Contagion And Bank Failures During The Great Depression: The June 1932 Chicago Banking Panic

Summary for Joint Cliometric Society/AEA Session

Charles W. Calomiris, University of Illinois, Urbana-Champaign and National Bureau of Economic Research, and Joseph R. Mason, University of Illinois, Urbana-Champaign

I. Bank Panics and Bank Failures, Before and During the Depression

Recent theoretical and empirical work (reviewed in Calomiris, 1993) argues that interbank cooperation often prevented the externalities that caused panics from producing unwarranted bank failures of solvent banks. Does the experience of the 1930s - which witnessed many bank failures, sometimes occurring at the same place over a short interval of time - represent an important counterexample?

Despite the large numbers of bank failures during the 1930s, some historians of American banking have argued that the banking collapse during the Great Depression may not fit the definition of a "true" banking panic (or series of panics). Instead the simultaneous collapses of many banks during the Depression may have reflected large, sudden asset value reductions that made many banks insolvent.

In this paper we consider the question of whether banks failed during the Great Depression because they experienced common exogenous declines in asset values, or because of contagions of fear that swept banks away irrespective of their fundamental solvency. We address this question by focusing on one of the clearest and most important instances of a bank panic during the Great Depression: the Chicago panic of June 1932. We employ data from individual bank failure experience, balance sheets, income and expense statements, and stock prices for failing and surviving Chicago banks around the time of the panic. We analyze the characteristics of failing and surviving banks to determine whether the banks that failed during the panic were similar ex ante to those that survived the panic, or alternatively, whether they differed from survivors and shared characteristics with other banks that failed outside the panic window. To the extent that panic bank failures were like non-panic failures and unlike survivors, we argue, their failure cannot be attributed to contagion.

II. The June 1932 Banking Crisis in Chicago

As Figures 1 and 2 show, mid-to-late June of 1932 witnessed an unparalleled concentration of bank failures in Chicago, whether measured by the number or total assets of failed banks. In contrast, the number of bank failures in June 1932 was not particularly high at the state, Federal Reserve District, or national level in comparison to previous months ( Figure 3).

James (1938) argues that the panic was triggered by several factors, including declines in real estate values, falling local utility stocks, and a well-publicized local case of bank fraud and mismanagement. In addition to these problems, the Chicago municipal government had been undergoing significant strain since 1931. The government failed to make payments on its municipal bonds in January 1932, and beginning in 1931 intermittently withheld pay from government workers or issued scrip. In March 1932, payments to city workers were suspended indefinitely. The city government's revenue problem weakened the banks by limiting bank revenue from municipal bond coupons, and encouraging withdrawals by illiquid depositors.

By June 23, bank depositors had witnessed, in a matter of two weeks, the collapse of some of the largest businesses in their city, an enormously costly case of bank fraud, a new arrest on fraud charges, and the denial of relief to their city government by federal authorities. It is not surprising that depositors became increasingly concerned over those weeks about the ability of banks to pay out their deposits.

The dramatic withdrawals from Loop banks began on June 22 and reached their peak on Friday, June 24. James argues that interbank cooperation, and the intervention of informed third-parties, brought an end to the crisis, but his account of the crisis leaves unresolved whether the banks that failed during the panic were those most likely to be insolvent, or whether failing banks simply lacked the protection of the clearing house or correspondent banks for other reasons. In the following sections we address that question.

III. Failures and Survivors During the Panic

In our empirical work, we examine the ex ante observable attributes of non-panic failing banks, banks that failed during panics, and banks that survived the panic. We ask whether failures of banks during the panic reflected the continuation of the same process that underlay other failures, or whether panic failures were observably similar to panic survivors. We focus on four kinds of ex ante measures of bank condition - (1) the ratio of market value of equity to the book value of equity or assets, (2) the estimated probability of failure, or expected survival duration, of banks, (3) the debt composition of banks, (4) the rate of decline in bank assets and deposits, and (5) the interest promised on bank debts.

Market-to-Book Value Ratios

Figure 5 plots the mean market-to-book value ratios of three separate groups of Chicago banks, based on stock price data reported in the Bank and Quotation Record of the Commercial and Financial Chronicle. We divide banks into the following groups: non-panic failures (banks that failed on non-panic dates between January and July 1932), panic failures (banks that failed during the June panic), and survivors (banks that survived the June panic). Standard deviations for each group are represented by the bracketed areas about the mean of each plot. The striking fact illustrated by Figure 5 is that as early as January 1931 the banks that survived the June panic appeared to be a separate group with higher average market-to-book ratios. The banks that failed during the panic had slightly higher average ratios than those that failed at other times, but throughout the pre-panic period (January 1931-May 1932) the market-to-book value ratios of panic failures were closer on average to those of pre-panic failures than to panic survivors. By May 1932, most of the panic failures had ratios less than unity. Figure 5 shows that all Chicago banks suffered from capital decline during 1931 and 1932, and that the banks that failed during the June panic reached and maintained unusually low market-to-book value ratios long before the panic. Figure 6 plots the percentage of surviving banks, and all banks, with ratios of the market value of equity to the book value of assets less than 10% for the period January 1931 through July 1932. Clearly, among surviving banks, more banks had large capital buffers.

Failure Predictions

We estimate the probability of failure using a logit model of the links between bank characteristics (e.g., balance sheet ratios) and bank failure. We also estimate a survival duration model, which is similar to the logit model except that it forecasts the length of time the bank will survive rather than the probability it will fail. The danger of using ex post failures to estimate failure risk, of course, is that special events with low probabilities may have influenced actual failure experience in ways that were unpredictable ex ante. To avoid (or at least minimize) this problem, we report logit failure forecasts constructed from both "in-sample" and "out-of-sample" estimation. In the out-of-sample forecasts, we exclude banks that failed during the panic from the sample when we estimate the coefficients relating bank characteristics to the probability of failure. This constrains the panic failures to be "predicted" using model parameters that were constructed to explain non-panic failures, and thus prevents special unpredictable events during the panic from influencing predictions of failure.

Our in-sample and out-of-sample logit results are reported in Tables 1a, 1b, and 1c. In Table 1a, we include the following variables (all measured at year-end 1931): size (log of total assets), the reserve-to-demand deposit ratio, the real estate loan share (defined as the ratio of loans on real estate to total illiquid assets), the ratio of real estate owned to illiquid assets (which mainly includes repossessed real estate collateral, and excludes bank premises), the ratio of last year's retained earnings to net worth, and the long-term debt ratio (bills payable plus rediscounts plus time deposits, divided by total assets).

We also experimented with including two other variables, which are omitted from Table 1a: the ratio of book net worth to assets, and the percentage changes in deposits and assets of banks from December 1930 to December 1931. In neither case did the regressors prove significant or affect our other results. The rates of decline of assets and deposits are highly correlated with bank failure, but they are also highly correlated with other regressors (see Table a1), and added no explanatory power to our regressions. In the case of the book net worth ratio, the sign was positive (contrary to our expectation). This is consistent with results from earlier work on similar data by White (1984, 126). In Tables 1b and 1c, we restrict our sample to banks for which we have stock price information, and add to our list of regressors the ratio of the market value of net worth to the market value of assets (assuming par valuation for debt). We also redefine the earnings to net worth ratio using the market rather than the book value of net worth. The market equity to asset ratio has the predicted negative sign and is statistically significant.

The logit results in Tables 1a, 1b, and 1c are quite similar for the in-sample and out-of-sample specifications, which is consistent with the view that failures during panics were similar events to non-panic failures. The variables that are most significant in the logits are of the predicted sign (the reserve ratio, the ratio of retained earnings to net worth, the net worth ratio, and the term structure of debt).

Figures 7a and 7b plot the failure probabilities of the in-sample and out-of-sample logits from Table 1a against one another, and indicate each type of bank (non-panic failure, panic failure, and survivor). Not surprisingly, all observations (and especially the panic failures) tend to lie above the 45 degree line - that is, by construction, including panic failures when estimating model coefficients increases the estimated failure probabilities for banks. Interestingly, however the ordinal ranking of banks' failure probabilities (within and across the three classes) is quite robust to whether the in-sample or out-of-sample model is used. Similar plots using estimated probabilities from Tables 1b and 1c (not reported here) provide the same picture.

We report survival regressions in Tables 2a, 2b, and 2c - which estimate the number of days the bank will survive beyond December 31, 1931. The results are quite similar to the logits (with coefficients of opposite sign, since the dependent variable is the survival time of the bank rather than the probability of failure). Figures 8a and 8b plot the in-sample predicted durations of survival against the out-of-sample predicted durations from Table 2a. As before, the rankings are similar, although now (by construction) observations lie below the 45 degree line.

Table 3 reports the mean and median predicted failure probabilities and durations for the logit and survival models by category of bank (panic failure, non-panic failure, and survivor), and the significance levels of tests for differences across categories in means and medians. These results indicate that the banks that failed during the panic were less risky than banks that failed outside the panic and more risky than survivors. Results using predicted values from in-sample and out-of-sample regressions are similar, although (by construction) the in-sample results show less of a difference between panic and non-panic failures compared to either as against survivors. Overall, our results are consistent with the notion that failures during the panic were a continuation of the same process that underlay other failures. The relatively late timing of panic failures can be explained by their low risk relative to the banks that failed earlier.

Debt Composition and Deposit Withdrawals

Detailed data on the composition of bank liabilities are available for all banks in our sample, either from Federal Reserve or state call reports. Tables 4a and 4b present data on the liability composition of banks as of December 1931, divided into groups in two ways - by the probability of failure (divided into high, medium, and low risk using the out-of-sample logit model), and according to actual failure experience (survivors, panic failures, and non-panic failures). Two interesting patterns emerge.

First, the shares of the various debt categories are monotonic in the risk of failure. The shares of demand deposits of the public and deposits of banks are decreasing in the probability of failure, while the shares of time deposits and "borrowed money" (defined as bills payable and rediscounts) are increasing in failure risk. One interpretation of this finding - which is consistent with observed differences in the deposit withdrawal rates across categories during 1931 reported in these tables - is that demandable debt is withdrawn first when banks become risky (long-term debts do not give depositors the option of costless early withdrawal). Additionally, banks that suffer large withdrawals of demandable debt are forced to raise additional funds through bills payable and rediscounts (essentially the CD market of that era). Banks with low probability of failure virtually never use this high-cost means of raising funds. Other studies have found that the share of "borrowed money" is a reliable predictor of bank failure during the 1920s and 1930s (Wheelock, 1992, Mason, 1994).

Second, the liability shares of panic failures are between those of survivors and those of non-panic failures, and indicate a liability profile of a medium-to-high insolvency risk bank. Although significance levels of tests for differences in means are sometimes low, given the small sample size, the patterns are consistent with viewing panic failures as banks that were considered riskier than survivors at least as early as December 1931. In particular, panic failures experienced larger withdrawals in 1931, and were forced to rely on borrowed money from an early date.

Interest Rates on Debt

Interest rates on debt indicate debtholders' perceived risk of bank failure. For a small sample of Chicago banks (31) which were Fed members, we have data on the interest paid during the last six months of 1931 on each of the categories of debt discussed above (individual deposits, bank deposits, time deposits, and borrowed money), which we report in Tables 4a and 4b. The banks are grouped, as before, both according to failure risk and failure experience. It is important to keep in mind that our reported interest rate differences likely understate the true differences as of late 1931; the interest paid on each category of debt was paid over the last six months of 1931, and therefore, may not provide an accurate picture of interest rates paid in December 1931. Interest rates on borrowed money (the marginal source of funds for high-risk banks) are significantly higher for medium- and high-risk banks, and the cost of funds on this category of debt far exceeds the costs paid on demand deposits and time deposits (which are of shorter maturity). The interest rates on time deposits increase with bank failure risk, but differences are small and insignificant. Surprisingly, the interest paid on demand debt is lower for high-risk banks. This likely reflects sample-selection bias; as we argued before, the higher the risk of failure, the more demand deposits leave the bank - only the uninformed ("risk-inelastic") demand depositors remain. This interpretation is consistent with the significantly lower withdrawal rates of deposits for low-risk banks during 1931, shown in Table 4b.

Banks failing during the panic paid interest rates in 1931 that were identical on average to non-panic failures, and different from survivors. Interest rates paid (by debt category) for panic failures and non-panic failures matched those of high- and medium-risk banks. Despite small sample size and weak statistical significance, the results on interest rates provide additional evidence that panic failures and non-panic failures were viewed as similarly, high-failure risk categories of banks as early as 1931.

IV. Conclusion

We have compared the attributes of banks that failed during the Chicago panic of June 1932 to those of banks that failed at other times in early 1932, and to those of banks that survived the period. Each of our categories of comparison - the market-to-book value of equity, the estimated probability of failure or duration of survival, the composition of debt, the rates of withdrawal of debt during 1931, and the interest rates paid on debt - lead to the same conclusion: banks that failed during the panic were more similar to others that failed than to survivors. The special attributes of failing banks were distinguishable at least six months before the panic and were reflected in stock prices, failure probabilities, debt composition, and interest rates at least that far in advance.

We conclude that failures during the panic reflected relative weakness in the face of a common asset value shock rather than contagion. That does not mean contagion was absent, nor does it mean that the run on Chicago banks is a myth. Rather, we think it means that - consistent with James (1938) account of the management of the banking crisis - cooperative intervention by the Chicago clearing house prevented the failure of banks that were known to be solvent until the runs by uninformed depositors subsided. Absent such cooperation, the failure experience during the panic of June 1932 could have been very different. As in many other examples of banking panics prior to the Depression (Calomiris and Gorton, 1991, Calomiris and Schweikart, 1991, Calomiris, 1993), bank failures in Chicago in June 1932 were not a costly consequence of panic-induced contagion or confusion on the part of depositors about the riskiness of banks. Indeed, it may have been that identifying and closing insolvent banks helped to resolve the depositor information problems that had threatened solvent banks with runs during the panic.