How Much Did Uncertainty Shocks Matter in the Great Depression?

Transcription

1 How Much Did Uncertainty Shocks Matter in the Great Depression? Gabriel P. Mathy American University Abstract The United States of the 93s experienced unprecedented uncertainty. Uncertainty shocks buffeted the economy during recessionary periods but these shocks receded during the recovery periods of the Great Depression. Using vector autoregressions on monthly data for 99-94, I show that a one standard deviation increase in uncertainty decreased investment, GDP, industrial output, employment, hours worked, wages, and the price level. I perform a historical decomposition which demonstrates how much uncertainty shocks mattered for explaining movements in major variables during the Depression. Roughly 5-9% of the decline in output can be explained by uncertainty shocks in the Great Depression. JEL classification: D8, E32, N2 Keywords: Uncertainty Shocks, Great Depression, Historical Decomposition Address: 44 Massachusetts Avenue, Washington, DC mathy@american.edu. Phone: I have benefited from fruitful conversations with Gustavo Cortes, Herman Stekler, Chris Meissner, Ellis Tallman, Nicolas L. Ziebarth, Nicholas Bloom, and many others, and have had many useful discussions at seminars and conferences. All errors remain my own.

2 Introduction The Great Depression was an exceptional macroeconomic event, a crisis unparalleled in depth and length. As a historical episode, the American Great Depression saw a stock market crash, record-breaking variability in stock prices, the breakdown of the gold standard, and haphazard and inconsistent monetary policies (Mathy, 26). While the Depression has warranted and received extensive study, several puzzles remain. This paper will present the case for uncertainty shocks as a significant factor in the American Great Depression. The theory of investment under uncertainty was pioneered by Dixit and Pindyck and coauthors. They argue that firms prefer to postpone or reduce investment ( wait-and-see ) for largely irreversible projects if the dispersion of their expected return from a project increases. This effect is driven by the bad-news principle, as proposed by Bernanke (983a), where downside risk is key, as firms are adverse to the potential for low profits on their irreversible investment. Thus firms have a real-option as they can delay their decision until the uncertainty is resolved before investing (McDonald and Siegel, 986). This theoretical work was confirmed by numerous empirical studies which found this effect operative in practice such as Guiso and Parigi (999); Leahy and Whited (996); Hu (995), Kellogg (24), Goel and Ram (999), Ferderer (993), and Scharler and Raunig (2). This largely microeconomic literature was first connected to macroeconomics in Bernanke (983a), who showed that with enough firms facing aggregate uncertainty, the cumulative effect of individual firms reducing their investment would cause a significant reduction in aggregate investment, and the resulting decline in expenditure could generate a recession. This argument was then applied to the Depression literature with Romer (99). Romer s analysis focuses on uncertainty over income or employment by households using stock volatility as a measure of uncertainty which households use to measure broader economic conditions. Given the extreme uncertainty in 929, households cut back sharply on their largely irreversible consumer durables purchases in 93, exacerbating other headwinds in Other papers also followed See Pindyck (99, 993); Abel et al. (996); Caballero and Pindyck (996); Pindyck (988); Majd and Pindyck (987); Dixit and Goldman (97); Dixit (992, 993); Dixit and Pindyck (994). 2 Precautionary savings theories, based on a literature begun by Leland (968), would predict that households increase savings and corresponding reductions in consumption to guard against the possibility of a bad future outcome. The focus on 93 by Romer is meant to address the finding in Temin s book that 93 was the most puzzling year of the Depression as it saw an abnormally large drop in consumer durable purchases. Romer uses the case of 93 to show that uncertainty can explain the remaining decline in consumer durable purchases which remained unexplained by Temin s analysis. 2

3 Romer s in analyzing the uncertainty hypothesis and support a significant role for uncertainty as a significant factor in the Great Depression, both in the United States and the rest of the world, such as Flacco and Parker (992), Ferderer and Zalewski (994), Ferderer and Zalewski (999), Greasley et al. (2), Greasley and Madsen (26), Nabar and Nicholas (2), and Lennard et al. (28). This paper extends Romer s analysis past 93, incorporating newer measures of uncertainty for the Depression and using a historical variance decomposition to quantify the impact of uncertainty shocks on output during the downturn phase of the Depression. Measurement is a major issue in analyzing uncertainty and uncertainty shocks. Uncertainty cannot be measured directly and thus reasonable proxies or indicators must be used to quantify unobservable uncertainty. The seminal uncertainty measure is the volatility of stock returns, which has been used by Leahy and Whited (996), Bloom et al. (27) and Bloom (29) among many others. As equity prices are determined by an interplay between expectations of profitability and the rate at which investors discount those profits, uncertainty over any fundamental such as income, sales, or employment would generate volatility in expectations of profitability and discount rates which would make equity prices more volatile. Veronesi (999) formalizes this intuition in a model where the uncertainty over being in either a high or a low state generates stock volatility. I also construct an indicator variable for high stock volatility in order to focus on large spikes in volatility. To ensure the robustness of the stock volatility measure as a barometer of underlying uncertainty, I also construct a similar measure as Alexopoulos and Cohen (25) who propose a Main Street measure of the number of articles that mention terms related to economic uncertainty in the New York Times. Finally, I use credit spreads between the bonds of low-risk companies those of riskier firms in the form of the interest rate differential between AAA and BAA rated bonds as rated by Moody s Investors Service. As the difference between these two interest rates is determined by the relative default risk between the two types of bonds and the risk appetite of investors, during uncertain times the likelihood of bankruptcy and default by riskier firms is higher and the risk appetite of investors is lower, which will widen these spreads considerably. Despite these caveats, all three uncertainty measures generally move together and tend to be high when output is low and vice versa. All three measures are high during the periods of the 93s when output is declining and declining when output is growing during the Depression, consistent with a negative relationship between uncertainty and output. I also connect the older uncertainty 3

4 literature with the more recent literature on the importance of time-varying uncertainty, which expanded rapidly in the wake of the global financial crisis. 3 Several authors have also discussed the interactions between uncertainty and financial frictions, where uncertainty affects the ability of firms to raise funds which then affects their ability to invest and hire (Stock and Watson, 22; Christiano et al., 23). Another contribution of this paper is to link uncertainty shocks to financial frictions in the Great Depression. This paper extends previous empirical analysis of the uncertainty hypothesis of the Great Depression, such as Romer (99), in several ways. I consider the entire period, with a focus on the Great Collapse and the recovery. I use vector autoregression (VAR) techniques, which allow for regressions between several variables to be conducted over time, yielding dynamic results that build on previous cross-sectional regressions. These regressions provide evidence for a significant negative effect of all four uncertainty shocks on investment, GDP, wages, and prices. This demonstrates statistical significance, but more can be shown, given the large uncertainty shocks which buffeted the American economy of the 93s. Based on the regressions, I construct historical variance decompositions that simulate the path of output, employment, and hours worked predicted by the observed path of uncertainty shocks in the data. These simulations find that uncertainty can explain roughly 4-7% of the observed declines in output, prices, and wages. 4 I also consider an alternative hypothesis, which is that the Federal Reserve bears much of the blame for the Great Depression. Variation in Federal Reserve policy variables predicts much less of the decline in output from as uncertainty shocks predict, consistent with Fed passivity in the face of the crisis, while uncertainty shocks can explain roughly 4-7% of the peak decline in industrial output. Uncertainty shocks mattered a great deal during the Great Depression. 2 Depression Literature Review There are several competing theories of the American Depression. Some center on changes in productivity or market power shocks, perhaps stemming from misallocation, such as (Ohanian, 3 See Bloom et al. (27), Bloom (27), Bloom (29), Bloom et al. (22), Bloom (24), Fernández-Villaverde et al. (25), Basu and Bundick (27), Leduc and Liu (26), Nakamura et al. (27), and Gilchrist et al. (24), though Knotek and Khan (26), and Bachmann and Bayer (23) find that uncertainty, working through the waitand-see effect, is not important for the business cycle. 4 Historical decompositions of this type generally assign a large fraction of the decline to lagged values of the simulated variable (e.g. lagged industrial production) which makes these results even more striking. 4

5 2; Cole and Ohanian, 24; Ziebarth, 22). These theories would predict negative comovement between nominal variables like prices and wages and real quantities like output, while these variables are strongly correlated in the data. Figure show the path of industrial output, personal income, prices and wages from The first three decline from , recover from , decline again from , and then rise again through 94. Wages generally follow the same trends, though they do rise through the recession, likely due to New Deal legislation favorable to labor unions. Other theories center around the pernicious impacts of deflation, such as the interactions between debt and deflation (Fisher, 933), theories of sticky or inflexible wages (Eichengreen and Sachs, 985; Ohanian, 29), banking crises (Friedman and Schwartz, 97; Bernanke, 983b), and many others. Flawed monetary policy, as argued by Friedman and Schwartz (97), Romer (992), Hamilton (987) another primary explanation. Temin (976), however, argues that monetary conditions during the period were not tight, as the monetary base was relatively constant and nominal interest rates fell steadily throughout the period until they could no decline further. 5 On the other hand, Mccallum (988) finds that had the Federal Reserve expanded the monetary base more quickly that the recession of 929 would have been mild, and Christiano et al. (23) find similar results. There is some debate over whether the gold standard constrained monetary policy as a fixed exchange rate regime with open international capital flows, with Hsieh and Romer (26) arguing for this position and Bordo et al. (22) arguing against, though the end of the gold standard was a major impetus for recovery, and coincides with recoveries across the world (Eichengreen, 996). An alternative class of theories would be non-monetary but still demand focused. An obvious example would be Keynes s theory that exogenous changes in animal spirits caused pessimistic expectations and a decline in investment (Keynes, 937) or Temin s theory that an autonomous decline in demand unrelated to changes in wealth or income is to blame. But these are proximate but not ultimate causes: What then caused this autonomous change in demand? These more fundamental explanations have remained elusive, but this paper will attempt an explanation: uncertainty shocks reduced expenditures which then directly led to declines in output and prices. 5 The monetarist position that monetary policy was tight would imply that interest rates were rising, but in fact interest rates were falling as money demand fell pari passu with declining output. 5

6 3 Uncertainty Measures Uncertainty, which is based on expectations of future outcomes, cannot be measured directly. I will construct four uncertainty measures for the interwar period, and then use them to analyze the economic impact of uncertainty shocks during the Great Depression. These measures are: stock volatility, a high volatility indicator, a newspaper index of uncertainty, and a measure of credit spreads. I show that all of these measures are high during recession periods and low during recoveries while the other recessions of the period (including the severe recession) were caused by non-uncertainty shocks (Wicker, 966) Stock Volatility Volatility has long been used to measure uncertainty. Lopez and Mitchener (28) used an exchange rate volatility measure to measure policy uncertainty as a driver of hyperinflations in post-world War I Europe. While stock volatility is the most common uncertainty proxy, stock volatility remains an imperfect measure of uncertainty. There are some episodes of high volatility in stock returns corresponded to relative certainty in economic conditions, such as the economic stability in the wake of the crash of October 987 or the large volatility of the late 99s. Several previous studies have linked uncertainty, stock volatility, and the Great Depression. Nabar and Nicholas (2) find that uncertainty, as measured by stock volatility, depressed the R&D investment of firms in the 93s. Officer (973) finds that stock volatility was similarly low before and after the Great Depression, while the 93s themselves were extremely volatile. Schwert (989) confirms this result, singling out the period between 929 and 939 as extraordinarily volatile. Schwert (99) even finds a strong connection between stock volatility and severe recessions: It is clear... that severe recessions are associated with higher stock volatility.... Of course, the Great Depression, and , is the most severe contraction and it is also the period when volatility was the highest during the sample period. Merton (985) also argues that the 93s stock volatility was partially driven by uncertainty over potential nationalizations and bankruptcies. Cutler et al. (989) shows that, not only was the volatility of aggregate stock indexes high in the 93s, but the cross-sectional volatility of stock prices was high was well. Bloom (29, p.657) 6 The newspaper index does rise during the recession, but the other measures remain relatively low. 6

7 suggests the Great Depression was an era characterized by persistent volatility and uncertainty which would be amenable to analysis by an uncertainty shock model such as in this paper. The only historical example of a persistent second-moment shock was the Great Depression, when uncertainty-as measured by share-returns volatility-rose to an incredible 3% of 9/ levels on average for the 4 years of 929 to 932. Schwert also characterized of stock market volatility as directly reflecting economic uncertainty: [T]he volatility of stock returns reflects uncertainty about future cash flows and discount rates, or uncertainty about the process generating future cash flows and discount rates (Schwert, 99, 85). Cortes and Weidenmier (27), in a related paper, argue that the volatility of building permits explain stock volatility in this period, demonstrating the connection between uncertainty over economic fundamentals and stock price volatility. The link between stock volatility and business cycles has been noted in previous literature, even by authors that weren t considering uncertainty directly. Traditionally, financial economics has modelled equity returns as following a geometric Brownian motion, such as in Black and Scholes (973). This implies that stock returns have a well-defined mean and variance. A more recent literature has examined stochastic volatility models where volatility itself varies, usually following an autoregressive process with errors. I follow this latter method by assuming that volatility is generally fairly stable, but occasionally is hit by a shock that increases the dispersion of returns. 7 The stock volatility measure that I use for this paper is the monthly standard deviation of equity returns (log differences) Volatility Indicator For comparison with the high volatility indicator in Bloom (29), I compute a volatility indicator that is coded as when monthly Dow stock volatility in the 95th percentile or higher over the sample period of 5/896/23 and otherwise. For the sample period from 8963, the 95th percentile for stock volatility is.2 and mean volatility is.8. Bloom (29) used the threshold of.65 standard deviations above mean volatility, but I use the 95th percentile as the data does not 7 Naturally these errors must be highly skewed to match these large shocks in the data. 8 The Dow Jones Industrial Average is used here as it is available prior to 926, unlike the S&P 5, and the resulting stock return volatility for the Dow is very close to that of the S&P 5 in any case. For the modern period, a volatility index such as the VIX or VOX is often used, as they are a forward looking measure of return volatility based on the implied volatility derived from option prices. As stock options are not available for the interwar period, I use observed stock volatility. Observed volatility is not very different than the implied volatility, so this assumption should not affect the results much. 7

8 exactly follow a Gaussian distribution. The volatility indicator is based on the significance level for a one-sided t-test with a 95% significance, so that the ones would correspond to rejections at this significance level, while the zeros would not reject. These high volatility months are listed in Figure 2. It is clear that most of the uncertainty shock events occur during the and recessions, which appear as the blue shaded regions in the figure. While stock volatility does fluctuate even in relatively tranquil times, this indicator variable shows how this period was fundamentally different than most recessions. Given the unprecedented nature of the crisis, being in uncharted waters could only have increased the sense of uncertainty in the economy. 3.3 Newspaper Indexes The newspaper uncertainty index I construct follows the Main Street index of Alexopoulos and Cohen (25) which is also one of the components of the policy uncertainty index of Baker et al. (25). The newspaper index not only produces an alternative uncertainty measure to the stock volatility measure, but also provides a measure that is not based on Wall Street, but is based on journalists discussing conditions in the broader economy. This index measures the number of times uncertain or uncertainty and economic or economy appeared in articles in the New York Times from While the other uncertainty proxies are based on financial markets, the newspaper index based on journalists covering the broader economy, and is not directly linked to financial markets as is the case with stock volatility. To construct the index, I used searchable ProQuest databases of major newspaper archives. The New York Times index is the baseline newspaper index as it is the paper of record for the United States, with other major American newspapers displaying similar trends, as can be seen in Figure 4. To account for the changing number of articles, the total number of articles is divided by the total number of articles per month in that publication. 3.4 Credit Spreads Credit spreads are a measure of financial frictions or financial strain. When the interest rates that firms face is higher, firms ability to finance new investment is reduced. The measure of credit spreads used here is the difference in interest rates between BAA and AAA rated corporate bonds 8

9 as rated by Moody s. The BAA-AAA spread rose during the 29 crisis after the Lehman Brothers uncertainty shock, and the BAA-AAA spread rose during the and during the recessions as well. The BAA rating is the lowest rating of investment-grade debt. The highest rating of investment grade debt for the most reliable debtors is a AAA rating which is resrved for organizations with a low default risk and is generally reserved for the stable and profitable corporation and fiscally sound governments. These credit spreads should rise if the profitability of corporations becomes more uncertain, as this will increase the default risk of more marginal corporations substantially. These links between uncertainty and financial frictions are analyzed in Gilchrist et al. (24), Arellano et al. (2), and Nodari (24). 3.5 Discussion The stock volatility index, the high volatility indicator, the newspaper index, and the credit spread measure are plotted together in Figure 2 of the appendix. The link between the four series in very close, as all rise during the shaded recession periods and fall during the recovery periods of the 93s. In fact, the temporal ordering seems to be very close, so one might worry about reverse causality, where a recession itself generates uncertainty about the future, which is also discussed in Ferderer and Zalewski (999). This reverse causation argument is acceptable for the uncertainty hypothesis for the Depression however, as uncertainty could be a propagation mechanism for the initial shocks of the Great Depression to get worse, or as Bachmann et al. (23) put it, Recessions breed uncertainty. Indeed, these uncertainty measures tend to rise when after the start of the 929 business cycle peak, so this channel is at least partially operative. Ferderer and Zalewski (994) argue that uncertainty surrounding the gold standard served to propagate the Great Depression, and the high persistence of the uncertainty shocks in this period could result from a similar channel. Indeed it does appear that uncertainty rises slightly after the start of the 929 and 937 starting dates for the recessions of the 93s. In any case, all the uncertainty proxies are relatively low during the 92s, which is consistent with monetary or other explanations that do not involve uncertainty for the recessions of the 92s. This both confirms their robustness as uncertainty measures and that uncertainty was very high in the and recession phases of the Great Depression. 9

10 4 Empirical Framework Vector autoregressions are used to examine the empirical relationship between uncertainty shocks and the broader macroeconomy in the 93s. 4. Data The variables considered are the four uncertainty proxies, stock returns, investment, GDP, the industrial output, manufacturing employment, hours worked in manufacturing, and the consumer price index (CPI). Stock returns are calculated as the change in the log of the Dow Jones Industrial Average divided by the number of trading days elapsed. All variables are in logs with the exception of the discount rate. There are two regressions run, one at the quarterly and one at the monthly level, based on data availability. Table 2 is a correlation table of most of the variables considered in this paper. Stock volatility, the newspaper index, and credit spreads are all positively correlated, but these uncertainty shocks are negatively correlated with the other variables. Data sources are described in the supplementary appendix in Table. 4.2 Empirical setup I first examine the Granger Causality (Granger, 969) between stock return volatility, the newspaper index, and credit spreads. This test finds that previous values of stock volatility are correlated with future values of both credit spreads and the newspaper index, that credits spreads Granger cause the newspaper index, and that the newspaper index does not Granger cause (i.e. does not have future predictive power) over either of the other two uncertainty proxies. I interpret this to mean that information about uncertainty shocks, say about banking panics during the Depression, is first reflected in return volatility. The uncertainty about firms future profitability is then reflected in worsening credit spreads. The media report on the uncertainty arising in financial markets last. Thus we can see how uncertainty shock act as a financial friction through credit spreads, and why all three measures proxy for underlying unobservable uncertainty shocks. This VAR covers the full period. As uncertainty was highest from , a period which begins after the peak and and which ends after the trough of output, it is likely that causality runs between uncertainty and output and vice versa.

11 Next I perform two vector autoregression, ordered for Cholesky decomposition. The first has the ordering of the uncertainty shock (each of the four), then investment, GDP, the consumer price index, and wages. 9 This is meant to capture the effect of uncertainty working through either a wait-and-see effect or through financial frictions, which should affect investment first, and then the broader economy as expenditures fall. As uncertainty shocks are demand shocks (Leduc and Liu, 26), we should then see the price level fall as well. The price level is ordered after GDP as ordering the price level before GDP would potentially include the effect of deflation in propagating the Depression, which was important but is not the channel under consideration. As GDP and prices fall, this would put pressure on wages to fall. This VAR uses data from due to data limitations, and is at a quarterly frequency. After performing these test for each of the uncertainty variables, I then rerun the VARs but now adding the level of stock returns before all other variables. This helps to control for the first-moment of the stock market to differentiate these effects from those of the second-moment of uncertainty, as in Romer (99). These robustness checks can be found in the supplementary appendix. The second specification is ordered as the stock return, uncertainty shock, then a monthly proxy for GDP (industrial output, manufacturing employment, and manufacturing hours worked) and finally the consumer price index. This specification has about three times are much data as the quarterly specification, so provides tighter estimates of the underlying relationships. For each of the VARs, I plot an impulse response function which displays the effect of a onestandard deviation shock to the impulse variable on other variables of interest, assuming all other variables are set to zero in the initial period. While these impulse response results are informative, performing a historical decomposition can yield further information about the importance of uncertainty shocks in the Great Depression. Historical decomposition has been used before to see how important various shocks were in determining the length and depth of the Great Depression in the United States. Sims (98), Burbidge and Harrison (985), and Fackler and Parker (994) look at the role money played in the Great Depression, Cecchetti and Karras (994) decompose 9 The preponderance of the lag length tests found that three lags was optimal, so I use 3 quarterly lags. To ensure stationarity and to remove seasonal factors, I regress all variables on a time trend and four quarterly dummies, then detrend all variables using their corresponding trend series. The preponderance of the lag length tests found that three lags was optimal, so I use 3 quarterly lags. To ensure stationarity and to remove seasonal factors, I regress all variables on a time trend and four quarterly dummies, then detrend all variables using their corresponding trend series.

12 Depression declines into aggregate demand and aggregate supply shocks, and McMillin and Parker (994) looks at the impact of oil prices during the Depression. Historical decomposition is based on a vector autoregression. The structural shocks obtained from the VAR are multiplied by the size of the shock in the data to construct a counterfactual series. This counterfactural series is then compared to the actual path of output for the Depression. If uncertainty shocks are a significant determinant of the path of output in this period, uncertainty shocks will explain a large fraction of this variance, but if uncertainty shocks do not explain a large portion of the behavior of output in this period, then we can say that uncertainty shocks did not matter much. 5 Results Figure 5 displays the results for the effects of the Granger Causality tests for the three main uncertainty shocks. An impulse to stock volatility generates a roughly 7% increase in the newspaper index which is statistically significant for 7 quarters. An impulse to stock volatility will induce a.-.2 percentage point increase in credit spreads that is statistically significant for two years. An impulse to credit spreads is correlated with a peak increase of 8% in the newspaper index after three quarters, with this results statistically significant over six quarters. This shows how uncertainty shocks generate first stock volatility, then an increase in credit spreads, then a rise in the newspaper index as discussed above. 5. VAR I Figure 5 displays the effect of an impulse to stock volatility on investment, with a peak decline of roughly 2% after quarters, For the high volatility indicator the effect is over %. The impulse response results for the newspaper index are for about a % decline in investment. Investment declines by more than 5% when there is an impulse to credit spreads. Statistical significance continues for about 2 years after the simulated shock. These results confirm the theoretical link between uncertainty shocks and investment declines as posited by the wait-and-see theories of uncertainty. Figure 6 plots the effect of an impulse to stock volatility on GDP of a decline of 4% through three 2

13 years. The effect of an impulse to the high volatility indicator is always negative with a magnitude that peaks at a less than 4% decline in GDP around 8 quarters. The impulse response results for the newspaper index are a 4% decline in GDP. On impact of an impulse to credit spreads, GDP declines by about 6% with statistical significance through 9 quarters. Fernández-Villaverde et al. (2) finds that a 2-standard deviation increase in fiscal volatility decreases output by about.2% which is much lower than the results here, for a different type of shock. Their results are also lower than other empirical estimates of uncertainty, such as Bloom (29), who found that increases in stock volatility of.654 standard deviations decreases output by about %. However, my estimates are larger still, which may be due to fragility or sensitivity of the economy to uncertainty shocks as output declines in this period are much larger in magnitude than in the postwar. The effect of an impulse to stock volatility on the consumer price index peaks at a 2% decline after quarters with statistical significance from 4 to quarters as can be seen in Figure 7. For the high volatility indicator, the peak decline is 2% with statistical significance for 3 to quarters. The peak effect of the newspaper index is similar, with an almost 2% decline after two years with statistical significance through quarters. Credit spreads see a peak decline of over 2% after 9 quarters with statistical significance in all quarters except the first. The reader will note that the magnitudes of the price declines across the different uncertainty shocks is very consistent. Deflation corresponded with declines in economic activity in the Depression period, and uncertainty shocks correspond with deflation as expected. The effect of an impulse to stock volatility on the level of wages peaks at a 2% decline after about 8 quarters with statistical significance from 3 to quarters as can be seen in Figure??. For the high volatility indicator, the peak decline is 2% with statistical significance for 3 to quarters. The peak effect of the newspaper index is similar, with an almost 2% decline a bit less than two years with statistical significance through quarters. Credit spreads see a peak decline of over 2% after 9 quarters with statistical significance in all quarters except the first. The reader will note that the magnitudes of the price declines across the different uncertainty shocks is very consistent. Deflation corresponded with declines in economic activity in the Depression period, and uncertainty shocks correspond with deflation as predicted. Romer (99) includes the level of stock returns in her estimation results to cleanly separate first-moment effects from the second-moment effects of stock volatility. I use stock returns based 3

14 on the log difference of stock prices and the volatility of those returns which helps address this concern. Take the example of a constant decline in the Dow, which will increase volatility in Romer s specification, but will appear as constant volatility in my measure of return volatility. Even so, separating level effects from variance effects is important. I add the level of average daily stock returns over the quarter to the beginning of the var, before the uncertainty shock. These results are added to the supplementary appendix. Figures 7-9 display the result for the same VARs as above but where the level of stock returns has been added. As can be clearly seen, including this variables does not affect the results much, showing that the second moment effect is being identified independently of first moment effects. The magnitudes are similar of slightly smaller, and the standard errors rise as is to be expected when a variable is added to a sample which has not increased in size. 5.2 VAR II The second specification, the reader will recall, is ordered as the level of the monthly average stock return, the uncertainty shock, then a monthly proxy for GDP (industrial output, manufacturing employment, and manufacturing hours worked) and finally the consumer price index. Figure 9 shows the results for the impulse response of the four uncertainty shocks (stock volatility, high volatility indicator, newspaper index, and credit spreads) on industrial output. A one-standard deviation shock to stock volatility or the high volatility indicator yields a peak decline of over 3% in industrial output, credit spreads and the newspaper index a peak decline of 4%, with all statistically significantly different than zero for most horizons. Figure shows the results for the impulse response of the four uncertainty shocks on manufacturing employment. Here, the different measures all roughly conform, with peaks declines of 2% in employment occurring approximately 8 months after the shock, with statistical significance strongest for stock volatility and the high volatility indicator. Figure shows the results for the impulse response of the four uncertainty shocks on hours worked in manufacturing. Both stock volatility and the high volatility indicator have impulse that peak at over a % decline in hours worked on impact, with credit spreads seeing between % peak declines, with statistical significance common in the horizon of the first year. 4

15 5.3 Historical Variance Decomposition First I plot the historical decomposition results for a VAR with uncertainty shocks, industrial production, wages, and the CPI from , between the business cycle peaks in this period. All are calculated as the percent change in the series between the initial quarter (the business cycle peak of 929Q3) and the quarter plotted. Figure 2 display the results for industrial production using all four uncertainty shocks. All four series track the declines in output through 933 and the recovery through 937. The share of the actual decline in industrial production from business cycle peak in 929Q3 to the trough in 933Q is 5% for stock volatility, 64% for the stock volatility indicator, 9% for the newspaper index, and 62% for credit spreads. Figure 3 shows the historical prediction for the consumer price index, with again all four series predicting significant declines in the CPI. The share of the actual decline in the CPI from business cycle peak to trough is 56% for stock volatility, 6% for the stock volatility indicator, 94% for the newspaper index, and 86% for credit spreads. Credit spreads and the newspaper index track the decline almost exactly in many periods. Figure 4 shows the historical decomposition for wages with all four series tracking the actual decline in wages well, with the newspaper index and the credit spreads simulations tracking the actual changes almost exactly in many periods. The share of the actual decline in wages from business cycle peak to trough is 58% for stock volatility, 62% for the stock volatility indicator, 95% for the newspaper index, and 83% for credit spreads. Overall, the four uncertainty measures fit the data well and predict a large decline in output, wages, and the price level, befitting their role as a quantitatively significance aggregate demand shock in the Great Depression. 6 Conclusion This study has examined the effect of uncertainty shocks on the American economy of the 93s, which was beset by extensive uncertainty shocks in the form of banking failures and panics, uncertain monetary policies, the breakdown of the gold standard, and many other uncertainties. I constructed several measures of uncertainty to estimate how large uncertainty shocks were in the Great Depression, including stock volatility, a high volatility indicator, a newspaper index, and credit spreads. 5

16 Empirical evidence showed that the uncertainty measures are correlated with statistically significant declines in a range of important macroeconomic variables including industrial output, employment, hours, investment, GDP, wages, and the price level over the next year or two. This result is robust to the uncertainty proxy used, as all four uncertainty proxies yield roughly similar results for the endogenous variables. Uncertainty shocks cause declines in output and prices consistent with aggregate demand declines seen during the Great Depression in the United States. The historical decomposition techniques made it possible to see if uncertainty shocks were large enough to cause the sheer magnitude of declines seen during the 93s. Uncertainty shocks can explain a considerable amount of the approximately 75% decline in industrial output from 929 to 933. About half of the decline in industrial output from peak output in 929 and trough output in 933 is predicted based on the actual path of stock volatility, which is a 37% predicted decline. The high volatility indicator predicts a 48% decline or roughly 64% of the total decline. 9% of this decline, or a 68% decline from 929 to 933, is predicted based on the path of the newspaper uncertainty index. The variation in credit spreads yields the prediction of a 46% peak decline in output or 6% of the actual decline. Naturally other factors, such as the financial and banking collapse and the collapse of the money supply clearly played a role as well. But it is also clear that uncertainty shocks, despite being relatively neglected in the Depression literature up to this point, have an important role to play in providing a more complete understanding of the Great Depression in the United States. While other factors played a role in the demand demand, none were quantitatively significant enough to explain such a large and persistent decline in economic activity. This paper has shown the uncertainty shocks can explain a significant percentage of the behavior in important macroeconomic variables. But moreover, this paper has provided a fundamental explanation for the aggregate demand declines of the Great Depression, deepening our understanding of the causes of the Great Depression in the United States. 6

17 References Abel, Andrew B., Avinash Dixit, Janice C. Eberly, and Robert S. Pindyck, Options, the Value of Capital, and Investment, The Quarterly Journal of Economics, August 996, (3), Alexopoulos, Michelle and Jon Cohen, The power of print: Uncertainty shocks, markets, and the economy, International Review of Economics & Finance, 25. Arellano, Cristina, Yan Bai, and Patrick Kehoe, Financial Markets and Fluctuations in Volatility, Federal Reserve Bank of Minneapolis Working Paper, 2. Bachmann, Rüdiger and Christian Bayer, Wait-and-See business cycles?, Journal of Monetary Economics, 23, 6 (6), , Steffen Elstner, and Eric R. Sims, Uncertainty and Economic Activity: Evidence from Business Survey Data, American Economic Journal: Macroeconomics, 23, 5 (2), Baker, Scott R, Nicholas Bloom, and Steven J Davis, Measuring economic policy uncertainty, Technical Report, National Bureau of Economic Research 25. Basu, Susanto and Brent Bundick, Uncertainty shocks in a model of effective demand, Econometrica, 27, 85 (3), Bernanke, Benjamin S., Irreversibility, Uncertainty, and Cyclical Investment, The Quarterly Journal of Economics, February 983, 98 (), 85 6., Nonmonetary Effects of the Financial Crisis in Propagation of the Great Depression, American Economic Review, June 983, 73 (3), Black, F. and Myron Scholes, The Pricing of Options and Corporate Liabilities, The Journal of Political Economy, 973, pp Bloom, Nicholas, Uncertainty and the Dynamics of R&D, American Economic Review, May 27, 97 (2), , The Impact of Uncertainty Shocks, Econometrica, 5 29, 77 (3), , Fluctuations in uncertainty, Journal of Economic Perspectives, 24, 28 (2), , Max Floetotto, Nir Jaimovich, Itay Saporta-Eksten, and Stephen J. Terry, Really Uncertain Business Cycles, National Bureau of Economic Research Working Paper, 22., Stephen Bond, and John Van Reenen, Uncertainty and Investment Dynamics, Review of Economic Studies, 4 27, 74 (2), Bordo, Michael D, Ehsan U Choudhri, and Anna J Schwartz, Was expansionary monetary policy feasible during the Great Contraction? An examination of the gold standard constraint, Explorations in Economic history, 22, 39 (), 28. Burbidge, John and Alan Harrison, An historical decomposition of the great depression to determine the role of money, Journal of Monetary Economics, 985, 6 (),

18 Caballero, Ricardo J. and Robert S. Pindyck, Uncertainty, Investment, and Industry Evolution, International Economic Review, August 996, 37 (3), 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, 994, 76 (), 8 2. Christiano, Lawrence J., Roberto Motto, and Massimo Rostagno, Risk shocks, National Bureau of Economic Research Working Paper, 23. Christiano, Lawrence, Roberto Motto, and Massimo Rostagno, The Great Depression and the Friedman-Schwartz Hypothesis, Journal of Money, Credit and Banking, 23, pp Cole, Harold L. and Lee E. Ohanian, New Deal Policies and the Persistence of the Great Depression: A General Equilibrium Analysis, Journal of Political Economy, August 24, 2 (4), Cortes, Gustavo S and Marc D Weidenmier, Stock Volatility and the Great Depression, Technical Report, National Bureau of Economic Research 27. Cutler, David M, Lawrence Katz, Danny Quah, Louise Sheiner, and Jeff Wooldridge, Stock Market Volatility, Cross-Section Volatility, and Stock Returns, Massachusetts Institute of Technology Working paper, 989. Dixit, Avinash, Investment and Hysteresis, The Journal of Economic Perspectives, 992, 6 (), pp , Art of Smooth Pasting, Routledge, 993. and Robert S. Pindyck, Investment under Uncertainty, Princeton University Press, 994. and Steven M. Goldman, Uncertainty and the demand for liquid assets, Journal of Economic Theory, December 97, 2 (4), Eichengreen, Barry, Golden Fetters: The Gold Standard and the Great Depression, , Oxford University Press, 996. and Jeffrey Sachs, Exchange Rates and Economic Recovery in the 93s, The Journal of Economic History, December 985, 45 (4), Fackler, James S and Randall E Parker, Accounting for the Great Depression: a historical decomposition, Journal of Macroeconomics, 994, 6 (2), Ferderer, J. Peter, The Impact of Uncertainty on Aggregate Investment Spending: An Empirical Analysis, Journal of Money, Credit and Banking, 993, 25 (), pp and David A. Zalewski, Uncertainty as a Propagating Force in The Great Depression, The Journal of Economic History, 994, 54 (4), and, To Raise the Golden Anchor? Financial Crises and Uncertainty During the Great Depression, The Journal of Economic History, September 999, 59 (3),

19 Fernández-Villaverde, Jesús, Pablo A. Guerrón-Quintana, Keith Kuester, and Juan Rubio-Ramírez, Fiscal volatility shocks and economic activity, National Bureau of Economic Research Working Paper, 2., Pablo Guerrón-Quintana, Keith Kuester, and Juan Rubio-Ramírez, Fiscal volatility shocks and economic activity, The American Economic Review, 25, 5 (), Fisher, Irving, The Debt-Deflation Theory of Great Depressions, Econometrica, 933, (4), Flacco, Paul R. and Randall E. Parker, Income Uncertainty and the Onset of the Great Depression, Economic Inquiry, 992, 3 (), Friedman, Milton and Anna J. Schwartz, A Monetary History of the United States, , Princeton University Press, 97. Gilchrist, Simon, Jae W Sim, and Egon Zakrajšek, Uncertainty, financial frictions, and investment dynamics, National Bureau of Economic Research Working Paper, 24. Goel, Rajeev K. and Rati Ram, Variations in the Effect of Uncertainty on Different Types of Investment: An Empirical Investigation, Australian Economic Papers, December 999, 38 (4), Granger, Clive W. J., Investigating causal relations by econometric models and cross-spectral methods, Econometrica, 969, pp Greasley, David and Jakob B. Madsen, Investment and Uncertainty: Precipitating the Great Depression in the United States, Economica, 26, 73 (29), ,, and Les Oxley, Income Uncertainty and Consumer Spending during the Great Depression, Explorations in Economic History, 2, 38 (2), Guiso, Luigi and Giuseppe Parigi, Investment and demand uncertainty, The Quarterly Journal of Economics, 999, 4 (), Hamilton, James D., Monetary factors in the Great Depression, Journal of Monetary Economics, 987, 9 (2), Hsieh, Chang-Tai and Christina D. Romer, Was the Federal Reserve constrained by the gold standard during the Great Depression? Evidence from the 932 open market purchase program, The Journal of Economic History, 26, 66 (), Hu, Zuliu, Stock market volatility and corporate investment, International Monetary Fund Working Paper, 995. Kellogg, Ryan, The effect of uncertainty on investment: evidence from Texas Oil Drilling, The American Economic Review, 24, 4 (6), Keynes, John M., The General Theory of Employment, The Quarterly Journal of Economics, 937, 5 (2), Knotek, Edward S. and Shujaat Khan, How do households respond to uncertainty shocks?, Federal Reserve Bank of Kansas City Economic Review, 26, (II). 9

20 Leahy, John V. and Toni M. Whited, The Effect of Uncertainty on Investment: Some Stylized Facts, Journal of Money, Credit and Banking, February 996, 28 (), Leduc, Sylvain and Zheng Liu, Uncertainty shocks are aggregate demand shocks, Journal of Monetary Economics, 26, 82, Leland, Hayne E., Saving and uncertainty: The precautionary demand for saving, The Quarterly Journal of Economics, 968, 82 (3), Lennard, Jason et al., Uncertainty and the Great Slump, Technical Report, Lund University, Department of Economic History 28. Lopez, Jose A and Kris James Mitchener, Uncertainty and Hyperinflation: European Inflation Dynamics after World War I, Technical Report, National Bureau of Economic Research 28. Majd, Saman and Robert S. Pindyck, Time to build, option value, and investment decisions, Journal of Financial Economics, March 987, 8 (), Mathy, Gabriel P, Stock volatility, return jumps and uncertainty shocks during the Great Depression, Financial History Review, 26, pp. 28. Mccallum, Bennet T., Robustness properties of a rule for monetary policy, Carnegie-Rochester Conference Series on Public Policy, January 988, 29 (), McDonald, Robert and Daniel Siegel, The Value of Waiting to Invest, The Quarterly Journal of Economics, 986, (4), McMillin, W Douglas and Randall E Parker, An empirical analysis of oil price shocks in the interwar period, Economic Inquiry, 994, 32 (3), Merton, Robert C., On the Current State of the Stock Market Rationality Hypothesis, 985. Nabar, Malhar and Tom Nicholas, Uncertainty and Innovation at the Time of the Great Depression, Harvard Business School Working Paper, 2. Nakamura, Emi, Dmitriy Sergeyev, and Jon Steinsson, Growth-rate and uncertainty shocks in consumption: Cross-country evidence, American Economic Journal: Macroeconomics, 27, 9 (), 39. Nodari, Gabriela, Financial regulation policy uncertainty and credit spreads in the US, Journal of Macroeconomics, 24, 4, Officer, Robert R., The Variability of the Market Factor of the New York Stock Exchange, Journal of Business, July 973, 46 (3), Ohanian, Lee E., Why did Productivity Fall so much during the Great Depression?, The American Economic Review, 2, 9 (2), , What or who started the great depression?, Journal of Economic Theory, 29, 44 (6), Pindyck, Robert S., Irreversible Investment, Capacity Choice, and the Value of the Firm, American Economic Review, December 988, 78 (5),

21 , Irreversibility, Uncertainty, and Investment, Journal of Economic Literature, 99, 29 (3), 48., Investments of uncertain cost, Journal of Financial Economics, August 993, 34 (), Romer, Christina D., The Great Crash and the Onset of the Great Depression, The Quarterly Journal of Economics, 99, 5 (3), , What ended the great depression?, The Journal of Economic History, 992, 52 (4), Scharler, Johann and Burkhard Raunig, Stock Market Volatility, Consumption and Investment; An Evaluation of the Uncertainty Hypothesis Using Post-War US Data, 2. Schwert, G. William, Why Does Stock Market Volatility Change over Time?, Journal of Finance, December 989, 44 (5), 5 53., Business cycles, financial crises, and stock volatility, National Bureau of Economic Research Working Paper, 99. Sims, Christopher A., Comparison of Interwar and Postwar Business Cycles: Reconsidered, The American Economic Review, 98, pp Monetarism Stock, James H. and Mark W. Watson, Disentangling the Channels of the 27?9 Recession, Brookings Papers on Economic Activity: Spring 22, 22, p. 8. Temin, Peter, Did Monetary Forces Cause the Great Depression?, New York: Norton, 976. Veronesi, Pietro, Stock Market Overreactions to Bad News in Good Times: a Rational Expectations Equilibrium Model, Review of Financial Studies, 999, 2 (5), Wicker, Elmus R, A Reconsideration of Federal Reserve Policy during the Depression, The Journal of Economic History, 966, 26 (2), Ziebarth, Nicholas L., Misallocation and productivity during the Great Depression, Northwestern University Working Paper, 22. 2

22 7 Appendix m 93m7 932m 933m7 935m 936m7 938m 939m7 94m 942m7 Industrial Production Personal Income CPI Wages Figure : Graph of Industrial Production, CPI, Personal Income, and Wages during the Great Depression. =August 929, business cycle peak. Source: Federal Reserve Economic Database/BEA/Federal Reserve Board/

23 Jan-9 Jul-9 Jan Jul Jan Jul Jan2 Jul2 Jan3 Jul3 Jan4 Jul4 Jan5 Jul5 Jan6 Jul6 Jan7 Jul7 Jan8 Jul8 Jan9 Jul9 Jan-3 Jul-3 Jan-3 Jul-3 Jan-32 Jul-32 Jan-33 Jul-33 Jan-34 Jul-34 Jan-35 Jul-35 Jan-36 Jul-36 Jan-37 Jul-37 Jan-38 Jul-38 Jan-39 Jul-39 Jan Jul Jan Jul Figure 2: Four Uncertainty Measures: % 7 7% 6 6% 5 5% 4 4% 3 3% 2 2% % % Credit Spreads Newspaper Index High Stock Volatility Indicator Stock Return Volatility Figure 3: Left axis corresponds to credit spreads, newspaper index, and high stock volatility indicator. Right axis is stock volatility. Credit spreads are the difference between BAA and AAA interest rates, the newspaper index is the number of articles mentioning economic uncertainty normalized by the number of articles, and stock return volatility is the annualized quarterly standard deviation of daily returns in percent. 23

24 8 6 New York Times Washington Post Los Angeles Times Chicago Tribune Wall Street Journal /2 2/2 2/2 /22 8/23 6/24 4/25 2/26 2/26 /27 8/28 6/29 4/3 2/3 2/3 /32 8/33 6/34 4/35 2/36 2/36 /37 8/38 6/39 4/4 2/4 2/4 Figure 4: Newspaper indexes formed from number of articles contained the terms economy or economic and uncertain or uncertainty. The number of articles per month is then divided by the number of days in that month, and multiplied by the average number of days in an average month (3.25). A 7-month moving average is applied using the three months before and after, and then all series are normalized such that the trough of the Great Depression is March 933=. Source: ProQuest Historical Newspapers.