Was the Great Defl ation of 1929 – 33 inevitable ?

Q is paper re-examines the recent and provocative hypothesis that the Great De\ ation of 1929–33 was an inevitable consequence of the return to the gold convertibility of currencies at pre-war parities. An alternative hypothesis, that the relative prices of gold tended to gravitate to one another, is put forward in this paper. Q is hypothesis is derived from the conventional gold standard model and Cassel’s well known insights on purchasing power of currency. Empirical evidence lends support to the alternative hypothesis: even though the relative price of gold returned to its pre-war level by 1931, the adjustment process was mainly driven by di[ erences between countries rather than the absolute deviation from the pre-war level.


INTRODUCTION
One of the most intriguing economic phenomena was the Great Depression.ere is a myriad of hypotheses on the causes and nature of the Great Depression, e.g. a collapse in consumption (Romer, 1990 and1993), failures in monetary policy (Friedman and Schwartz, 1963;Hamilton, 1987), competition for gold reserves (Irwin, 2012), failures in the operation of international monetary system (Hamilton, 1988;Temin, 1993), a lack of cooperation between central banks (Eichengreen, 1992).
Even today, after more than eight decades from the onset of the Depression, the debate has not seemed to die out.e reason was vividly stated by Bernanke (1995): '[t]o understand the Great Depression is the Holy Grail of macroeconomics.' In a recent paper Mazumder and Wood (2013) made an attempt to explain the Great De ation anew.eir hypothesis is that 'the Great De ation of 1929-33 was inherent in the operation of the gold standard once a country decided to return to pre-war parity following its suspension and wartime in ation.' ey point to the fact that 'the value (general purchasing power) of a convertible currency must be the relative cost of producing gold and other products, which did not change signi cantly between 1914 and the 1930s.' us, the price fall of 1929-33 was an inevitable consequence of the mismanagement of the gold standard which lay in the resumption of gold convertibility at pre-war parities and the resultant overvaluation of cur-rencies.Moreover, macroeconomic policies were of secondary (if any) importance with their impact limited, at most, to the de ation's timing.
One terminological comment should be made here.Mazumder and Wood (2013) use the term Great De ation and not Depression since they focus on price levels only.One should not overemphasize that difference because de ationary forces were at the heart of the Great Depression.Indeed, Mazumder and Wood (2013) admit that even though real e ects are not discussed in their analysis they 'may be surmised.'For Bernanke (1995) 'no account of the Great Depression would be complete without an explanation of the worldwide nature of the event, and of the channels through which de ationary forces spread among countries' (emphasis added).e channels linking de ation and depression are analysed in detail by Bernanke and James (1991).It is also interesting to observe that Knakiewicz (1967) used the term 'de ation' in her description of the course of economic events in Poland in 1930-1935.e objective of this paper is threefold.First, the hypothesis that the Great De ation was inevitable is re-examined.It is demonstrated that even though this hypothesis conforms well to the conventional gold standard model developed by Barro (1979), it is oriented at the long-run (ultimate) equilibrium and does not explain the adjustment process towards it.us, its main weakness is the lack of a credible explanation of the medium-run equilibrium and the de ation's timing.Second, a competing hypothesis is put forward, which says that the relative prices of gold tended to gravitate to one another.Its theoretical underpinnings are derived from the conventional gold standard model and Cassel's insights on purchasing power of currency.ird, both hypotheses are confronted on empirical grounds.
e paper is organized in a standard way.In Section 2, Mazumder and Wood's hypothesis is shortly presented.Its drawbacks are discussed in Section 3. eoretical foundations of the alternative hypothesis are explained in Section 4; Section 5 compares these two hypotheses.Results of empirical tests are discussed in Section 6 and conclusions are o ered in the last Section.Mazumder and Wood (2013) use a simpli ed version of the model presented by Barro (1979) to explain the inevitability of the Great De ation.e model describes a world economy with full employment, fully exible prices and xed exchange rates.Money supply, M s , and money demand, M d , are respectively

ARGUMENTS IN FAVOUR OF THE GREAT DEFLATION'S INEVITABILITY
where is the money multiplier or the reciprocal of the cover ratio, P g is the xed nominal price of gold, G m is the monetary gold stock, P is the general price level of commodities, Y is the aggregate income (output), and L is a decreasing function of the nominal rate of interest, i.In equilibrium the relative (real) price of gold,

P P g
, is and it holds at all times (it is assumed that equilibrium in money market holds continuously).e monetary gold stock can change if there is a di erence between the supply of new gold, g s (supply is derived from the pro t maximisation by gold producers), and the demand for non-monetary gold, Using the above equations, Mazumder and Wood (2013) explain adjustments to the 1913-25 in ation.Assume that monetary policy is relaxed, i.e. the cover ratio is allowed to decrease, and goes up.us, the relative price of gold falls (equation ( 2)), gold producers are discouraged from supplying gold and demand for non-monetary gold rises.Both these factors cause the change in monetary gold stock to decrease.It could even become negative, m G .According to Mazumder and Wood (2013), this is su cient to justify the conclusion that 'the nal position includes a return to the original P and stock of money.' eir line of reasoning is based on the long-run solution of the Barro's model: the relative price of gold in the world economy under the gold standard returns in the long run to its initial level when disturbed by changes of velocity, changes in the price of gold or in the gold backing of monetary base.It does this because 'the value (general purchasing power) of a convertible currency must be the relative cost of producing gold and other goods, which did not change signi cantly between 1914 and 1930s' (Mazumder and Wood, 2013).

ARGUMENTS IMPAIRING THE GREAT DEFLATION'S INEVITABILITY
ough long-run equilibrium is no doubt important, a reliable explanation of the Great De ation cannot neglect the speci cs of medium-run adjustment or, more generally, the De ation's timing.Unfortunately this is an issue that Mazumder and Wood (2013) do not try 'to be precise about.' ere are three problems with the adjustment process that cast some doubt on its pace: sluggishness of changes in non-monetary demand, loose observance of the rules of the game by the central banks and rough empirical evidence.
First, following Barro (1979) and Barsky and Summers (1988), the demand for non-monetary gold, g n d , is assumed to be a di erence between the target, f(•), and actual stocks of gold, G n , held by private agents where is the pace of adjustment, π g is the expected change in the relative price of gold and r is the real interest rate.Due to wartime in ation g P P was indeed low and stimulated non-monetary demand via the increase in target stock.e pace of adjustment, however, could be and actually was rather small because there was no opportunity of pro table arbitrage.e central bank was ready to stabilize the price of gold so there was no reason to change the stock of non-monetary gold in an abrupt way.As explained by Barro (1979), currency was equally good as gold so there was no reason for speculative behaviour.
e additional factor that contributed to rather small changes in non-monetary demand was a positive correlation between interest rates and price level (the Gibson paradox).As demonstrated by Barsky and Summers (1988), a rise in the real interest rate pushed down both the long-run level of the relative price of gold and the target stock of non-monetary gold, f(•).us, the return to the gold standard at a relatively high real interest rate due to wartime expenditure and higher public debt put downward pressure on the di erence between target and actual stocks of non-monetary gold.
Second, so far we have focused on the implications of equations ( 3) and ( 4).But even if one assumes that low g P P makes the monetary gold stock smaller, as argued by Mazumder and Wood, this does not necessarily result in a proportionate change in the stock of money.For this assumption to hold the central bank has to observe the rules of the game, i.e. change domestic credit to reinforce the impact of gold ows onto the monetary base.In the words of McCloskey and Zecher (1981), '[a]n alternative indicator of the extent to which central bankers played the rules is the extent to which the relationship between in ows of gold and increases in domestic credit ... was positive'.
is, however, was not the case in the inter-war period.Nurkse (1945) pointed out that '[i]n the inter-war period neutralization of gold movements by central banks became, in fact, the rule rather than exception ... any change in a central bank's gold and foreign reserve was accompanied by a change in the opposite direction in the bank's domestic loans and securities' (emphasis in original).More detailed data on changes in central banks' international and domestic assets tabulated by Nurkse (1944, p. 69) show that the changes in the same direction were rather rare and their share in the total number of observations ranged from 19 to 35 per cent between 1928 and 1931, i.e. the period when the gold standard system was in full operation. 1e weaker the observance of the rules of the game, the slower the adjustment of the stock of money and the price level.
Finally, the simple empirical evidence is in stark contrast with the adjustment process suggested by Mazumder and Wood (2013).Using the data provided by these authors (see their Table 2), one can see that the average annual rate of change of non-monetary gold decreased from 2.5 per cent in 1914-25 to 1.3 per cent in 1926-33.e rate of accumulation of monetary gold increased from 2 to 2.7 per cent in the same period.us, the data point to changes in the demand for non-monetary gold and changes in monetary gold stock that are opposite to those suggested by Mazumder and Wood's hypothesis.
Preliminary inspection of data on de ation also reveals the problem with Mazumder and Wood's explanation.e case of France is very instructive here.In ation during and after the war pushed French prices far above the levels observed in the United Kingdom or the United States.In June 1928, when the gold standard was de jure re-established in France, the wholesale price index stood at the level of 611 (in 1914 it was 100).e parity adopted was 4.925 new francs for 1 old franc.Taking into account parity adjustment, one could expect that the French economy was less prone to de ation than the U.S. or British economies.e data, however, do not con rm such a conjecture (Table 1).Price de ation in France measured by the changes in wholesale prices was comparable to that experienced by the U.S. economy and almost the same as in the United Kingdom in the period that ended with the sterling crisis in September 1931.In the whole period of the Great De ation, 1929-33, it was not weaker but more severe in France than in the U.K., U.S., or German economies.
Table 1 Relative prices of gold (based on the wholesale price indices, 1914 = 100) and de ation.
Note: Common base is adopted, 1914 = 100, and data are seasonally adjusted.
Source: Wholesale prices on a monthly basis: for France series m04057 and for the UK -m04053 from the on-line NBER Macrohistory Database.Data for France are corrected for the change in gold parity to make it comparable to other countries.For Germany, data are from the Statistical Year-book of the League of Nations.U.S. data are from the on-line Federal Reserve Economic Database.

CASSEL'S INSIGHTS ON PURCHASING POWER OF CURRENCIES
Since mere gold resumption at pre-war parities does not seem to be a convincing explanation for the dynamics of price changes both before and during the Great De ation of 1929-33, an alternative hypothesis is o ered in this Section.It is based on the concept of the purchasing power of currency promoted by Cassel and his two insights into the stability of monetary system.First, more than one country under the gold standard system is needed to stabilize the value of gold (a concept of a 'centre of stability').Second, the value of a currency stems from 'the fact that this money possesses a purchasing power as against commodities and services' (Cassel 1922, p. 138; see also Cassel, 1916 and1918).
Cassel's plan for the restoration of the gold standard was to reintroduce the pre-war parity between internal purchasing powers of sterling and dollar by stabilizing their values and creating in this way a centre of stability.e price level chosen for that purpose was not necessarily the pre-war level (Cassel, 1928).It should rather be the one that 'can most easily and most rapidly be obtained on both sides' (Cassel 1922, p. 262).e centre of stability 'would manifestly bring about a certain stabilisation, not only of the internal values of these currencies [dollar and sterling], but also the value of gold itself ' (Cassel 1922, p. 261). 2 Moreover, it would encourage other countries to resume gold convertibility.His point was that the parity should be set with an eye to the relation between domestic and the U.K./U.S. price levels and not to the relative price of gold or pre-war parity (Cassel 1922, pp. 140-1, 260, 264-5).
Barro's model can be used to for malize the alternative hypothesis put forward in this paper.Admittedly, Barro assumes that the relative price of gold is the same across countries, but this can be easily modi ed.It follows from money market equilibrium in each country j (see equation ( 2)) that m j j g j j G z P P j n (5) where j j j j j z L i Y .e crucial observation is that since gold can ow from one country to another the relative prices of gold will tend to equalize in the medium term Using the medium-run equilibrium condition (6) it is straightforward to nd the medium-run stock of monetary gold stock, m j G , for each country Equations ( 5) and ( 8) imply that if the relative price of gold in a given country is above its medium-run level ( g j j g P P P P ) then the medium-run level of monetary gold stock is above actual stock ( m j m j G G ). us, gold will ow into such a country putting a downward pressure on the relative price of gold.e opposite is true for a country with the relative price of gold below its medium-run level.
It is true that in the long run g P P will indeed return to its steady state level which-as demonstrated by Barro (1979)-does not depend on the monetary policy of a single or even all the countries under the gold standard.Such an adjustment, however, does not undermine the medium-run adjustment.Both these adjustment processes can work simultaneously.
Mazumder and Wood's hypothesis rests on the long-run adjustment process, whereas the alternative hypothesis rests on the medium-run adjustment.As it is clear from the analysis above, the medium-run adjustment process is driven by di erences between countries and not within a given country.us, according to the alternative hypothesis under the gold standard the relative prices of gold gravitate to one another and not necessarily to the long-run level implied by the conventional gold standard model.

TWO HYPOTHESES CONFRONTED
Both hypotheses are presented in a similar way to facilitate their comparison and empirical analysis in the next Section.Mazumder and Wood (2013) assume that the pre-war relative price of gold is an approximately good description of the steady state of g P P .us, their hypothesis, denoted as H MW , can be summarised as and all P g 's and P's are indices 1913 = 100.In other words, the post-war relative prices of gold were expected to return to the common pre-war level.Any deviation from this level would mean that gold was too cheap (expensive) and more de ation (in ation) was needed to restore the long-run equilibrium.
According to the alternative hypothesis, it is the relation between the relative prices of gold across countries that matters.It can be expressed as where the United States are treated as a baseline country.e ratio C j t is an equivalent to the real exchange rate of a currency j against the U.S. dollar.Since the real exchange rate compares purchasing power of two currencies, it is justi ed to say that the hypothesis H C is focused on the relative purchasing power of a currency.e currency j is overvalued in real terms against the U.S. dollar if C j t is above 100.If so, one should expect de ation in that country in excess of that in the United States or at least lower in ation than in the United States in the medium run.e opposite would be true for a country with undervalued currency.
e hypotheses H MW and H C are not independent one from another: the former can be seen as a special case of the latter.e return of the relative price of gold to its pre-war level (H MW ) is su cient but not necessary for the convergence of relative purchasing powers (H C ).Moreover, the convergence of relative purchasing powers (H C ) is necessary but not su cient for the return of the relative price of gold to its pre-war level (H MW ).Because of these relations one can expect three basic empirical results.First, both hypotheses can be rejected by data suggesting that the gold standard model needs revision.Second, both can be supported by the empirical analysis.In such a case, long-run adjustment would be strong enough to explain the Great De ation, so the H MW hypothesis should be given precedence over the alternative H C .ird, empirical evidence can lend support to the H C hypothesis and reject the H MW hypothesis. is would suggest that the medium-run adjustment is robust with tight linkages between relative prices of gold across countries and the long-run adjustment is sluggish and cannot explain the Great De ation.
e common feature of both hypotheses is that a certain ratio, MW j t or C j t , adjust to its equilibrium level.us, both imply the following price-adjustment equation: where MW for Mazumder and Wood's hypothesis and C for the hypothesis based on Cassel's insights.A constant is the equilibrium level, and under both hypotheses it is expected to be 100.Two important restrictions on the parameters of equation ( 11) are imposed by each hypothesis.First, should be negative.Otherwise the adjustment would run in the wrong direction, i.e. any deviation of the relative price of gold or relative purchasing power of currency from their equilibrium levels would increase over time even if there were no disturbances ( j t u ).It is also important that is large enough to imply a non-negligible speed of adjustment (economic signi cance).Second, since the equilibrium levels implied by both hypotheses are 100, i.e. the relative prices of gold return to their pre-war level (H MW ) or they gravitate to one another (H C ), coe cient is expected to be insigni cant.If it turns out to be signi cant then the implied equilibrium level would be .For instance, if (and ) then the implied equilibrium level is above the one suggested by the hypothesis.

DATA AND EMPIRICAL RESULTS
Annual data on the relative price of gold were collected for 26 countries: Argentina, Australia, Austria, Belgium, Bulgaria, Canada, Chile, Denmark, Egypt, France, Germany, Greece, Hungary, India, Italy, Japan, the Netherlands, New Zealand, Norway, Peru, South Africa, Spain, Sweden, Switzerland, the United Kingdom and the United States.e sample covers period of 1925-35.All the data are from various issues of the Statistical Yearbook published by the League of Nations.
Equation ( 11) resembles the equation used in unit root tests.Indeed, if π H is a non-stationary I(1) process, then one should expect to be insigni cant.In other words, if a unit root is found in π H the adjustment mechanism implied by the hypothesis H does not work and therefore the hypothesis should be rejected.Since the dataset covers 26 countries the relevant tests are panel unit root tests.e results are provided in Table 2.
e general nding from these tests is that the results are inconclusive for both relative price of gold and relative purchasing power.In speci cations with the intercept only, the null of unit root is not rejected, but this is not the case in speci cation with the trend.Moreover, no matter which speci cation is used, the Levin, Lin and Chu tests suggest that both variables are stationary but the opposite is implied by the PP tests.
e ambiguity of the results could be related to the fact that unit root tests have low power to reject the (false) null of unit root.us, equation ( 11) is used to estimate directly and then residuals, j t u , are checked for stationarity.e results for the period 1927-35 are presented in Table 3. ree alternative estimation methods have been used for each hypothesis: OLS, cross-section xed e ects and period xed effects. 3Regression (1) shows that the adjustment coe cient 1 is negative as expected but insigni cant.Even if it were signi cant, the implied pace of adjustment would be very slow with the half-life of a shock more than six years.us, there is little empirical support for the H MW hypothesis.Results of the OLS regression (4) for the relative purchasing power are di erent: not only is adjustment statistically signi cant but the coe cient is large and important from an economic point of view.e half-life of a shock is slightly less than two years.e alternative hypothesis H C gains more support from the data.Since the unexplained variation of the dependent variable is rather large in both regressions (R's-squared are low) it could be useful to add xed e ects.us, cross-section xed e ects are introduced in regressions (2) and ( 5). e picture, however, does not change substantially: the H MW hypothesis still receives less support than the alternative H C .e speed of adjustment is greater in both cases with the half-life of a shock equal to less than six years for H MW and one year for H C .Tests of signi cance of xed e ects show that the e ects are redundant in regression (2) but not in regression (5).
Finally, the regressions with period xed e ects are run.e results of regressions ( 3) and ( 6) are quite similar one to another: adjustment speed is both statistically and economically signi cant (the half-life of a shock is slightly less than three years) and xed e ects are found not to be redundant.Before concluding that both hypotheses gain comparable support when this speci cation is used, it is worthwhile examining period xed e ects in more detail.Generally, they are supposed to account for any time-speci c e ect (Baltagi, 2005, p. 33).According to equation ( 11) they, together with the intercept 0 , modify the equilibrium level of variable π.
In Figure 1 such modi ed equilibrium levels of π implied by regressions ( 3) and ( 6) together with minus/plus two standard error bands are illustrated.Figure 1 covers the period in which the gold standard was fully in operation.According to the H MW hypothesis the restoration of the gold standard with pre-war parities was 'a su cient explanation' of the Great De ation (Mazumder and Wood, 2013).us, one would expect the implied equilibrium level of the relative price of gold to stay close to 100, i.e. the pre-war level.
is, however, is not the case: the implied equilibrium was around 80 in 1929 and then jumped to more than 130 in 1930 (the degree of uncertainty around this estimate is large).e picture looks much more promising for the H C hypothesis: the implied equilibrium level of relative purchasing power was not only quite close to 100 but its uctuations were moderate.All the regressions have been run on the shorter sample of 1927-1931, i.e. excluding the period when the international gold standard started to disintegrate after the sterling crisis in September 1931.e results are very similar to those in Table 3 (not reported).
Overall, empirical evidence supports the view that the relative prices of gold were not so much driven towards their pre-war levels as gravitated to those in other countries.e hypothesis H C , therefore, provides a more accurate description of the actual price behaviour than the H MW hypothesis.

CONCLUSION
is paper re-examines the hypothesis that the seeds of the Great De ation of 1929-33 were sown by the central banks who decided to resume gold convertibility at pre-war parities.ese were inappropriate and implied overvaluation of the currencies because the price levels were far above the levels recorded in 1914.
It is demonstrated that the adjustment mechanism implied by that hypothesis was far from being effective due to the sluggishness of changes in non-monetary demand and loose observance of the rules of the game of the gold standard by the central banks.Moreover, rough data on rates of change in the demand for non-monetary gold, changes in monetary gold stock and on de ation in four major economies do not lend support to the adjustment implied by that hypothesis.
e alternative hypothesis, that the relative prices of gold tended to gravitate to one another, is put forward.It builds on the conventional gold standard model and Cassel's insights on purchasing power of currency.Both hypotheses are tested empirically.It is found that long-run adjustment implied by Mazumder and Wood's hypothesis was rather weak, whereas the medium-run adjustment implied by the alternative hypothesis was indeed strong and robust.In other words, changes in the relative prices of gold were driven by the di erences between countries and not within a given country (the absolute deviation from the pre-war level).
ough the results invalidate the hypothesis on the inevitability of the Great De ation the interesting question about its causes remains.On the one hand, price movements were not driven by the absolute deviation from the pre-war level; on the other hand, the tendency of the relative prices of gold to attract one another says nothing about the changes of the common medium-run level of the relative price of gold.us, one needs to look for the explanation that clari es what triggered the movement of the common mediumrun level of the relative price of gold, not only to its pre-war level but much above it.is is left for further research.
the (common) medium-run level of the relative price of gold.Assuming for simplicity that the world stock of monetary gold, approximately constant over time, one can use equation (5) to arrive at the medium-run level of the relative price of gold lines are minus/plus 2 standard errors; delta method was used to derive standard errors.

Figure 1 .
Figure 1.Implied equilibrium levels of relative price of gold and relative purchasing power.Source: Author's calculations.

Table 2
Panel unit root tests for the relative price of gold and real exchange rate Notes: tests are for the log of variable.Schwartz Information Criterion is used for lag length selection.*** , ** , * indicate statistical signicance at 1, 5 and 10 per cent, respectively.Calculations performed with the Eviews 8.

Table 3
Competing hypotheses: relative price of gold adjustment to pre-war level (H MW ) vs. adjustment to common level (H C ), 1927-1935.Notes: t-statistics are in brackets.e cross-section SUR (PCSE) method has been used to obtain robust standard errors (this estimator is robust to cross-equation (contemporaneous) correlation and heteroskedasticity).One lag of dependent variable is included to eliminate serial correlation.Null in the F test and likelihood ratio (LR) test is that the xed e ects are redundant.*** , ** , * indicate statistical signi cance at 1, 5 and 10 per cent, respectively.Calculations performed with the Eviews 8. Source: Author's regressions based on data collected from various issues of the Statistical Year-book of the League of Nations .