reservemonetary actions that could have put their balancesheet at risk of insolvency, or at least illiquidity.Addressing the fiscal ramifications of these mon-etary policy choices in an economic model remainsa challenging research topic.

B O X 1

What Did They Know, and When Did They Know It?

1. Some modern economic theories of business cycles have similar implications. For example, the idea that the credit needsof the real economy will be satisfied passively by financial intermediaries is consistent with an early real business cyclemodel (King and Plosser 1984).

6. Schwartz (1981) argues that the Fed initiated the contraction; Gordon and Wilcox (1981) argue that Friedman and Schwartz(1963) were unsure of the Fed’s role as initiator of the contraction but were convinced that inappropriate Fed monetarypolicies magnified the contraction.

7. See Currie (1934) and Warburton (1945) for comparable arguments.8. See Romer (1993) for a contrary perspective; also see Romer (1992).


Great Depression as “a complicated event” thatcame out of the complex economic and politicalaftermath of World War I. But even in a researchenvironment evolving toward consensus, the notabledifferences of opinion motivate a relentless searchfor more concrete evidence. The research appearsfocused on finding the key underlying sources andunderstanding with greater clarity the propagationmechanism (or how the Great Depression snow-balled from mild recession to calamitous contrac-tion). Ongoing empirical efforts keep the debateslively among economic historians as well as macro-economists. Sharper characterization of the propaga-tion mechanism likely will center on the specificationof the banking and financial sectors of a macro-economic model.

Background Empirical Evidence

The debate over whether the depth and length ofthe Great Depression was largely due to mone-

tary policy errors took center stage long after thesubsequent economic recovery.

The contrasting views were brought to theforefront by Friedman and Schwartz (1963), whoargued that monetary forces magnified the GreatDepression, and Temin (1976), who proposed thatunfavorable real economic shocks were mainly thecause. For most macroeconomists, these two lead-ing aggregate explanations for the Great Depressiondiffer largely with respect to whether FederalReserve System monetary policies were mainlyresponsible for the severe economic contraction.The distinction drawn in the discussion that followsis sharp in order to make the separate explanationsmore distinguishable. So far, the empirical literaturehas been unable to settle this debate althoughproponents on either side likely consider such anambiguous conclusion controversial.

Friedman and Schwartz (1963) propose a mon-etary explanation for the severity of the GreatDepression, placing much responsibility upon theFederal Reserve’s ineffective monetary policies. Inthis view, the monetary authority (the Fed) failedto supply sufficient liquidity in the form of bankreserves and currency during the initial contractionas well as throughout the Great Depression. Withoutthat liquidity, banks were unable to maintain loanactivity, thereby allowing the initial economic down-

turn to magnify by causing large-scale contractionof bank balance sheets and thus leading to sharpincreases in observed bank failures and the pre-cipitous drop in the price level. The assumptionunderlying this view is that the resulting creditcontraction and the associated breakdown in theintermediation process magnified the initial down-turn and turned it into a depression.6

The monetary explanation of the Great Depressionwas not new, but the data, rigorous scholarship, andexhaustive research presented by Friedman andSchwartz brought renewed vigor to the argumentsand combined to challenge most other explanations.7

The main alternative explanation, presented in

Temin (1976), suggests that the Great Depressioncan be explained as a large negative shock to aggre-gate demand. Like Friedman and Schwartz, theaggregate demand view had antecedents and wasconsistent with an earlier, dominant view that mon-etary factors were not essential elements in expla-nations of the Great Depression. Temin argues thatmoney supply growth and real output are largelyjointly determined so that the observed decline inmoney growth may reflect underlying forces thataffect both real output/income and money growth.Notably, Temin concentrates attention on interestrates for indications of monetary policy intervention.

Hamilton (1987) presents evidence suggestingthat monetary authority actions in 1927 and 1928especially led to the contraction in real output atthe onset of the Great Depression.8 The empiricalevidence links monetary-influenced data (interestrates, bank reserves, “high-powered” money) withsubsequent real output contractions. The research,using interest rates in addition to monetary aggre-gates as an indicator for monetary policy, provides

Economists continue to investigate the hypothet-ical causes of the Great Depression in empiricalresearch, and the culpability of Federal Reservepolicies remains an unsettled issue.


Empirical research surveyed in this reviewshares a number of common elements, yet,

despite common elements, the studies presentdivergent conclusions about the role of monetaryfactors in explaining the Great Depression. In theVAR-based research, the conflicting results andconclusions arise from the different restrictions,including differences in the included data series,lag length restrictions, Bayesian prior restric-tions, and the imposition of exclusion restrictionsthat limit the contemporaneous interactionsamong the variables in the system. These latterrestrictions are often described as identificationrestrictions that allow the modeler to isolate anassociation between what the model estimates tobe unpredictable movements in the data series(reduced form errors) and what may be inter-preted as the underlying source of uncertainty inthe model—shocks that can be given a more eco-nomic interpretation.

Researchers want to use identification restric-tions that arise from structural specificationsmotivated from economic theory and to deter-mine whether underlying shocks to monetary pol-icy generate economically plausible results. Aneconomically rigorous conclusion requires thatresearchers design a structural economic modelin which monetary policy actions have moneysupply–based interpretations as distinct from amoney demand source. The model implies a mon-etary policy function and thereby isolates howmonetary policy affects aggregate economic out-comes (for real output, prices, employment, andso on), both actions being crucial for makinginferences about monetary policy’s role in theGreat Depression. Depending on the theoreticalassumptions that underlie the research, sharpchanges in monetary and banking aggregates mayhave been caused by monetary policy, or theymay have been instead the symptoms of the realoutput contraction. The discipline of a structuralmodel allows researchers to see explicitly theimportance of concrete assumptions—such as,for example, the economic measures that themonetary authority is assumed to use in settingmonetary policy (its policy reaction function).

These assumptions are often crucial for theinterpretation of the results. Researchers can

assess the econometric results and, by alteringthe assumptions, can determine whether theinferences change if these assumptions arechanged. For example, economic theory suggeststhat any contractionary monetary policy shockwould produce an increase in the nominal inter-est rate and a contraction in the monetary aggre-gate. Also, that shock would, other things beingequal, reduce output. In this survey, the identifi-cation restrictions in the empirical VAR analysestypically employ just-identified models—that is,there are a variety of alternative specificationsthat are equally likely representations of thereduced-form model. Identifications that areoveridentified provide testable hypotheses andhelp distinguish between identifications that fitthe data from those that do not. There may beinstances in which empirical models retain over-identifying restrictions that have been rejectedby the data because of the strength of theresearcher’s belief in that assumption.

So far, few researchers have settled on oneuniversally accepted specification of the mone-tary policy function, and the long-standing debateregarding the role of monetary factors in theGreat Depression continues. Similarly, there isongoing research in macroeconomic theory tocharacterize more realistically the role of finan-cial intermediation for real economic decisions.

Eichengreen’s synthetic consensus view (dis-cussed on page 9) of the Great Depression mayseem difficult to bring to the data within an empir-ical model. But several elements of that viewappear amenable to a structural VAR model. Insuch a model, an economic structure with a mean-ingful real shock in a monetary propagation mech-anism could suggest that real economic shocksweakened the solvency of the U.S. banking sys-tem, which then was weakened further by therestrictive international financial and monetarystructure. The other unstated but ambiguous issuesurrounds whether the real or monetary shocksdetermined the subsequent deflationary path orwhether that path was an additional outcomearising from the international monetary standard.Identifying a mechanism in a VAR to uncoverpotentially informative effects from these pathsremains a challenge for further research.

B O X 2

Identification Restrictions

9. Ohanian (1999) suggests that analysis of the Great Depression should help economists understand why bad policy choices areimplemented in crisis situations and thereby help prevent such choices in future. See also Bordo, Erceg, and Evans (2000).

10. In the final line of the abstract Eichengreen writes, “For the United States, there is no denying the role of monetary policymistakes in the onset of the Depression, whereas for other countries international monetary instability played the mostimportant part” (2002).


support for a monetary source for the economiccontraction. These results make a credible, narra-tive case supporting the monetary source of thecontraction, and Hamilton points to a variety offinancial measures that indicate a notable change inmonetary policies prior to the Great Depression. Theresearch provides a simple model for a monetarypolicy reaction function, describes a mechanism fortransmitting monetary policy changes to the realeconomy, and then makes inferences about how thechanges in monetary policies affected key macro-economic aggregate measures of real activity. Theconnection between monetary policy actions andthe effects on the real economy does not arise froma precisely specified and estimated economic model.

Recent theoretical and empirical approaches tomacroeconomics suggest a less central role for mone-tary and nominal quantities and have implications forthe debate over the Great Depression that are similarto Temin’s. Research continues to uncover more pre-cise evidence of policy ineptness—in fiscal policy(particularly, regulatory restrictions) as well as mon-etary policy—and whether real economic phenomenaare sufficient for explaining the severity of the realoutput contraction.9 An example of this more recentresearch, Cole and Ohanian (2001) apply generalequilibrium modeling tools to their analysis, in whichbanking and monetary shocks appear less importantfor explaining the severe contraction. The refreshingaspect of this research is that the economic modeldoes not assume a central role for the monetary andbanking sector as a source of real economic contrac-tion. Using the same approach, Christiano, Motto, andRostagno (2003) develop a model in which mone-tary and banking factors are central for explainingthe outcomes observed during the Great Depression.Hence, even in a new literature, no consensus existsregarding whether monetary factors explain theinterwar collapse.

In recent decades, both new empirical techniquesand newly discovered (or created) data have led tonotable advances in the examination of internationaldata and the experiences of other countries duringthe Great Depression as well as the explicit investi-gation of the gold standard’s role in transmitting theGreat Depression (see Bernanke 1995; Eichengreen1992, 2004; Eichengreen and Sachs 1985). Thisresearch provides insights into the worldwide con-

ditions (some institutional, such as the effects of thegold standard, and some political, such as how keynations failed to follow the rules of the gold stan-dard) that led to the Great Depression. This broad-ening literature has pointed out the shortcomings inour understanding and in the available data. Byanswering some questions and linking the answerstogether within a coherent international perspec-tive, these studies contribute toward building a con-sensus. Eichengreen (2004) concludes that theremay be a growing “synthetic” consensus view thatconsiders both monetary policy errors and existinginstitutional structures—the international monetaryand financial system—as generating and transmit-

ting destabilizing impulses around the world. In theEichengreen (2002, 2004) synthetic view, monetarypolicy mistakes in the United States provided theinitial international disturbance that then propagatedacross nations through the gold standard.10

International evidence helps explain the GreatDepression, and additional cross-country researchshould continue to influence empirical research forsome time. But Eichengreen’s “anything approach-ing consensus” view may not yet be the last word onthe topic. Introducing advances from internationalevidence into economic models remains a challenge.In its current form, the synthetic explanation appearstoo inclusive to allow empirical modelers to estimateand test the predictions in standard economic andeconometric methods.

In both Hamilton and Eichengreen, the researchprovides well-reasoned analysis, but only indirectevidence identifies monetary factors as the sourceof the dislocation. More compelling answers to thisquestion require isolating prospective measures forthe monetary policy disturbance and the relationship

Difficulties in evaluating competing explanationsof the Great Depression arise because dataalone are insufficient to distinguish the impor-tance of monetary policy during that period.


not the central focus of his paper, Sims’s work initi-ated what has become a common method of exam-ining the notorious contraction. Figure 7 presentsthe long-term yield on the lowest-rated commercialbonds from 1926 to 1935. The path highlights thedramatic increase in the interest rate in 1931, indi-cating a perception of the rising costs of intermedi-ation at that time. The pattern of perceived riskhints at the additional information that the yielddata convey about the Great Depression.13

Burbidge and Harrison (1985) estimate a VARof the same four variables as in Sims; their researchaims directly at uncovering how monetary distur-bances affected real output during the GreatDepression. In raw form, a VAR model estimates atime series model in which a variable like IP is a lin-ear function of past observations of IP as well as pastobservations of all other variables in the system. Tointerpret the results as “structural,” the researchersmust impose restrictions on the model, and it ishelpful if these restrictions have economic implica-tions. Specifically, one can restrict contemporaneouscorrelations of the data, that is, how IP movementsaffect (or are affected by) movements in other series(like monetary aggregates) within the same timeperiod. In most VAR analyses, restrictions on thecontemporaneous correlations provide the key ingre-dients for interpretation of the results. In a modelthat reflects Temin’s view, contemporaneous move-ments in monetary aggregates would respond tocontemporaneous real output movements, but notvice versa; the economic interpretation is that realoutput movements do not respond contemporane-ously to unpredicted monetary movements. In con-trast, a monetarist model might restrict monetaryaggregate movements to be unresponsive to realoutput movements and allow real output to respondto contemporaneous monetary aggregate move-ments, thus allowing monetary disturbances to havea real output effect contemporaneously.

In both Sims (1980) and Burbidge and Harrison(1985), the identification restrictions impose arecursive ordering as the structure. In this setting,movements in a data series that is first in an order-ing are insensitive to contemporaneous movementsin the variables that follow it in the ordering. Therecursive structure implies an economic intuition;the isolation of money supply versus moneydemand hinges on the inclusion or exclusion of con-temporaneous variables in the monetary policy–related equations. The empirical results examinea selection of potential orderings. One orderingimplies that innovations in M1 precede innovationsin the short-term interest rate (R), the price level

of these measures to key economic aggregates (thatis, a structural economic model) that jointly initi-ated the economic contraction. Further refining ourunderstanding of the Great Depression will help usanswer key questions more concretely.11 Our hunchis that the careful specification of monetary andbanking sectors of the era in a fully specified eco-nomic model will be an important element inanswering these questions.

Evidence That Monetary Policy Mattered

The VAR-based studies of the Great Depressionbegin with Sims (1980); the paper compares

the interwar and post–World War II business cycles

with particular focus on the explanatory power ofmonetary aggregates for real output movements.Sims examines two models: a three-variable modelof the monetary aggregate (M1), industrial produc-tion (IP), and the wholesale price index (WPI) anda four-variable model of R (a short-term interest rate,here the four–six month commercial paper rate),M1, WPI, and IP. The VAR is estimated in logarithmsof the series (except for the interest rate) over theinterwar period (from 1920 to 1941) using twelve lagsof monthly data and a constant term. The empiricalresults for both models indicate that innovations tothe monetary aggregate explain over 50 percent ofthe forecast error variance of industrial productionat the four-year horizon whether or not the interestrate variable is included in the VAR. The result refersto empirical model estimation that uses the entiredata sample, and the empirical evidence supports amonetary explanation of the Great Depression.12

In his conclusion, Sims questions whether, for theinterwar period, monetary innovations would con-tinue to explain so much of real output movementsin a larger model that encompasses additional finan-cial surprises. Such a model would likely isolate adifferent estimate of monetary innovations, account-ing for financial data series like risk spreads, depositsof suspended banks, and other indicators of inter-mediary crisis. Although the Great Depression was

Recent theoretical and empirical approachesto macroeconomics suggest a less centralrole for monetary and nominal quantitiesand have implications for the debate overthe Great Depression.


(WPI), and then real output (IP). In this structure,the shock to the monetary aggregate is equivalentto the residual error from the reduced-form equation.The innovation in the monetary aggregate incorpo-rates responses to other variables only through lags.The innovation in the interest rate reflects that theinterest rate responds contemporaneously to mon-etary aggregate movements. Similarly, the price levelinnovation reflects that the price level responds tocontemporaneous movements in both the monetaryaggregate and the interest rate. Finally, the real

output innovation reflects that the real outputresponds to contemporaneous movements in themonetary aggregate, the interest rate, and the pricelevel. A second ordering reverses Rand M1. A thirdordering simply moves M1 to the last position sothat no other shock variables respond contempora-neously to shocks in the monetary aggregate.14 Therecursive VAR identifications do not generate strongstructural interpretations for the money demandand money supply functions from the first orderingor the other orderings.

1926

13

11

9

1927 19311928 1929 1933 1935

3

1930 1932

7

1934

5

Pe

rce

nt

F I G U R E 7

Long-Term Bond Yield, Lowest-Rated

Source: Board of Governors of the Federal Reserve System, Banking and Monetary Statistics, 1914–1941 (1943, Table 128, 468–71)

11. Examples of such questions are, If monetary policy mistakes were important, then did monetary policy err more by failingto increase bank reserves through open market purchases, by failing to provide reserves through the discount window, byreducing the discount rate too slowly, or all of the above? Of course, the central question is then, What monetary policiescould have avoided the Great Depression?

12. Sims contrasts the implications of the interwar period results with the post–World War II sample; in the more recent sample, inno-vations in the monetary aggregate explain a lower percentage of real output variance when the VAR includes the interest rate.

13. It is notable that the yield spread between the Moody’s Baa bond (in Figure 7) and the long-term Treasury bond (not shown)follows the pattern in the risky commercial bond.

14. Typically, the innovation in a data series (M1) will have greater power for explaining the variance in other data series in themodel when it (M1) is first in the ordering. The ordering implies identification restrictions, and the identification is “justidentified,” which means that the identification restrictions allow a unique mapping between the “reduced-form” or unre-stricted parameter estimates and the structural model. However, the just-identified system is indistinguishable from all otherjust-identified structures because there are no restrictions implied by the structure on the reduced-form parameters. In thiscase, the chosen ordering is as likely to occur as alternative orderings or, more generally, just-identified models. Hence, thetheoretical justification that explains and supports the identification is important, but it is notable that there are fewtheoretical models that motivate the choice of a recursive structure.


among the data series—that is, a persistent, long-run relationship between the data series. In thatstory, the cointegrating relationship is a standardmoney demand function that includes both an inter-est rate term and a real output term.

mt– p

t= α0 + α1yt

+ α2rt+ v

t,

where α represents a parameter, m is the log ofmoney, p is the log of the price level, y is the log ofoutput, r is the nominal interest rate, and v is theerror term.

Bordo, Choudhri, and Schwartz (1995) examinethe counterfactual question, “Could stable moneygrowth have averted the Great Depression?” Theauthors display counterfactual simulations of thissmall econometric model estimated over the sam-ple period that includes the Great Depression.One simulation assumes a stable money demandfunction throughout the full estimation sampleand enforces only that the money growth rate is amean growth rate and a monetary shock insensi-tive to past variations in the other series. Theresults indicate that holding the money growthrate to an average estimated from 1929 to 1933moderates the loss in real economic activity byabout one-third, which is still a contraction ofGreat Depression magnitude. Further restrictingthe money growth process by shutting off themonetary shock reduces the contraction notably.Under the assumption that M2 money growth isfully determined by Fed policy, their empiricalmodel suggests that, had the Fed maintained astable growth rate of money, then the contractionin real output and the fall in the price level wouldhave been muted during the Great Depression.18

The researchers find strongest support that amoney growth rule would alleviate the huge realoutput contraction when they estimate separatelythe basic model over two subperiods and usereduced-form errors from two separate estima-tions for the model simulations for the whole peri-od. These results rely on strong restrictions thatare not imposed in the baseline specification andlimit the generality of the inferences.

Several identifying assumptions seem crucialfor generating the findings in Bordo, Choudhri, andSchwartz, and these key assumptions are not tested.One critical assumption is the one underlying thesimulations—namely, that money growth can becontrolled directly by Fed policy. The model speci-fication does not have a banking sector in whichbank deposits may be partly determined by the pri-

The empirical evidence uses historical decompo-sitions, a procedure that uses model parameter esti-mates from the entire data sample and makes anunconditional forecast (for a given forecast horizon)using only data up to the period prior to the firstforecast period.15 Then, the researcher can estimatehow much of the forecast error can be explained byeach identified shock.16 Burbidge and Harrison sug-gest that the identification restrictions (that is, theorderings) have a substantial influence on the esti-mated role of money in explaining the behavior ofreal output and prices over the Great Depression.When the monetary aggregate is first in the recur-sive ordering, the addition of the monetary aggre-

gate improves the forecast for the WPI price leveland the IP output measure more than if the mone-tary aggregate is placed last in the ordering. In otherwords, if the innovation in the monetary aggregate(the model forecast error) is assumed to reflect bestthe money supply disturbance, then the VAR modelsuggests that monetary disturbances explain muchof the subsequent decline in real output.

The overall conclusions favor a measurable rolefor monetary disturbances in the explanation of thedeclines in both real output and the price level dur-ing the Great Depression in general. For the initialcontraction (1929 to 1931), the authors find that thebaseline model forecast improves little when errorsassociated with the monetary aggregate are added.Hence, the results suggest that monetary factors inthis model cannot help explain the decline in realoutput from 1929 to 1931 (see Burbidge and Harrison1985, 52). Still, the evidence supports the idea thatmonetary mistakes magnified the real output con-traction in 1932 and after.

Bordo, Choudhri, and Schwartz (1995) estimate avector error correction model (VECM) using quar-terly data from 1921:Q1 to 1941:Q4 for three vari-ables—real gross domestic product (GDP), theimplicit GDP deflator, and the M2 monetary aggre-gate.17 A VECM is a time-series model similar to aVAR that explicitly accounts for cointegration

International evidence helps explain theGreat Depression, and additional cross-country research should continue to influ-ence empirical research for some time.

15. Note that the forecasting model employs coefficient values estimated from the full-estimation sample, an inherent tensionwith the goal of the forecast experiment. However, the procedure allows the model to assign explanatory power to thesubsequent forecast error associated with each variable in the system.

16. The procedure sequentially adds observed, in-sample forecast errors for each variable in the system until the forecast matchesthe actual data series.

17. The researchers also introduce a rate of interest measure and a proxy measure for financial fragility by introducing the dif-ference in the logarithm of the ratio of deposits in suspended banks to total deposits. These measures are offered in variationsof the basic specification.

18. Evans (1997) finds evidence that supports the claim that consistent money growth through the Great Depression wouldhave alleviated much of the real contraction during that period.

19. Calomiris (1993, 73–75) explains this point in detail, providing a comprehensive summary of the key debates about the roleof financial and intermediation shocks in explaining the Great Depression.


vate sector. Federal Reserve notes (currency) andbank reserves are liabilities of the Federal ReserveSystem. Currency, specie, and bank deposits makeup the M2 monetary aggregate. A major portion ofM2 is “inside” money (time and demand deposits inthe banking system) that is not ultimately a claimon the Federal Reserve System and instead reflectsdebt and loan contracts between private sectorentities. Observed movements in deposit aggre-gates may have more to do with disturbances in pri-vate sector activity and could have been largelyresponding to price level and real output move-ments on the quarterly frequency.

During the Great Depression, movements inbank deposits were the dominant source of mone-

tary aggregate movements. Figure 8 shows that thecurrency component of M2 behaves differently frombank deposits over this period. The deposit compo-nent may respond to changes in the currency stock,bank reserves, and the amount of currency that theprivate market demands. However, monetary policycontrolled only the supply of the sum of currencyand bank reserves; market forces determined therelative proportions. As bank runs persistedthroughout the early 1930s, the composition of M2reflected the public’s growing aversion to holdingbank deposits. It remains unclear how FederalReserve policies would have alleviated this aversionunequivocally.19 In addition, bank failures and thecorrespondent loss of access to deposits may need

1926

50

40

30

15

1927 19311928 1929 1933 1935

0

1930 1932

25

1934

5

45

35

20

10

De

po

sit

s i

n b

illi

on

s o

f d

oll

ars

6

2

0

3

4

1

Cu

rre

nc

y i

n b

illi

on

s o

f d

oll

ars

5

Deposits in bank

Currency

F I G U R E 8

Demand Deposits versus Currency



effects of shocks associated with it.21 The empiricalapplication may have overextended standard VARprocedures, using eight variables, estimating severalsimultaneous coefficients in the identification, andincluding a moving average variance of stock prices.Although the VAR introduces additional financialdata to fortify the financial sector, the money demandand supply functions include unusual contempora-neous relationships with variables, such as realdeposits in failing banks, stock price volatility, andreal liabilities of failed nonfinancial businesses. Theestimation and interpretation of these functions aredebatable, thereby leaving research opportunitiesto reinforce and extend these findings. Still, theconclusions from the research anticipate and areconsistent with later empirical research using addi-tional econometric procedures.

Cecchetti and Karras (1994) investigate the keycauses for the depth and the persistence of theGreat Depression. The paper employs three differ-ent structural specifications, each using alternativeidentification methods in an attempt to isolate mon-etary sources of the real contraction from othersources. One identification method associated withBlanchard and Quah (1989) imposes restrictions onthe long-run properties of a model that includes realoutput, prices, and the nominal interest rate. In this

special handling relative to withdrawal of depositsfrom the banking system. Further research on mod-els with fully specified money demand functionsthat examine the stability of the functions over theentire interwar period appears warranted.

Fackler and Parker (1994) employ an eight-variablestructural VAR that includes a variety of data seriesrelatively underutilized in the empirical literaturefor the interwar period. For instance, the depositsof failed banks, the liabilities of failed nonfinancialbusinesses (Figure 9), and changes in the movingvariance of the Standard and Poor’s stock index areuseful indicators of the degree of financial distress.20

The authors emphasize financial fragility—thebreakdown in the intermediation process facilitatedby financial markets and the banking system—as anexplanation for the Great Depression as distinct fromthe monetary policy mistakes view. Their resultsindicate that an autonomous demand shock likelyinitiated the real output contraction, that monetarydisturbances exacerbated the contraction startingin 1931 or so, and that a nonmonetary financialintermediation shock explained some of the earlycontraction and most of the positive real outputresponse to the bank holiday in March 1933.

Their model uses a number of simultaneous equa-tion restrictions to identify monetary policy and the

1926

300

250

200

100

1927 19311928 1929 1933 1935

0

1930 1932

150

1934

50

Mil

lio

ns

of

19

57

–5

9 d

oll

ars

F I G U R E 9

Real Liabilities of Failed Nonfinancial Businesses

Source: Dun’s Liabilities of Failed Nonfinancial Businesses deflated with the wholesale price index, 1957–59 = 1.0

20. See Coe (2002), described below. Notably, Coe’s estimates suggest that the most sustained periods of financial distress occurafter March 1931.

21. Fackler and Parker may face the concern of “weak instruments” for methods to estimate the parameters in simultaneous spec-ifications of VAR structural error, noted by Pagan and Robertson (1998). See footnote 24 for a more extensive discussion.

22. See Galí (1992) for another example of an identification that combines long-run restrictions and contemporaneous restrictions.23. Eichengreen (2004) notes that the authors rule out any concern for a speculative attack on the gold reserves in their model.


setting, shocks associated with aggregate demandare assumed to be temporary, whereas supply shocks(for example, those that have durable effects on thefactors of production) have permanent effects. Still,uncertainty exists about how well the identificationcan isolate monetary shocks because that sector ismeasured only by an interest rate. Shocks referredto as aggregate supply and aggregate demand mayalso reflect monetary-based shocks that are notcaptured by interest rate innovations. A separatespecification attempts to isolate labor supply fromtechnological shocks to specify further the aggregatesupply. An alternative model attempts to isolatemonetary shocks from the other shocks using aninnovative identification scheme within a model ofreal output, the nominal interest rate, the real inter-est rate, and real M2.22 The model tries to separateaggregate demand shocks from money supply shocksto account for the role of monetary policy errors.The paper concludes that monetary disturbanceswere important for explaining the severe and lengthyoutput contraction, but the initial downturn wasmainly driven by a sharp contraction in aggregatedemand less associated with monetary disturbances.

Cecchetti and Karras, like Fackler and Parker,also find that monetary shocks alone are insufficientto explain the Great Depression although mone-tary factors are important for magnifying the ini-tial contraction. In both cases, however, the iden-tification of monetary policy is relatively unusual.In Cecchetti and Karras, the variables in the moneydemand and money supply functions appear insuf-ficient to isolate monetary policy shocks. In con-trast, Fackler and Parker add to the money demandand supply functions variables that make it diffi-cult to interpret the resulting shocks. More precisecharacterizations of monetary policy and analysisof the transmission mechanism of monetary shocksto real output contractions are opportunities forfurther research.

Bordo, Choudhri, and Schwartz (2002) specifyan overidentified structural model in which simula-tions suggest that the gold standard would not havebeen threatened by expansionary open market mon-etary policies that the Fed failed to pursue. Thispaper and Hsieh and Romer (2001) offer evidenceto refute the argument in Eichengreen (1992) that

the Fed perceived its gold reserves position as aconstraint on its monetary policy. In Bordo, Choudhri,and Schwartz (2002), the specification of the mon-etary process focuses on the determination of high-powered money (base money) rather than on themonetary aggregate. High-powered money is thequantity that the Fed influences directly, and thepaper focuses on a central institutional questionthat likely influenced operational decisions during theDepression. However, the simulation results requireexplicit assumptions regarding a counterfactual pathfor the money multiplier (how high-powered moneytranslates to the M2 aggregate) in order for thecounterfactual base growth to affect the monetary

aggregate.23 It is possible that more appropriate Fedmonetary policies might have kept the money mul-tiplier from falling so precipitously in 1931, but it isfar from certain. The determination of the moneymultiplier is a central issue in empirical research onthe Great Depression and remains an opportunityfor further research on the transmission mechanismas well as on the financial intermediation processmore generally.

From a new literature, Bordo, Erceg, and Evans(2000) employ a DSGE model with sticky wages(that is, wages that adjust only slowly to marketforces) to investigate the Great Depression. Thespecification of the real economy can account forsuch nominal frictions to evaluate the role of thesefrictions in the extent and duration of the contrac-tion. Their findings suggest that the main explana-tion for the decline in output was ineffective andinappropriate monetary policy. Yet the model spec-ifies a simple financial sector. There is no insidemoney creation, and the money shock measure is

As bank runs persisted throughout the early1930s, the composition of M2 reflected thepublic’s growing aversion to holding bankdeposits. It remains unclear how FederalReserve policies would have alleviated thisaversion unequivocally.


model parameters. The estimated model employsdata from 1919 to 1939 for six time-series—indus-trial production, the consumer price index, a com-modity price index, currency, M1, and the discountrate. In contrast to several earlier VAR papers,Sims’s paper focuses the identification to isolatethe “structural” shocks of monetary policy by char-acterizing a policy reaction function using specifi-cations of Federal Reserve operating policies inpost–World War II data. The application of such apolicy function to the interwar period is anachro-nistic, and ultimately unrealistic, yet it providesclearly defined estimates for the supply and demandfunctions for money.

The structural identification isolates monetarypolicy disturbances—that is, changes in monetarypolicy behavior that would not have been pre-dictable—from money demand fluctuations. Notably,there are estimation complexities that arise fromsimultaneous coefficient estimation.24 Other shocksin the system are less easy to interpret. Sims alsouses a Bayesian prior in which the hyperparameterseffectively smooth out the estimated coefficient val-ues for the lags in the system.25 The results suggestthat innovations related to monetary policy do notexplain much of the subsequent variation of realoutput or in the price level. Instead, real output andprice fluctuations are explained mainly by their owninnovations, consistent with the idea that the econ-omy was subject to substantial shocks aside frommonetary policy during the Depression.

Notably, Sims does not infer from these resultsthat monetary policy had little effect on real outputand the price level. Rather, he indicates that themeasure of monetary policy included in his modelmay not be sufficient to capture monetary policy’seffects on the solvency of the banking system or tomeasure the impact of bank solvency regardless ofmonetary policy influence.26 In other words, someimportant effects of the banking system on real out-put and the price level are not addressed in hisVAR. This conclusion hints at an intuition offered byBernanke (1983) that nonmonetary factors, such asa breakdown in financial intermediation, of whichbank intermediation is a crucial component, mayunderlie some of the evidence supporting monetarycauses for the Great Depression.

Coe (2002) has investigated further the idea ofa separate role for financial fragility. Coe isolates anestimate of the conditional probability of financialcrisis from two indicators of disruptions in financialintermediation—the yield spread between the Baabond and the comparable-term Treasury bond andthe ratio of currency to deposits. Then the author

simple and stylized. The Federal Reserve Systemfailed to offset the decline in the money multiplier,resulting in a loss of confidence in the financial sys-tem. However, the question is whether the Fed couldhave done anything about the money multiplier.This paper shows that the model results are consis-tent with a money shock story. But the model hasno explicit transmission mechanism or monetaryand financial sector, and, as a result, the evidence isonly suggestive.

In summary, the evidence from these papers sug-gesting that inappropriate monetary policies werethe key causes for the Great Depression still leavessome fertile area for research. For example, more

detailed characterizations of financial intermediationmay help isolate the precise mechanisms wherebyFederal Reserve actions that increase the monetarybase can then affect the behavior of the money mul-tiplier. Further research on the transmission mech-anism of monetary policy could help in understandingwhether the counterfactual expansionary policy forthe monetary base would have produced the desiredeffects on the growth in demand deposits and totalM2 money supply. However, true believers in themonetary source of the Great Depression may viewassumptions that Fed policies to increase the mon-etary base would thereby increase the money multi-plier as well justified without empirical tests. Amongall these papers, the results support a causative roleof monetary factors, to varying degrees, in theGreat Depression.

Evidence That Monetary Factors Were Not Important

Sims (1998) revisits the topic of his earlier paperthat examines the difference in the relation-

ships between economic performance and mone-tary policy in the postwar and interwar periods. Inthis paper, he exploits advances in identifying VARmodels as well as in applying Bayesian techniquesto estimation in pursuit of more accurate estimatesof model structure and more precise estimates of

The determination of the money multiplier isa central issue in empirical research on theGreat Depression and remains an opportunityfor further research.

24. Simultaneity implies that the same contemporaneous variable appears in two equations or the dependent variable in oneequation (M2) appears in the independent variable equation (the discount rate) and vice versa. See Sims (1986) andBernanke (1986). See also Pagan and Robertson (1998) for some of the complications that arise.

25. See Sims and Zha (1998). For a less technical description of the prior, see Robertson and Tallman (1999). These priors tendto aid models in forecasting and limit the variability of coefficient estimates.

26. Sims may also have concerns that, as the Fed was near the zero bound on nominal interest rates, any activist monetary pol-icy would have a fiscal dimension. That is, the Fed policies could place the balance sheet of the Fed at risk of substantialloss, a consideration that may have constrained monetary policy choices and brought a fiscal dimension to monetary policy.

27. Coe does not attempt to address the question of whether monetary factors explain the real contraction.28. “Out-of-sample” implies estimating the statistical model using data available up to the date when the forecasts are to begin.

Next, the researcher uses that model to forecast a given number of periods beyond the estimation sample. The techniqueapproximates the real-time process of estimating a forecasting model and then forecasting several periods into the future.


examines whether the estimated probability of finan-cial crisis helps explain the behavior of real output inaddition to lags of real output and lags of monetaryaggregate growth. The results suggest that the prob-ability of financial distress has significant explanatorypower in addition to the other variables.27 Furtherresearch may clarify the mechanisms that relatefinancial distress to real output contraction.

Among recent empirical work on the GreatDepression, Ritschl and Woitek (2002) employ aVAR to uncover the explanatory power of money inthe presence of previously untapped time-seriesdata. The Great Depression in the United Stateswas associated with a steep and substantial con-traction in the manufacturing sector. The authorsestimate a Bayesian VAR that examines whether thetime-series behavior of monetary data is effectivein predicting the sharpness of the real output con-traction during the Great Depression. The modelincludes data on real output (IP), prices (CPI),wholesale prices (WPI), total reserves (TR), andnonborrowed reserves (NBR). An alternative modelreplaces the monetary measures with the discountrate (DR) to investigate a specification using theinterest rate as the key mechanism for monetarytransmission. The estimation imposes a Bayesianprior on the model and employs recursive estima-tion; that is, the model is estimated using only datameasures dated prior to the forecast period in orderto forecast the contraction in real output. The pro-cedure is anachronistic, but the experiment uncov-ers an unusual finding that the monetary variables(or interest rate measures) are not very useful inforecasting real output out of sample.

Next, Ritschl and Woitek use data on residentialbuilding permits along with measures from the steelindustry—shipments of machines, sheet steel pro-duction, steel ingot production, and the prices ofmetal products—to measure whether a VAR modelof these variables can more accurately forecastthe steepness of the real output contraction. Theauthors find that the VAR made up of these real

indicators can forecast the degree of output con-traction effectively even in the out-of-sample set-ting using a model estimated up to January 1929.28

Figure 10 displays steel ingot production versustotal IP and illustrates the significant comovements(sheet steel production offers similar insights).

Ritschl and Woitek conclude that the real datacan predict the steepness of the contraction and,therefore, that the Great Depression may have hadat its core real phenomena driving the degree of con-traction. In their paper, the main inference appearsoverextended. The VAR model forecast fromJanuary 1929 raises interesting issues regarding theinitial shock responses, but prediction accuracydoes not measure fully our understanding of theeconomic process. The paper presents a VAR thatexploits a correlation between the aggregate outputmeasure (IP) and key measures of sectoral output—steel. At that time, the U.S. automobile industry wasa growth industry, and that industry was pummeledduring the Great Depression. Ritschl and Woitekshow that steel production indicators help predict asteep decline in output. Still, the forecast does notindicate the identity of the underlying sources of theshock that drives the real output contraction. Byforecasting the path of the real output contractionfrom what might be called component data series,the paper uncovers an interesting statistical artifactthat may stimulate further investigations into theultimate source of the real shock.

Recently, Cole and Ohanian (2001) approachedthe enigma of the Great Depression with an alter-native perspective—that of the real business cycletheory. The approach imposes the discipline of gen-eral equilibrium analysis applied to an aggregatemodel, often with minimal frictions. In empiricalimplementations, the approach must often findvalues or estimates of key model parameters fromother economic studies and impose those parametervalues on the general equilibrium model. The empir-ical strategy is to examine whether the observationsin key data series are consistent with predictions


that their model cannot explain the observation ofthe Great Depression through the contraction in theirstylized banking sector has as much to say aboutthe adequacy of their modeling framework as itdoes about the importance of the banking collapseas an explanation of the Great Depression. Both themodeling framework and the banking hypothesismay need additional features to constitute a satis-factory explanation. But the progress and attentionin this general equilibrium modeling literature sug-gest that advances are likely to occur in our under-standing of the interwar economic collapse.

The differences across the empirical results arisefrom selection of the set of variables used in theestimation, the identification (or overidentification)of the structure, and the restrictions that are imposedon the estimation (that is, Bayesian priors). For theGreat Depression, money and other banking mea-sures influence the estimation of monetary policymeasures, but present specifications are insufficientto account for the monetary transmission process.As a result, the empirical results appear inconsis-tent: Sims (1998) and Ritschl and Woitek (2002)suggest that real sources were central causes forthe Great Depression, whereas Bordo, Choudhri,and Schwartz (1995, 2002) show evidence that theGreat Depression was mainly the result of huge

arising from model simulations. In many cases, theanalysis amounts to assessing how closely simulateddata from the model can replicate the statisticalcharacteristics of observed data series. In their analy-sis, Cole and Ohanian find that, conditional on theirmodeling framework and their calibrated parametervalues, the banking contraction observed during theGreat Depression is not associated with such asevere contraction in output in their model as wasexperienced during the Great Depression. As a result,they conclude that other sources of shocks shouldbe investigated to explain the Great Depression in ageneral equilibrium setting.

One view of their research is that the generalequilibrium modeling framework is currently toorestrictive and therefore does not mimic closely theeconomic mechanisms that the banking system pro-vides in the actual economy. From this view, Coleand Ohanian’s model is not close to explaining theGreat Depression. For example, the assumption of asingle representative agent framework may notallow the model to capture the complex facets offinancial contracts across agents. As a result, whenbanking failures take place in the actual economy,the closure of a bank may produce an inability toenforce these contracts and may prevent the trans-fer of banking information to another bank. The fact

1926

250

200

150

1927 19311928 1929 1933 1935

0

1930 1932 1934

100

50

Ste

el

ing

ot

pro

du

cti

on

(th

ou

sa

nd

s o

f lo

ng

to

ns

pe

r a

ve

rag

e w

ork

ing

da

y)

18

6

0

10

14

4

Ind

us

tria

l p

rod

uc

tio

n i

nd

ex

(1

97

7 =

10

0)16

12

8

2

Steel ingot production

Total industrial production

F I G U R E 1 0

Steel Ingot Production versus Total Industrial Production

Source: Steel ingot production from 1938 Statistical Report of the American Iron and Steel Institute, NBER series no. 01135, NBERMacrohistory Database; industrial production from the Board of Governors of the Federal Reserve System


monetary mistakes. The modest conclusions ofSims (1998), along with the provocative findings inCecchetti and Karras (1994) and Fackler andParker (1994), suggest that detailed specification ofthe transmission mechanism may be a productivepath for further research. Ultimately, the empiricalresearch may lead toward more of a consensus thatthe flawed policies of the Fed were largely respon-sible for the depth and duration of the GreatDepression. New evidence using an estimated VARmay continue to contribute to the literature, espe-cially if it embodies more explicit modeling of thefinancial sector.

Introducing a Banking Sector into the Empirical Models

The inclusion of a banking sector into estimatedmodels may help uncover whether the data are

consistent with the assumed behavior of the moneymultiplier. For example, expansionary monetarypolicy may increase the growth in the monetarybase (increasing both currency stock and bankreserves). To raise money supply growth effectively,the growth in the monetary base must affect thebehavior of borrowers and lenders. Increased mon-etary base growth reduces the intermediation costsbanks face so that banks increase their loans.Isolating the effects of Fed policy during the GreatDepression still centers on what caused the con-traction in bank deposits. The contraction could berational banker responses to weakening loan demand,or it could be that the Fed failed to maintain suffi-cient growth in the monetary base and inadvertentlyincreased banks’ costs of intermediation. Researchaimed at detecting the monetary elements of theGreat Depression investigates whether the actions(or inaction) of the Fed in the reserves market failedto lower the costs of intermediation sufficiently toallow banks to maintain the levels of financial activ-ity consistent with (or in support of) a recoveringeconomy. Few empirical papers specifically addressthis issue, but some recent empirical papers incor-porate a banking sector more completely.

Recently compiled banking data may help in thespecification of the financial sector and the transmis-sion mechanism. Anari, Kolari, and Mason (forth-coming) calculate a data series that accumulates thedeposits of banks that remain in liquidation, essen-tially assessing the quantitative effects of illiquiddeposits. The data are carefully constructed andoffer insights that differ from those offered by thetypical data series that aggregates the deposits ofsuspended banks. The authors note that the keycontribution of their data series is to disentangle

the deposits of failed banks (deposits that were inac-cessible for long periods of time) from deposits inbanks that were suspended for only a brief time.For example, all banks were suspended during thethree-day bank holiday in March 1933, but relativelyfew were liquidated. Figure 11 displays total bankdeposits versus the deposits-in-suspension series.

The paper employs this new data series in a four-variable VAR model composed of the industrial pro-duction index, the wholesale price index, M1, andthe stock of bank deposits in suspension. In addi-tion, the authors find that the model variables arecointegrated and so estimate both a VAR and a vec-tor error correction specification that imposes the

proposed restrictions. The empirical results fromthe restricted model (the vector error correctionmodel) indicate that the stock of deposits in failedbanks in liquidation explains an important portionof the fluctuations in industrial production, sup-porting the role for a credit availability indicatorin explaining output dynamics during the GreatDepression. These results suggest that intermedia-tion frictions that arose from the inaccessibility ofcredit and liquidity were important.

In research by Calomiris and Mason (2003a, b),further investigations in detailed banking data exploitbalance sheet data for individual Federal Reservemember national banks during the Great Depression.The authors attempt to uncover whether bankingfailures were explained by aggregate liquidity short-ages or by insolvency due to the real economiccharacteristics of the bank balance sheets. The twopapers find the key measurable characteristics thatseparate banks that fail from those that survivemainly in the items on individual bank balance sheets.That is, bank balance sheet data (individual bankfinancial conditions) explain bank failure; this find-ing is consistent with an explanation that adverse realeconomic shocks may have affected some banksmore than others.

Despite this finding, Calomiris and Mason stillconclude “there can be little doubt” that aggressive

The search for one conclusive empirical study ofthe Great Depression is likely futile, akin to thesearch for a single source of the contraction.


that return is not contingent on the observation ofcurrent period shocks. This friction allows themodel a mechanism to capture the credit risk facedby banks during the Great Depression.

The empirical results from this paper suggestthat the monetary factors in the Great Depressionwere largely responsible for the depth and length ofthe contraction. It is notable that the empiricaltechniques have not yet become standard practice.A criticism of the empirical results is that a largeproportion of the ability to explain the movementsin the actual data series is due to exogenous shocks,most notably, a shock associated with liquidity pref-erence. An increase in this shock indicates thathouseholds prefer to hold a higher proportion ofliquid assets as currency as opposed to bank deposits.In this model, the liquidity preference shock leadsto higher currency holdings, a lower volume of bankdeposits, fewer bank loans, and less investment.The illustration of bank disintermediation effectsin the model help identify how monetary-relatedshocks may affect real variables in a hypotheticaltransmission mechanism. In his comments onChristiano, Motto, and Rostagno, Ohanian (2003)notes that their paper offers advances to generalequilibrium–based investigations of the Great Depres-sion and expresses concerns about the substantial

open market operations or the departure from thegold standard could have avoided the aggregate col-lapse in 1931–33. It is notable that their empiricalresults support the hypothesis that real economicshocks were the source of bank failure, which con-flicts with the hypothesis of an aggregate liquiditysource of bank failures in their sample. The strongconclusion that an alternative Fed policy may haveavoided the aggregate economic collapse in1931–33 appears to conflict with their own findingsbut is consistent with other empirical and historicalresearch. The comment displays how the view thatFed policy mistakes were largely to blame for themagnified contraction in real output has becomeaccepted among many economic historians.

From an alternative modeling perspective,Christiano, Motto, and Rostagno (2003) design aDSGE model with a sophisticated financial sectorand a more complicated transmission mechanismfor monetary policy along with labor market rigidi-ties. The model incorporates a mechanism throughwhich monetary policy can have substantially non-neutral effects on real output and the price level.The banks issue consumer deposit accounts fromwhich the banks hold part as reserves and lend therest of the funds to firms. The deposit accountsoffer a fixed nominal return to the consumer, and

1921

60

50

40

20

1923 19311925 1927 1935 1939

0

1929 1933

30

1937

10

De

po

sit

s i

n b

illi

on

s o

f d

oll

ars

1941

Total bank deposits

Deposits in suspended banks

F I G U R E 1 1

Bank Deposit Contraction and Deposits in Suspended Banks

Source: Deposits in suspended banks from Anari, Kolari, and Mason (forthcoming); total bank deposits from Friedman and Schwartz (1970)


variability and explanatory power of key exogenousshocks. More effort on deciphering the sources andrefining estimates of the variability of these shockswill add to the influence of this new path in research.Despite criticisms, the modeling strategy brings amore realistic financial sector with implications thatallow for reasonable monetary non-neutralities,and the empirical results are in line with the estab-lished “monetary causes” empirical literature. Asa result, the paper demonstrates further that, likeother empirical literatures, the real business cycle/DSGE model approach can generate results on bothsides of the monetary versus real debate about theGreat Depression.

Conclusions

The empirical literature on the Great Depressionis far from consensus about the source of the

contraction or the quantitative role of monetaryshocks in the real output contraction. The VAR lit-erature offers evidence both in favor of and in oppo-sition to a central role for monetary policy amongthe causes for the Great Depression. Similarly,recent work using the real business cycle or DSGEmodeling approach displays a similar lack of con-sensus regarding Fed culpability for the GreatDepression. The search for one conclusive empiri-cal study of the Great Depression is likely futile,akin to the search for a single source of the con-traction. The “synthetic consensus” offered byEichengreen (2002, 2004) may offer a coherentexplanation of the Great Depression using the accu-mulation of empirical evidence to suggest that theGreat Depression arose from circumstances andundesirable shocks too complex to capture ineconometric models. Still, ongoing efforts to inno-vate in both economic theory and in VAR econo-

metric statistical analysis, along with newly con-structed data series, open up opportunities to suggestseveral fertile paths for uncovering useful elementsof the monetary transmission process from thattime period.

The purpose of further study on this most heav-ily researched period in our economic history is toclarify our understanding of the sequence ofevents—both the unforecastable and exogenousshocks and the policy errors that followed. A rigor-ous accounting of the effects on the economy fromeach source may help policymakers today evaluatethe risks faced in our current financial structure.From our vantage point, the most promising path isa sharper focus on the financial sector and a morepotent specification of the transmission mechanismof monetary policy along with methods to incorpo-rate elements of the labor market frictions noted inother research.

Our view is that the monetary history and analy-sis of the Great Depression (separate from the VARempirical literature) suggest strongly that mone-tary factors played an important role in magnifyingand extending the painful economic contraction.The inability of the VAR literature to reliably verifythis insight suggests that the identification of mone-tary policy during this period remains elusive. Otherresearch using aggregate economic models alongwith in-depth analysis of banking and other financialintermediaries suggests that models need to accountexplicitly for the disruptions in the banking andfinancial sectors. Further work to isolate sources ofthe monetary disturbance in identifications withmore detailed formulation of the financial and bank-ing sectors in the monetary transmission mechanismmay contribute to our understanding of interactionsbetween monetary and real economic factors.


hypothesis. Journal of Money, Credit, and Banking

35, no. 6, part 2:1119–97.

Coe, Patrick J. 2002. Financial crisis and the Great Depres-sion: A regime-switching approach. Journal of Money,

Credit, and Banking 34, no. 1:76–93.

Cole, Harold L., and Lee E. Ohanian. 2001. Re-examiningthe contributions of money and banking shocks to theU.S. Great Depression. In NBER Macroeconomics

Annual 2000, 183–227. Cambridge, Mass.: MIT Press.

Currie, Lauchlin. 1934. The failure of monetary policy toprevent the depression of 1929–32. Journal of Political

Economy 42, no. 2:145–77.

Del Negro, Marco, and Frank Schorfheide. 2003. Takeyour model bowling: Forecasting with general equilibriummodels. Federal Reserve Bank of Atlanta Economic

Review 88, no. 4:35–50.

Eichengreen, Barry. 1992. Golden fetters: The gold stan-

dard and the Great Depression, 1919–1939. New York:Oxford University Press.

———. 2002. Still fettered after all these years. NBERWorking Paper No. 9276, October.

———. 2004. Viewpoint: Understanding the Great Depres-sion. Canadian Journal of Economics 37, no. 1:1–27.

Eichengreen, Barry, and Jeffrey Sachs. 1985. Exchangerates and economic recovery in the 1930s. Journal of

Economic History 45, no. 4:925–46.

Evans, Paul. 1997. What caused the Great Depression?Managerial Finance 23, no. 2:15–24.

Fackler, James, and Randall E. Parker. 1994. Accountingfor the Great Depression: An historical decomposition.Journal of Macroeconomics 16, no. 2:193–220.

Friedman, Milton, and Anna J. Schwartz. 1963. Monetary

history of the United States, 1876–1960. Princeton, N.J.:Princeton University Press.

———. 1970. Monetary statistics of the United States:

Estimates, sources, methods. New York: ColumbiaUniversity Press.

Galí, Jordi. 1992. How well does the IS-LM model fit post-war U.S. data? Quarterly Journal of Economics 107,no. 2:709–38.

Gordon, Robert J., and James A. Wilcox. 1981. Monetaristinterpretations of the Great Depression: An evaluationand critique. In The Great Depression revisited, editedby Karl Brunner, 49–107. Boston: Martinus Nijhoff.

Hamilton, James D. 1987. Monetary factors in the GreatDepression. Journal of Monetary Economics 19,no. 2:145–69.

Hsieh, Chang-Tai, and Christina Romer. 2001. Was theFederal Reserve fettered? Devaluation expectations inthe 1932 monetary expansion. NBER Working Paper No. 8113, February.

Anari, Ali, James Kolari, and Joseph Mason. Forthcoming.Bank asset liquidation and the propagation of the GreatDepression. Journal of Money, Credit, and Banking.

Barger, Harold, and Lawrence R. Klein. 1954. A quarterlymodel for the United States economy. Journal of the

American Statistical Association 49, no. 266:413–37.

Bernanke, Ben S. 1983. Nonmonetary effects of the finan-cial crisis in the propagation of the Great Depression.American Economic Review 73, no. 3:257–76.

———. 1986. Alternative explanations of the money-income correlation. Carnegie Rochester Conference

Series on Public Policy 25 (Autumn): 49–99.

———. 1995. The macroeconomics of the Great Depres-sion: A comparative approach. Journal of Money, Credit,

and Banking 27, no. 1:1–28.

———. 2002. On Milton Friedman’s ninetieth birthday.Delivered at the conference to honor Milton Friedman,University of Chicago, November 8.

Blanchard, Olivier J., and Danny Quah. 1989. The dynamiceffects of aggregate demand and supply disturbances.American Economic Review 79, no. 4:655–73.

Bordo, Michael, Ehsan U. Choudhri, and Anna J. Schwartz.1995. Could stable money have averted the GreatContraction? Economic Inquiry 33, no. 3:484–505.

———. 2002. Was expansionary monetary policy feasibleduring the Great Contraction? An examination of the goldstandard constraint. Explorations in Economic History

39, no. 1:1–28.

Bordo, Michael, Christopher Erceg, and Charles Evans.2000. Money, sticky wages, and the Great Depression.American Economic Review 90, no. 5:1447–63.

Burbidge, John, and Alan Harrison. 1985. A historicaldecomposition of the Great Depression to determinethe role of money. Journal of Monetary Economics

16, no. 1:45–54.

Burgess, W. Randolph. 1946. The Reserve Banks and the

money market. New York: Harper Brothers.

Calomiris, Charles W. 1993. Financial factors in theGreat Depression. Journal of Economic Perspectives 7,no. 2:61–86.

Calomiris, Charles, and Joseph Mason. 2003a. Conse-quences of U.S. bank distress during the Great Depression.American Economic Review 93, no. 3:937–47.

———. 2003b. Fundamentals, panics and bank distressduring the Depression. American Economic Review 93,no. 5:1615–47.

Cecchetti, Stephen, and Georgios Karras. 1994. Sourcesof output fluctuations during the interwar period: Furtherevidence on the causes of the Great Depression. Review

of Economics and Statistics 76, no. 1:80–102.

Christiano, Larry, Roberto Motto, and Massimo Rostagno.2003. The Great Depression and the Friedman-Schwartz

R E F E R E N C E S


Humphrey, Thomas M. 2001. The choice of a monetarypolicy framework: Lessons from the 1920s. Cato Journal

21, no. 2:285–313.

King, Robert G., and Charles I. Plosser. 1984. Money, credit,and prices in a real business cycle. American Economic

Review 74, no. 3:363–80.

Meltzer, Allan H. 2004. A history of the Federal

Reserve—Volume 1: 1913–1951. Chicago: Universityof Chicago Press.

Ohanian, Lee. 1999. The defining moment: A reviewessay. Journal of Political Economy 107, no. 1:178–85.

———. 2003. Liquidity shocks and the Great Depression:Comment on “The Great Depression and the Friedman-Schwartz hypothesis” by Larry Christiano, Roberto Motto,and Massimo Rostagno. Journal of Money, Credit, and

Banking 35, no. 6:1205–15.

Pagan, Adrian R., and John C. Robertson. 1998. Structuralmodels of the liquidity effect. Review of Economics and

Statistics 80, no. 2:202–17.

Ritschl, Albrecht, and Ulrich Woitek. 2002. Did monetaryforces cause the Great Depression? A Bayesian VARanalysis for the U.S. economy. Humbolt UniversityWorking Paper, January.

Robertson, John C., and Ellis W. Tallman. 1999. Vectorautoregressions: Forecasting and reality. Federal ReserveBank of Atlanta Economic Review 84, no. 1:4–19.

Robertson, Ross M. 1964. History of the American econ-

omy. New York: Harcourt Brace Jovanovich.

Rolnick, Arthur J. 2004. Interview with Ben S. Bernanke.Federal Reserve Bank of Minneapolis The Region 18, no. 2:18–27.

Romer, Christina. 1992. What ended the Great Depression?Journal of Economic History 52, no. 4:757–84.

———. 1993. The nation in depression. Journal of

Economic Perspectives 7, no. 2:19–39.

Schwartz, Anna J. 1981. Understanding 1929–1933. InThe Great Depression revisited, edited by Karl Brunner,5–48. Boston: Martinus Nijhoff.

Sims, Christopher A. 1980. Comparison of interwar andpostwar business cycles: Monetarism reconsidered.American Economic Review 70, no. 2:250–57.

———. 1986. Are forecasting models usable for policyanalysis? Federal Reserve Bank of Minneapolis Quarterly

Review 10, no. 1:2–16.

———. 1998. The role of interest rate policy in the gen-eration and propagation of business cycles: What haschanged since the ’30s? In Beyond shocks: What causes

business cycles? edited by Jeffrey C. Fuhrer and ScottSchuh, Federal Reserve Bank of Boston Conference Series42 (June): 121–60.

Sims, Christopher A., and Tao Zha. 1998. Bayesian meth-ods for dynamic multivariate models. International

Economic Review 39, no. 4:949–68.

Steindl, Frank. 1995. Monetary interpretations of the

Great Depression. Ann Arbor: University of MichiganPress.

Temin, Peter. 1976. Did monetary forces cause the

Great Depression? New York: W.W. Norton & Company.

Warburton, Clark. 1945. Monetary theory, full production,and the Great Depression. Econometrica 13, no. 2:114–28.

Wheelock, David C. 1991. The strategy and consistency

of Federal Reserve monetary policy, 1924–1933.Cambridge: Cambridge University Press.

———. 1992. Monetary policy in the Great Depression:What the Fed did, and why. Federal Reserve Bank ofSt. Louis Review 74, no. 2:3–28.

Wicker, Elmus. 1966. Federal Reserve monetary policy,

1917–1933. New York: Random House.

———. 1996. The banking panics of the Great Depres-

sion. New York: Cambridge University Press.

———. 1999. What did the Fed know? Money supply the-ory and the Great Depression. Paper presented at theSouthern Economic Association meetings, New Orleans,La., November 21–23.

Documents

Monetary Explanations of the Great Depression: A Selective ...€¦ · (1964) refers to a Schumpeterian perspective— that innovation and the movement toward mass production increased