Could stable money have averted the great contraction?

Bordo, Michael D. ; Choudhri, Ehsan U. ; Schwartz, Anna J. 等

I. INTRODUCTION

Although over fifty years have elapsed since the onset of the Great Contraction, controversy still swirls about the factors responsible for its depth and long duration. A basic issue is the hypothesis due to Friedman and Schwartz [1963] that the Great Contraction would have been attenuated had the Federal Reserve not allowed the money stock to decline. Although the validity of the hypothesis has been a subject of much debate, there is little empirical evidence directly addressing it.

A direct test of the Friedman and Schwartz hypothesis requires simulating the behavior of output under a counterfactual policy of stable money. That approach was adopted by Warburton [1966], Friedman [1960], and Modigliani [1969], none of whom constructed a model of the economy on the basis of which to conduct his test. Recently, McCallum [1990] has used simulations from a macro model to show that his base rule (with feedback) would have avoided the severe decline in nominal income that occurred between 1929 and 1933.(1) The extent to which his rule would have improved the behavior of nominal income depends on the value of the feedback coefficient. McCallum's simulations, however, do not address the issue considered here of how a stable-money rule would have performed.

In this paper we present simulations that focus on the stable-money counterfactual. Our model is based on McCallum's, but we use a more general framework than his that (a) estimates separate relations for output and the price level, and (b) places no prior restrictions on the feedback from output and the price level to money. Our stable-money counterfactual does not change the long-term rate of money growth (from that actually experienced) and thus involves only short-run changes in the behavior of the money supply. We assume that dynamics embedded in the output and price relation are not especially sensitive to this type of policy change.

We simulate two variants of Milton Friedman's [1960] constant money growth rule. The first, a strong form of the rule, assumes that the Fed could quickly offset changes in the money multiplier by changes in high-powered money and thus keep the money stock on the constant growth rate path in each quarter. The second, a weaker form of the rule, assumes that the money multiplier is observed with a one-quarter lag and the Fed can only set the expected rate of growth of money (conditional on last-period information) equal to the constant rate.

Our basic simulation is derived from a model estimated for the 1921.1-1941.4 interwar period. The results of this simulation buttress the views of Friedman and Schwartz. Had a constant money growth rule (with an annual rate equal to the average 1921-41 rate of 2.95 percent) been followed throughout the interwar period the Great Contraction would have been avoided. Real output would have declined from 1929.3 to 1933.1, but the order of magnitude of the cumulative change (-11.1 to -21.6 percent) and the annual rate of change (-3.3 to -6.6 percent), when the actual cumulative change was -36.2 percent and the annual rate of change was -12.1 percent, would have been comparable to other recessions in the nineteenth and twentieth centuries.

In the basic simulation we assume that the output and price relations remained stable over the entire interwar period. The extraordinary economic conditions of the Great Contraction could have arguably shifted the output and price relations. Thus, we also present an alternative simulation that uses estimates of these relations based on the 1920s data. This simulation shows that a constant money growth rule (with an annual rate equal to the 1921-29 average of 3.29 percent) would have prevented even a mild recession during the Great Contraction period.

Section II contains a brief survey of the continuing debate over the cause and propagation mechanism of the Great Contraction and its implications for the stable-money hypothesis. Section III presents an overview of the behavior of the variables considered in our model. Section IV describes our methodology for performing counterfactual simulations for the interwar years 1921 to 1941. Section V provides basic simulations of the paths of output and the price level under different variants of the constant money growth rule. Section VI explores the sensitivity of our results to alternative specifications. Section VII contains a brief conclusion.

II. THE CONTROVERSY OVER THE GREAT CONTRACTION

Since the publication of A Monetary History in 1963, a voluminous literature has appeared, some of it critical and some supportive of Friedman and Schwartz's contention that monetary forces--specifically the failure of the Federal Reserve to engage in expansionary open market operations to offset a series of banking panics beginning in October 1930--turned a not-unusual cyclical downturn in the United States into the greatest depression of all time. Various survey articles in recent years have dealt with many of the issues--notably Bordo [1989], Calomiris [1993], Eichengreen [1992a], Temin [1993], and Romer [1993]--so our discussion will be brief.

In the 1970s and 1980s debate focused on Peter Temin's [1976] contention that monetary forces could not have been the primary cause of the contraction because short-term interest rates fell rather than rose, as would be consistent with a left-ward shift in the LM curve. For his alternative causal mechanism he posited a collapse of autonomous expenditures.

The consensus of the subsequent literature was that, although monetary forces were paramount, nonmonetary shocks should not be discounted. Thus, according to Hamilton [1987] and Field [1984], tight monetary policy initiated by the Fed in 1928 to counter the stock market boom and to stem a gold outflow to France, following her return to gold at an undervalued parity, initiated the contraction. The stock market crash of October 1929 then aggravated the downturn in the year following the crash by both increasing uncertainty and wealth-loss effects that led to a reduction in expenditures on consumer durables, as in Romer [1990] and Mishkin [1978]. Romer [1993] agreed that banking panics in 1930, 1931, and 1933 turned a severe recession into depression. Though consensus prevails on the primacy of monetary forces, debate still persists over the issues of the exogeneity of the money supply, as in Schwartz [1981], Bordo [1989], and Calomiris [1993], and of the propagation mechanism.

In the past decade, the focus of the literature has shifted to the mechanism by which monetary collapse was transmitted to the real economy. What is in contention is whether declining output responded to a bank-panic-induced (by a fall in the deposit-currency and deposit-reserve ratios) decline in money supply in the face of sticky wages and prices, or whether the bank failures themselves reduced output via the nonmonetary channel of the disruption of the process of financial intermediation, as suggested by Bernanke [1983].

This debate hinges on the issue of whether or not deflation was anticipated. If not, as Hamilton [1987; 1992] argued, based on evidence from commodity futures markets, and as Evans and Wachtel [1993] found for 1930-32, based on nominal interest rates, then debt deflation (by reducing the net worth of firms and banks) was the propagation mechanism, as in Fisher [1933], and Bernanke and Gertler [1989] and [1990]. If it was largely anticipated, as Nelson [1992] argued, based on an examination of the business press, and as Cecchettti [1992] found, based on time-series models of prices, then the mechanism worked through a rise in the ex ante real interest rate.

Another unsettled issue concerns the causal and propagation roles of international factors--whether the U.S. depression spread to the rest of the world or vice versa, as in Kindleberger [1973] and Temin [1993], with the consensus favoring the U.S. depression as the primary cause with reflex influences on the United States from declining output abroad, as Eichengreen [1992a] concluded. In addition, both Eichengreen [1992b] and Temin [1989] have argued that adherence to the gold standard rule of convertibility was crucial both in the international transmission of shocks and in preventing the monetary authorities of adherents from following reflationary policies.

Our approach in this paper is compatible with much of the recent literature. We develop a methodology to ascertain the effect of the Friedman and Schwartz hypothesis of stable money on output and the price level. Such a framework is compatible with propagation mechanisms based on sticky wages and prices and on mechanisms stressing both unanticipated and anticipated deflation. It is also compatible with studies that identify the relative importance of different shocks, such as those of Cecchetti and Karras [1994] for the United States, and of Betts, Bordo and Redish [1995] for the United States and Canada. According to these studies, which identify the banking panics as supply shocks, output declines up to the middle of 1931 are generally accounted for by demand shocks and subsequent output declines mostly by supply shocks. Regardless of the source and importance of supply shocks, these results are not inconsistent with the Friedman and Schwartz hypothesis, since a stable money policy could have diminished the effect of all types of shocks to the economy.

One objection to our approach is that as long as the United States was committed to gold convertibility, it was beyond the Fed's power to undertake expansion of domestic credit to maintain stable money through the Great Contraction, as Eichengreen [1992b] argues.

For several reasons we do not believe this raises serious problems. First, given that the United States had the largest monetary gold stock in the world and that at its lowest point in 1932 the Federal Reserve gold reserve ratio was 56 percent, the threat of the United States being forced off the gold standard could not have been a binding constraint.

Second, we do not accept the view that Eichengreen [1992b] has revived, namely, that the Fed could not undertake an expansionary policy because of a shortage of "free gold"-gold not earmarked either as backing for Federal Reserve notes and deposits or as replacement for backing by ineligible assets. Had free gold been a genuine problem, why did not the Federal Reserve importune the Congress to alter the eligibility requirement before such legislation was enacted in February 1932? According to Friedman and Schwartz [1963, 406], free gold was "largely an ex post justification for policies followed, not an ex ante reason for them."(2)

Third, had the gold constraint been a real deterrent to pursuing expansionary monetary policy in the Great Contraction, the United States could have suspended convertibility, as Britain and other countries did. Once they did so, their economies rebounded, as described in Eichengreen and Sachs [1985] and Eichengreen [1992b].

A second objection to our approach, advanced by Calomiris [1993] among others, is that the Fed lacked the requisite understanding to maintain stable money. In favor of this view, Wicker [1965] and Brunner and Meltzer [1968] have argued that the Fed accepted the flawed Burgess-Riefler-Strong doctrine that focused on levels of member bank borrowing and short-term nominal interest rates as policy indicators. On this view the Fed conducted expansionary open market operations in the 1924 and 1927 recessions because neither member bank indebtedness nor nominal interest rates had declined. By contrast, in 1930 member bank borrowing and nominal interest rates fell. Hence the Fed believed no action was warranted, according to Wheelock [1992], and Bordo [1989].

Against this objection, we cite two facts: the connection between stable money and the real economy was well known at the time, as noted by Laidler [1993], even by isolated Federal Reserve officials; and other central banks had on earlier occasions acted successfully as lenders of last resort.

III. OUTPUT, PRICES, AND MONEY IN THE INTERWAR YEARS

Before discussing our model, we briefly review the behavior of real output, the price level, and the money supply during the interwar period. For data definitions and sources, see Appendix A.

Figure 1a shows quarterly time series of the log of real GNP (y) for the period 1921.1-1941.4. Output grew from 1921.1 to the cyclical peak in 1929.3 at an annual rate of 6.4 percent. The period was marked by two mild recessions: 1923.2 to 1924.3 and 1926.3 to 1927.4. Output then declined from 1929.3 to the cyclical trough in 1933.1 by 36.2 percent (at an annual rate of -12.1 percent), the sharpest and most prolonged decline in U.S. history. Rapid recovery then followed until the outbreak of World War II, with output growing at an annual rate of 8.6 percent. Recovery was marred by a brief, sharp recession from 1937.2 to 1938.2.(3)

Figure 1b presents the quarterly data for the log of the implicit GNP deflator (p) for the interwar period. Following a sharp post-World War I deflation, the price level was mildly deflationary until the summer of 1929 when it began a rapid plunge for the next three and a half years. It declined by 26.8 percent, or an annual rate of -8.5 percent. The deflation of 1929-1933 was the most severe in U.S. history. Fisher [1933] and Bernanke [1983] regard unanticipated deflation as an aggravating condition in the Great Contraction because of its deleterious effects on the balance sheets of households and firms. From the cyclical trough in 1933.1 until World War II prices advanced at an annual rate of 3.5 percent except for a deflationary episode from 1937.3 to 1939.2.

Figure 1c displays the behavior of the log of the M2 money supply (m).(4) Following a sharp policy-induced contraction after World War I, the money supply increased from 1922.2 until 1928.2 at an annual rate of 5.7 percent. The subsequent failure to grow in 1928-29 reflected contractionary Federal Reserve policy in reaction to the stock market boom and a gold flow to France after it returned to the gold standard at a parity that undervalued the franc as in Hamilton [1987] and Field [1984]. Money supply declined for the first year of the Great Contraction by 2.2 percent, a decline not much different from that experienced in earlier severe recessions (Friedman and Schwartz [1963]),(5) but, beginning in October 1930, with the onset of a series of banking panics, it began a plunge whereby m 1n M2) fell by 12.4 percent at an annual rate. The fall in m did not end until the Banking Holiday of March 1933.

Beginning in March 1933, the money supply expanded rapidly at an annual rate of 7.8 percent until World War II, with the exception of a sharp contraction in 193738. According to Friedman and Schwartz, the Federal Reserve's decision to double reserve requirements in an attempt to soak up what it regarded as excess liquidity in the banking system was responsible for the monetary contraction.

Friedman and Schwartz link the severe decline of the money supply from 1930 to 1933 to a series of banking panics which ultimately caused one-third of the nation's banks to fail. The banking panics reduced the money supply by their effects on the money supply multiplier (see Figure 2a). The bank failures, attributed to the absence of Federal Reserve lender-of-last-resort action, undermined the public's confidence, leading to a massive decline in the deposit-currency ratio. The banks in turn reduced their loans, resulting in a sharp fall in the deposit-reserve ratio (see Figure 2b).

Friedman and Schwartz emphasize the bank failures as the force which led to a collapse of the money supply; for Bernanke [1983] they were also important for their nonmonetary role of impairing the financial intermediation process and hence reducing the level and growth rate of real output.(6) As a standardized measure of the importance of bank failures, we use the log of the ratio of deposits in suspended banks to total deposits (s) (see Figure 2c).(7) This ratio, s, was quite high in the 1920s, reflecting deep-seated structural problems in the U.S. banking system--the weakness of unit banks in primarily agricultural areas of the country. The dramatic increase in s from 1930 to 1933 largely reflects the banking panics. Following the Bank Holiday of March 1933 and the advent of FDIC in 1934, s declined to a much lower level.(8) That the rise in s in 1930-33 was accompanied by a fall in the money supply multiplier underscores the importance of bank failures in the money supply process.

IV. METHODOLOGY

In this section we describe the methodology we use to simulate counterfactual historical situations for the interwar period. We estimate a small quarterly model of the interwar U.S. economy. To keep our framework simple, our basic model includes only three variables: output, y, the price level, p and money, m. However, we also consider two variations of the basic model. In one variation, we introduce the rate of interest (r) as an additional variable while in the other we add the variable s to capture the role of bank failures. We assume that money affects output in the short run. The short-run non-neutrality of money is assumed to result from sticky prices or wages.(9)

We assume that y, p, m, r, and s are all stationary in first differences.(10) Also, we assume initially that the first four of these variables are cointegrated via a money demand relation of the following form:

(1) [m.sub.t]-[p.sub.t]=[[alpha].sub.0]+[[alpha].sub.1][y.sub.t]+ [[alpha].sub.2][r.sub.t]+[v.sub.t]

where [v.sub.t] is a stationary component.(11) In view of (1), we include [v.sub.t-1] as an error correction term in the equations that model the behavior of output and the price level as follows:

(2) [Mathematical Expression Omitted]

(3) [Mathematical Expression Omitted]

where [a.sub.ij](L) are lag polynomials of order k, [z.sub.t]= [r.sub.r] or [s.sub.t], and [Mathematical Expression Omitted] and [Mathematical Expression Omitted] are mutually uncorrelated white-noise disturbances representing demand and supply shocks respectively.(12) The determination of the money supply is modeled as

(4) [Mathematical Expression Omitted]

where [Mathematical Expression Omitted] is a white-noise money supply shock (uncorrelated with-both [Mathematical Expression Omitted] and [Mathematical Expression Omitted]). Equation (4) allows for-potential effects of y, p and z on m-through high-powered money (via the Fed's reaction function) as well as the money multiplier.

In our basic model, we suppress the role of both the interest rate r and bank failures s by letting [[alpha].sub.2] = 0 in (1), and [[alpha].sub.14](L) = 0, i = 1,2,3, in equations (2) through (4). Friedman and Schwartz [1963] treated bank panics as exogenous shocks to money supply and did not consider output and the price level as significant determinants of either high-powered money or the money multiplier. Their view suggests including bank panic shocks in [Mathematical Expression Omitted] and setting [a.sub.31](L) = [a.sub.32](L) = 0 in (4).

Before discussing the estimation of our basic model, it is interesting to compare it with McCallum's [19901 model and to note the points of difference. First, McCallum estimates a relation for nominal income growth (i.e., [delta][y.sub.t] + [delta][p.sub.t] in our notation), which depends on its own lagged values and lagged money growth. This relation can be thought of as a special case of our equations (1) and (2), which could be derived by placing restrictions on appropriate lag polynomials in (1) and (2). Second, McCallum models money growth differently from the way we do. He assumes that the growth of the money base is exogenous but allows growth in the money multiplier to depend on a bank failure variable (denoted by s in our model), which in turn is influenced by deviations of nominal income from its target path. Thus, there is a feedback from nominal income to money growth in his model, but this effect is restricted to the above channels and is not as general as our equation (3).

We estimate our basic model in the form of a vector error correction model. Letting [x.sub.t] [equivalent to] [[delta][y.sub.t], [delta][p.sub.t], delta[m.sub.t] and [eta] [equivalent to] [Mathematical Expression Omitted], write equations (2) through (4) as [Ax.sub.t] = [[beta].sub.0] + [[beta].sub.1][v.sub.t-1] + [A.sup.-1] to obtain

(5) [x.sub.t]=[[gamma].sub.0]+[[gamma].sub.1][v.sub.t-1]+C(L) [x.sub.t-1]+[e.sub.t]

where [y.sub.i] = [A.sub.-1][[beta].sub.i], i=0, 1, C(L) = [A.sup.-1]B(L) and [e.sub.t] [Mathematical Expression Omitted] is a vector of reduced-form shocks. The reduced-form shocks are related to structural shocks by

(6) [e.sub.t] = [D[eta].sub.t]

where D(= [A.sup.1][delta]) [equivalent to] [[D.sub.ij]], i, j = 1,2,3.

To perform counterfactual simulations of output, y, and prices, p, based on (5), we define the constant money growth policy as the Fed targeting a rate of growth of the money stock equal to [theta], based on all available information. Under this policy

(7) E([m.sub.t] | [I.sub.t])=[m.sub.t-1]+[theta]

where [I.sub.t], is the Fed's information set in period t. The degree of monetary control under a constant money growth policy clearly depends on the information available to the Fed. We consider several possibilities. At a minimum, we assume that the Fed observes all variables with a one-period lag (i.e., [I.sub.t], contains values of all variables up to t-1). Using this weak informational assumption, the constant money growth rule (7) implies that the [delta]m equation of model (5) will change to

(8) [Mathematical Expression Omitted]

We replace the [delta]m equation in (5) by (8) to simulate the behavior of output and the price level under this policy. This simulation does not require identification of structural shocks, and would be consistent with any view about the contemporaneous relationship between m, y, and p.

A stronger assumption about the Fed's monetary control is that it knows contemporaneous money supply shocks (i.e., [I.sub.t] includes [Mathematical Expression Omitted] in addition to information on all variables up to-t-1). In this case the constant money growth policy would eliminate [Mathematical Expression Omitted] shocks. Using (6), the constant money growth rule under the stronger informational assumption implies that

(9) [Mathematical Expression Omitted]

where [Mathematical Expression Omitted]. Moreover, with money supply shocks eliminated, [delta]y and [delta]p equations in (5) will be subject only to shocks [Mathematical Expression Omitted] respectively. Note that the simulation based on (9) requires the identification of only [Mathematical Expression Omitted] shocks as [Mathematical Expression Omitted]* can be estimated as a residual in a projection of [Mathematical Expression Omitted] w = y,p,m,.

A number of approaches can be used to identify [Mathematical Expression Omitted], the money supply shock term. One interesting possibility is suggested by the Friedman-Schwartz view that the behavior of the money supply was essentially independent of output and prices. If it is assumed that y and p do not affect m at least contemporaneously, then [a.sub.31](0) = [a.sub.32](0) = 0, and these exclusion restrictions imply that [D.sub.31] = [D.sub.31] = 0 while [D.sub.33] = 1. Thus [Mathematical Expression Omitted], and (since [Mathematical Expression Omitted] = 0) the counterfactual money supply equation (9) reduces to

(10) [delta][m.sub.t]=[theta]

As (8) and (10) represent polar assumptions about monetary control, the next section focuses on simulations based on these equations to present the least and the most favorable cases for stable money.

As extensions of our basic model, we include a fourth variable z representing either interest rates r or bank failures s. The behavior of the fourth variable is modeled as

(11) [Mathematical Expression Omitted]

where [Mathematical Expression Omitted] is assumed to be white noise and uncorrelated with other shocks. In the case of [z.sub.t] = [r.sub.t], [Mathematical Expression Omitted] can be viewed as shocks to asset markets while for [z.sub.t]=[s.sub.t], [Mathematical Expression Omitted] ran be thought of as financial shocks relevant to bank suspensions and failures. The model represented by (1), (2), (3), (4) and (11) [with [[alpha].sub.2]=0 in (1) for [z.sub.t]=[s.sub.t]] can be estimated in the form of (5) with [x.sub.t] and [e.sub.t] redefined as [[delta][y.sub.t], [delta][p.sub.t], [delta][m.sub.t]] and [[Mathematical Expression Omitted]]. In this case, however, the restriction that y and p do not contemporaneously affect m is not sufficient to identify [Mathematical Expression Omitted] since z could still exert a contemporaneous effect on m. While identification of [Mathematical Expression Omitted] is not needed for counterfactual (8), additional assumptions would be required to justify counterfactual (101.

Our simulations assume that a change in the [delta]m equation does not alter other equations of the model. We suppose that the short-run adjustment process is not very sensitive to a change in monetary policy. The long-run rate of inflation would, however, depend on the long-run rate of money growth. Thus, in constructing our-counterfactual, we avoid a change in the long-run money growth rate by setting [theta] equal to the average value of [delta]m for the period over which our model is estimated.

Our model does not explicitly constrain money to be neutral in the long run. One implication of long-run neutrality of money is that [[eta].sup.m] would not permanently affect y. This restriction poses no problem for simulations based on (8) as any set of restrictions used to identify structural shocks would be consistent with this counterfactual. However, restrictions used to identify [Mathematical Expression Omitted] for counterfactual (10) would not constrain the long-run effect of this shock to equal zero.

V. BASIC SIMULATIONS

Before discussing our simulations, we present some evidence on the relationship between money and other variables in the model in the interwar period. Table I shows results of Granger causality tests between certain variables in the-model for 1921.1 to i1941.4. Regressions in this table include the lagged value of-the error correction-term, D, and three lagged values of other regressors. Here, assuming that [[alpha].sub.1] equals one and [[alpha].sub.2] equals zero, we measure v by (m-p-y).(13)

[TABULAR DATA I OMITTED]

The first three regressions in the table are based on the three-variable framework of the basic model. In the first regression, we examine the influence of output and the price level on money. As the results show, three lagged values of both [delta]y and [delta]p as well as the lagged value of v, the error correction term, are insignificant in the regression explaining [delta]m. This evidence on the absence of Granger causality from output and the price level to money is consistent wish the view that the money supply in this period was exogenous. The second and the third regressions in the table examine the links from money to output and the price level. The lagged values of [delta]m are significant (at the .011 level) in the [delta]y equation but insignificant in the [delta]p equation. However, even in the [delta]p equation,=the lagged value of v is significant (at the .023 level), and thus money does exert an effect on the price level through the error correction term. Although we have suppressed the role of the interest rate in our basic model, we nevertheless explore the influence of this variable on money, output and the price level in regressions 4, 5 and 6. As the results show, interest rates do not improve the predictive content of any of these regressions.

The remaining regressions :in the table explore the interaction of s with m, p and y. The results based on regressions 7 through 9 show that lagged As terms are significant in the [delta]m and [delta]p equations (at .017 and .360 levels, respect very) but not in the [delta]y equation. In regression 10, however, lagged values of both [delta]p and [delta]m are significant (at .015 and .052 levels) in the [delta]s equation. These results suggest a role for bank failures (s) in the model, but note that the effects operating between s and other variables are not very strong. For examine, inclusion of this variable in the [delta]m and [delta]p equations increases the [R.sup.2] of these equations only marginally: by .043 and .027, respectively (compare regression 7 with 1, and 9 with 3). The [R.sup.2] of the [delta]s equation, moreover, is only .160.

We begin with simulations based on the basic vector error correction model (5) with three variables. We estimate this model using three lags for=each variable.(14) Although our interest is primarily in the Great Contraction period, we simulate the behavior of [delta]y and [delta]p under the constant money growth policy for the whole interwar period. We focus on the two variants of the constant money growth policy discussed above. The weak case for this policy is represented by model I defined as model (5) with its [delta]m equation replaced by (8). Model II, on the other hand, represents the strong case. This model is defined as model (5) in which not only is the [delta]m equation replaced by (10) but also [Mathematical Expression Omitted] in the [delta]y and [delta]p equations are replaced by [Mathematical Expression Omitted] (estimated as residuals in regressions of [Mathematical Expression Omitted]). In both models I and II, [theta] is set equal to .00738 (the mean value of Am for 1921.11941.4).(15)

Figure 3 shows the simulated behavior of m, p, and y for models I and II, and compares it to the actual behavior of these variables. The actual and counterfactual monetary policies are contrasted in panel (a) of the figure. Note that even the weaker constant money growth rule (used in model I) would have resulted in only minor decreases in the money stock during 1931 to 1933.

The basic question raised in this paper is addressed in panel (b) of Figure 3. As this figure shows, a fall in output would have occurred starting in 1929 in model I, but the simulated decline (-21.6 percent cumulative and -6.61 percent at an annual rate) would not have been as severe and prolonged as the decline in the Great Contraction (-36.2 percent cumulative and -12.1 percent at an annual rate). In fact, under this counterfactual a recovery would have begun in 1933 and output would have reached its 1929 level by early 1934. The 1929-33 output decrease in this case would have represented a major but not an exceptional recession. Model I assumes that the constant money growth policy would not have had any influence on money supply shocks. However, if this policy had eliminated these shocks according to the identifying assumptions of model II, then as the model II simulations in Figure 3(b) show, output would have declined only modestly (-11.1 percent cumulative and -3.30 percent at an annual rate) during 1929 to 1931 and this episode would not have been considered a major recession. Interestingly, in both models a constant money growth policy does not appreciably change Output until 1931, which is consistent with Romer's [1990] position that the 1929-30 contraction was due to nonmonetary forces-specifically, the stock market crash. It is also interesting to note that the level of output in both models stays well above the actual level not only during the Great Contraction period but also throughout the subsequent recovery period. Figure 3(c) shows that the behavior of the price level would also be significantly altered under both counterfactuals, but deflation still persists in both cases.

As discussed above, money supply shocks in model II need riot be neutral in the long run. Panel (a) of Figure 4 shows the impulse response of output to a (one standard deviation) money supply shock as identified in model II. The figure suggests that money supply shocks affect output permanently. One explanation of this result (indicating non-neutrality of money in the long run) is that monetary shocks led to a collapse of financial (highlighted by Bernanke [1983]) during the Great Contraction, and this collapse had long-term effects. We examine this issue further in the next section.

Table 2 presents some measures of the performance of the two models (as well as some other models discussed below). First, to examine the average difference between the counterfactual and actual output levels, the table shows the mean of [y.sub.t] - [y.sub.t], where [y.sub.t] represents the simulated value of [y.sub.t]. This statistic is shown for the whole 1921.1-1941.4 period as well as subperiods 1929.3 to 1941.4 and 1929.3 to 1933.1. For both models I and II the mean difference for the subperiod including the recovery is even higher than the Great Contraction subperiod. Although the main issue is the effect of the constant money growth policy on the level of output, the table also shows the ratio of the variance of [delta][y*.sub.t] to that of [delta][y.sub.t] for the whole period. As the table shows, this ratio of variances is below 1.0 for model I and much lower than 1.0 for model II. Finally, the table also provides estimates (for the whole period) of the ratio of the variance of [delta][p*.sub.t] to that of [delta][p.sub.t], where [p*.sub.t] is the simulated value of [p.sub.t]. This ratio is also below 1.0 for both models I and II.

[TABULAR DATA 2 OMITTED]

Our simulations also have implications for issues recently raised by Romer [1992]. Romer wishes to explain the determinants of the recovery from 1933 to 1941. Based on money-income multipliers calculated for the years 1921 and 1938 she simulates the behavior of real output from 1933 to 1941 if money growth had not deviated from the average (M1) growth rate of 1923 to 1927 of 2.88 percent. Such a money growth rate would have led to a 50 percent gap below potential in 1941. The fact that actual output grew substantially faster makes her case that gold-flow-induced expansionary money supply produced the recovery.

Our simulation also covers the recovery period from 1933 to 1941 but, unlike Romer, who assumed that the money supply collapsed from 1929 to 1933, as it did, our simulation assumes that a constant money growth rule for M2 was maintained over the entire interwar period. Our simulations suggest that the extra monetary stimulus Romer documents would have been unnecessary had stable money prevailed throughout the interwar period.

VI. VARIATIONS OF THE BASIC MODEL

In this section we examine the sensitivity of our results to a number of variations of our model. First, as there is controversy about the view that the long-run money demand function is stable, we examine how our results would change if the assumption that m - p and y are cointegrated is dropped. In this case the error correction term v will not appear in (2), (3) and (5). We thus resimulate models I and II using estimates of a three-variable VAR including y, p and m (without the error correction term). The performance measures for the resulting simulations (referred to as models IA and IIA) are also shown in Table 2. Compared to model I, the mean (y* -y) for model IA is lower by a small amount (.006) for the whole period but a somewhat larger amount (.010) during the Great Contraction. Comparison of model II with IIA shows similar results. The performance of a constant money growth policy thus deteriorates somewhat if the error correction term is removed from the model.

Next, we extend our model to include s or r as an additional variable as discussed in section III. For this case we consider only the simulations based on the weak counterfactual since the strong counterfactual would require additional assumptions to identify money supply shocks. The four-variable vector error correction model under the weak monetary-control assumption is labelled model IB when the fourth variable represents s and model IC when it represents r.(16) Measures of performance for these two models are also shown in Table 2. Model IB leads to a slightly higher level of output than model I during the Great Contraction, but the overall performance of these two models is not much different. Thus the introduction of variable s does not make an appreciable difference to the results. The results change more significantly under model IC. While this model produces greater variability of both output growth and inflation, it also brings about a higher level of output. On balance, its performance compares favourably to that of model I.

Our data on y and p are the Balke-Gordon quarterly series on real GNP and the GNP deflator. As these series use interpolations of annual data, they allow future information to influence current values. To explore whether this data problem may have introduced a serious bias in our results, we also estimated our basic model using quarterly averages of monthly data on industrial production and the wholesale price index as alternative measures of y and p. The weak and strong versions of the=basic model based on the alternative data set are labelled ID and IID. Measures of performance for these two models in Table 2 show that although the effect of the constant money growth policy on the variability of [delta]y and [delta]p differs between the two models, the policy substantially increases the level of output, especially during the Great Contraction, in both models. Indeed, in terms-of the effect on the level of output, the case for the constant money growth policy is stronger if the data-on industrial production and the wholesale price index are used instead of the Balke-Gordon estimates (compare model ID with I and IID with II).

Our simulations assume not only that a constant money growth policy would not have significantly altered the output and price equations but also that these equations were stable throughout the interwar period. This assumption is subject to the Lucas [1976] critique since the Great Contraction could have represented a regime shift that would have caused the parameters of the reduced-form relations to change. To explore this issue, we tested our model for possible breaks at three different dates, 1930.1, 1931.4 and 1933.2, which represent the beginning, the middle and the end of the Great Contraction. The Chow test of a break at a fixed date indicates a shift in both the output and price equations in 1930.1 as well as in 1933.2. It may be thought that an equation explaining nominal income would have been more stable than our separate output and price equations. However, testing a comparable nominal-income equation for stability, the Chow test indicates a break in this relation in 1933.2, though not in 1930.1.(17)

If, as the above-evidence indicates, the Great Contraction was responsible for shifts in the output and price relations, then these shifts would not have occurred under a constant money growth policy that prevented the Contraction. This possibility suggests using :relations estimated for the stable 1920s to construct the constant money growth counterfactual for the whole interwar period. For this simulation, the basic model was estimated separately for the subperiods 1921.1-1929.4 and 1930.1-1941.4 to calculate reduced-form shocks for the whole period. Then, using output and price equations estimated for the 1921.1-1929.4 subperiod and letting [theta]= .00822 (the mean value of [delta]m for this subperiod), the behavior of m, y and p was simulated under both the weak and strong monetary control assumptions. These simulations are referred to as models IE and IIE, respectively, and are illustrated in Figure 5.

As Figure 5 shows, the behavior of output y and the price level p under both model IE and IIE is dramatically different from previous simulations. The most remarkable difference is in the behavior of y which, unlike in previous cases, does not exhibit any significant contraction during the 1929-33 period under both the weak and strong forms of the constant money growth rule. This result fully supports the underlying assumption of the simulation that by preventing the Great Contraction the constant money growth policy would have preserved the 1920s output and price relations. Performance measures for models IE and IIE in Table 2 show that although these models would have made a marginal difference to the variability of the price level, they would have produced a substantially higher level of output and a significantly lower output variability.

As noted above, the collapse of financial intermediation could have accounted for the result that money shocks had long-term effects on output in model II. If money is otherwise neutral in the long run, we would not expect money shocks to exert a permanent effect on output in a model based on the period before the financial collapse. This prediction is supported by panel (b) of Figure 4, which shows that in contrast to model II, [[eta].sup.m] does not have a marked long-run effect on y in model IIE.

VII. CONCLUSION

This paper reconsiders the long-debated question of whether stable money could have averted the Great Contraction in the United States. Our approach follows McCallum's methodology of using an empirical model of the economy based on interwar data to examine how a counterfactual policy would have performed. Our study, however, departs from McCallum's in considering a different monetary policy rule and estimating separate relations for output and the price level. We also allow the form of our rule to depend on information available to the Fed and on contemporaneous relations between money and other variables.

We consider two variants of Milton Friedman's constant money growth rule that represent limiting cases for a wide range of assumptions about monetary control. While one of these variants assumes that output and the price level do not affect money in the same period and lets the=Fed observe money supply shocks contemporaneously, the other allows any pattern of instantaneous causality between money and economic activity and assumes that the Fed can react to shocks with only a one-period lag.

Basic simulations of both variants produce results consistent with claims that, had a stable money policy been followed, the Great Contraction would have been mitigated and shortened. A severe recession would have occurred but not of the extraordinary character of the Great Contraction. If it is further assumed that a stable money policy would have prevented shifts in the output and price relations, our alternative simulations produce the much stronger result that there would have been no recession at all during the Great Contraction under both variants of the constant money growth rule.

If our simulations are valid, then the extra monetary stimulus that Romer [1992] argues was essential to produce recovery after 1933 would not have been necessary. In fact, under all of our constant money growth counterfactuals, simulated output is well above the actual output throughout the 1933 41 period.

Eichenbaum [1992] has argued that, "After all, aside from the time consistency issue, there is little theoretical reason to recommend a k percent rule." The view that a k percent rule (our constant money growth rule) is suboptimal compares economic performance under constant money growth with alternative rules or discretions that yield a superior outcome. Our focus, however, is on the constant money growth policy relative to actual performance during the Great Depression. On that basis, constant money growth was clearly preferable.

[Figures 1 to 5 ILLUSTRATION OMITTED]

APPENDIX A

Data Sources

Quarterly Data

Real-Gross National Product and Gross National Product Implicit Price Deflator: Balke-Gordon [1986], Table 2.

Suspended Bank Deposits: Federal Reserve Bulletin, September 1937, 909; McCallum [1990].

Money Stock (M2): Friedman and Schwartz [1970], Table 2, 61-73.

Monetary Base: Friedman and Schwartz [1963], Appendix B, Table B-3, 799-808.

Interest Rate: Commercial paper rate in Balke-Gordon [1986], Table 2.

Monthly Data

Consumer Price Index: 1910-40, NBER Business Cycle Series no. 04072; 1941-61, NBER Business Cycle Series no. 04052.

Industrial Production: 1919-66, NBER Business Cycle Series no. 01303.

Money Stock (M2): Friedman and Schwartz [1970], Table 1, 8-52.

(1.) See also Feldstein and Stock [1992] who simulate the effect of an optimal monetary policy rule on nominal GDP over the 1959-92 period.

(2.) The argument in chapter 10 of Eichengreen [1992b], that the Federal Reserve would not have been able to expand the money supply in he 1931 episode because it would have violated the free gold constraint and forced the United States off the gold standard, is incorrect, as shown by Bordo [19941. First, the author argues that since M1 fell by $2 billion between August 1931 and January 1932, a $2 billion open market operation would have been required to offset the decline, and this would have forced the United States, with only $400 million in free gold, off the gold standard. This argument ignores the arithmetic of money supply relationships. The ratio of M1 to high-powered money fell between August 1931 and January 1932 from 3.2 to 2.8--the average money multiplier over the period was approximately 3. Thus an open market operation to increase the monetary base by $600 650 million was all that would have been needed if no other forces were at work. According to Friedman and Schwartz 11963], other forces were present so that a $1 billion purchase would have been needed. But would the $1 billion purchase have exhausted free gold? It would have if the open market purchase, as Eichengreen assumes, were absorbed by a dollar-for-dollar increase in currency, which had to be backed 40 percent by gold and 60 percent by eligible paper. But if the purchase had increased bank reserves by $1 billion, which would have had the effect of calming depositors' fears about bank safety, then free gold would not have been exhausted.

(3.) We use the Balke-Gordon [1986] data based on the Kuznets-Commerce sources. Romer 11988] has criticized this data set and constructed her own series. Her data considerably reduce the severity of the 192324, 1926-27, and 1937-38 recessions but not that of the Great Contraction of 1929-33.

(4.) We also estimated the model using M1. The results were quite similar.

(5.) Romer [1990] attributes the decline in output from 1929 to 1930 to the effects of the stock market crash, not the decline in the money stock.

(6.) Temin [1978] disputes Friedman and Schwartz's contention that banking panics were the cause of the collapse in the money supply His regressions show that the banking failures of the 1930s entirely reflected the decline in economic activity. A number of subsequent papers refute this finding as in Anderson and Butkiewicz [1980], Wicker 11980], Mayer [1978], Boughton and Wicker [1979].

(7.) McCallum [19901 uses this measure. Bernanke [1983] uses the real value of suspended bank deposits as a proxy for the nonmonetary influence of the banking failures on economic activity.

(8.) According to White [19841, however, the majority of banks that failed after 1930 had characteristics similar to those that failed in the 1920s.

(9.) See, for example, Blanchard [1990].

(10.) For each of these variables, the augmented Dickey-Fuller test (both with and without a time trend) does not reject the hypothesis that the variable contains a unit root.

(11.) See Lucas [1988] who argues that U.S. long-run money demand is stable. Also see Hafer and Jansen [1991] for evidence supporting a cointegrated money demand relation for M2 over the period from 1915 to 1988. For a contrary view, see McCallum [1993] who argues that shocks to the money demand function are likely to include permanent stochastic changes in technology affecting transactions.

(12.) In the case where [z.sub.t]= [s.sub.t], s is included in (2) and (3) for reasons discussed by Bernanke [1983].

(13.) Existing evidence, for example, Lucas 11988] and Hoffman and Rasche 11991], suggests that the income elasticity of money demand is close to one. The assumption that [[alpha].sub.2] = 0 is relaxed later in simulations based on the extended model that includes the interest rate. Note that [[alpha].sub.0] (the intercept in the demand for money) is left out and is included in the constant terms of the regressions.

(14.) We also tried four lags for each variable but the additional lagged terms were not significant.

(15.) It might be argued that at the onset of the Great Contraction, the Fed would not have known the average interwar money growth rate and, thus, it would be more appropriate to set [theta] equal to some rate reflecting the 1920's experience. The average value of [delta]m for 1921-29 is, in fact, greater than the interwar average. As discussed above, however, we eschew using a higher rate of money growth in the present simulations on the grounds that estimated parameters of our model are likely to reflect the average experience of the whole interwar period. In the next section, we do set [theta] equal to the average [delta]m value for 1921.11929.4 in an alternative simulation based on a model estimated for the same period.

(16.) In model IC, v is measured as m - p - y [[alpha].sub.2]r with [[alpha].sub.2]=-.04. This estimate of a2 was derived by applying the dynamic OLS procedure in Stock and Watson [1989] to the money demand relation (1), constraining [[alpha].sub.1] equal to one and using four leads and lags.

(17.) Nominal income growth was regressed on a constant, lagged value of v, three lagged values of [delta]m and three of its own lagged values. McCallum's [1990] nominal-income equation is slightly different (it includes only one lagged value of [delta]m and no error-correction term) but this equation also breaks in 1933.2 according to the Chow test. However, as McCallum notes, a break at this date is not indicated if two dummy variables for the 1930.2 and 1930.3 quarters are also included in the equation.

REFERENCES

Anderson, Barry, and James Butkiewicz. "Money, Spending, and the Great Depression." Southern Economic Journal, October 1980, 388-403.

Balke Nathan S., and Robert J. Gordon. "Historical Data" in The American Business Cycle: Continuity and Change, edited by R Gordon. Chicago: University of Chicago Press, 1986, 781-850.

Bernanke, Ben. "Nonmonetary Effects of the Financial Crisis in the Propagation Of the Great Depression." American Economic Review, June 1983, 25776.

Bernanke, Ben, and Mark Gertler. "Agency Costs, Net Worth and business Fluctuations." American Economic Review, March 1989, 14-31.

--. "Financial Fragility and Economic Performance." Quarterly Journal of Economics, February 1990, 87-114.

Betts, Caroline, Michael Bordo and Angela Redish. "A Small Open Economy in Depression: Lessons from Canada in the 1930's." Canadian journal of Economics, (1995, forthcoming). Blanchard, Olivier. "Why Does Money Affect Output? A Summary," in Handbook of Monetary Economics, edited by B. M. Friedman and E H. Hahn. Amsterdam: North Holland Publishers, 1990, vol. 2, chapter 15.

Bordo, Michael. "The Contribution of A Monetary History of the United States, 1867-1960 to Monetary History," in Money, History, and International Finance: Essays in Honor of Anna J. Schwartz, edited by M. D. Bordo. Chicago: University of Chicago Press, 1989, 15-70.

--. "Review of Barry Eichengreen, Golden Fetters: The Gold Standard and the Great Depression." Journal of International Economics, February 1994, 193-97.

Boughton, James M., and Elmus R. Wicker. "The Behavior of the Currency-Deposit Ratio during the Great Depression." Journal of Money, Credit and Banking, November 1979, 405-18.

Brunner, Karl, and Allan H. Meltzer. "What Did We Learn from the Monetary Experience of the United States in the Great Depression?" Canadian Journal of Economics, May 1968, 336-48.

Calomiris, Charles. "Financial Factors in the Great Depression." Journal of Economic Perspectives, 7(2), Spring 1993, 61-85.

Cecchetti, Stephen G. "Prices During the Great Depression: Was the Deflation of 19.30-1932 Really Anticipated?", American Economic Review, March 1992, 141-56.

Cecchetti, Stephen G., and Georgios Karras. "Sources of Output Fluctuations During the Interwar Period: Further Evidence on the Causes of the Great Depression." Review of Economics and Statistics, February 1994, 80-102.

Eichenbaum, Martin. `Comment on `Central Bank Behavior and the Strategy of Monetary Policy: Observations from Six Industrialized Countries' by Ben Bernanke and Frederic Mishkin," in NBER Macroeconomics Annual 1992, edited by O. Blanchard and S. Fisher, 1992, 233.

Eichengreen, Barry. "The Origins and Nature of the Great Slump Revisited." Economic History Review, May 1992(a), 213-39.

--. Golden Fetters: The Gold Standard and the Great Depression, 1929-Z939. New York: Oxford University Press, 1992(b).

Eichengreen, Barry, and Jeffrey Sachs. "Exchange Rates and Economic Recovery in the 1930s." Journal of Economic History, December 1985, 925-46.

Evans, Martin, and Paul Wachtel. "Were Price Changes during the Great Depression Anticipated?" Journal of Monetary Economics, August 1993, 3-34.

Feldstein, Martin, and James Stock. "The Use of a Monetary Aggregate to Target Nominal GDP," in Monetary Policy, edited by N.G. Mankiw, Chicago: Chicago University Press, 1994, 7-62.

Field, Alexander J. "Asset Exchanges and the Transaction Demand for Money, 1919-1929." American Economic Review, March 1984, 43-59.

Fisher, Irving. "The Debt-Deflation Theory of Great Depressions." Econometrica, October 1933, 339

Friedman, Milton. A Program for Monetary Stability. New York: Fordham University Press, 1960.

Friedman, Milton, and Anna J.-Schwartz. A Monetary History of the United States, 1867-1960. Princeton: Princeton University Press, 1963.

--. Monetary Statistics of the United States. New York: Columbia University Press for NBER, 1970.

Hafer, Rik W., and Dennis W. Jansen. "The Demand for Money in the United States: Evidence from Cointegrating Tests." Journal of Money, Credit and Banking, May 1991, 155-68.

Hamilton, James D. "Monetary Factors in the Great Depression." Journal of Monetary Economics, March 1987, 145-70.

--. "Was the Deflation During the Great Depression Anticipated? Evidence from the Commodity Futures Market." American Economic Review, March 1992, 157-78.

Hoffman, Dennis, and Robert H. Rasche. "Long-Run Income and Interest Elasticities of Money Demand in the United States." Review of Economics and Statistics, November 1991, 665-74.

Kindleberger, Charles. The World in Depression. Berkeley: University of California, 1973.

Laidler, David. The Golden Age of the Quantity Theory. Princeton: Princeton University Press, 1991.

Lucas, Robert E., Jr. "Economic Policy Evaluation: A Critique," in Carnegie-Rochester Conference Series on Public Policy, edited by K. Brunner and A. H. Meltzer. Amsterdam: North-Holland, 1, 1976, 1946.

--. "Money Demand in the United States: A Quantitative Review," in Carnegie-Rochester Conference Series on Public Policy, edited by K Brunner and A. H. Meltzer, 29, 1988, 137-68.

McCallum, Bennett T. "Could a Monetary Base Rule Have Prevented the Great Depression?" Journal of Monetary Economics, August 1990, 3-26.

--. "Unit Roots in Macro-economic Time Series: Some Critical Issues." Federal Reserve Bank of Richmond Quarterly, 79(2), Spring 1993, 13-44.

Mayer, Thomas. "Money and the Great Depression: Some Reflections on Temin's Recent Book." Explorations in Economic History, April 19.78, 127-45.

Mishkin, Frederic S. "The Household Balance Sheet and the Great Depression." Journal of Economic History, December 1978, 918-37.

Modigliani, Franco. "Some Empirical Facts of Monetary Management and Rules Versus Discretion." Journal of Political Economy, April 1964, 211-45.

Nelson, Daniel B. "Was the Deflation of 1929-30 Anticipated? The Monetary Regime as Viewed by the Business Press." Research in Economic History, 13, 1992, 1-66.

Romer, Christina D. "World War I and the Postwar Depression: A Reinterpretation Based on Alternative Estimates of GNP." Journal of Monetary Economics, July 1988, 91-115.

--. "The Great Crash and the Onset of the Great Depression." Quarterly Journal of Economics, August 1990, 597-624.

--. "What-Ended the Great Depression?" Journal of Economic History, December 1992, 757-84.

--. "The Nation in Depression." Journal of Economic Perspectives, 2, Spring 1993, 19-39.

Schwartz, Anna J. "Understanding 1929-1933," in The Great Depression Revisited, edited by K. Brunner. Boston: Martinus Nijoff, 1981, 5-48.

Stock, James H., and Mark W. Watson. "A Simple MLE of Cointegrating Vectors in General Integrated Systems." NBER Technical Working Paper No. 83, 1989.

Temin, Peter. Did Monetary Factors Cause the Great Depression? New York: Norton, 1978.

--. Lessons from the Great Depression. Cambridge: MIT Press, 1989.

--. "Transmission of the Great Depression." Journal of Economic Perspectives, 7(2), Spring 1993, 871-102.

U.S. Department of Commerce. Long-Term Economic Growth, 1860-1970. Washington, D.C.: U.S. Government Printing Office, 1973.

Warburton, Clark. "Monetary Theory, Full Production, and the Great Depression," in Depression, Inflation, and Monetary Policy. Baltimore: Johns Hopkins University Press, 1966, chapter 5.

Wheelock, David. "Monetary Policy in the Great Depression: What the Fed Did and Why." Federal Reserve Bank of St. Louis Review, March/April 1992, 3-28.

White, Eugene N. "A Reinterpretation of the Banking Crisis of 1930." Journal of Economic History, March 1984, 119-38.

Wicker, Elmus R. "Federal Reserve Monetary Policy, 1922-33: A Reinterpretation." Journal of Political Economy, August 1965, 325-43.

ANNA J. SCHWARTZ, Professor, Rutgers University, Professor, Carleton University; Research Associate, National Bureau of Economic Research. We thank the Harry and Lynde Bradley Foundation for its support. For helpful comments and suggestions we thank Ben Bernanke, Steve Cecchetti, Charles Evans, Bennett McCallum, Bob Rasche, Pierre Siklos, Jim Stock, Paul Wachtel, Tom Willett, and participants at the Rutgers Mini-Conference on the Great Depression, February 1993, and the NBER Macro History Conference, April 1993. For able research assistance we thank Jakob Koenes.