Contagion and Bank Failures During the : The Banking Panic

By CHARLES W. CALOMIRIS AND JOSEPH R. MASON*

We examine the social costs of asymmetric-information-induced bank panics in an environment without government deposit insurance. Our case study is the Chicago bank panic of June 1932. We compare the ex ante characteristics of panic failures and panic survivors. Despite temporary confusion about bank asset quality on the part of depositors during the panic, which was associated with widespread depositor runs and bank stock price declines, the panic did not pro- duce significant social costs in terms of failures among solvent banks. (JEL G28, G21, N22, E58, E32)

Recent work in banking theory and history vent, but can observe a shock that affects has helped to define the potential causes and banks' portfolios, they may initiate runs on all costs of bank panics, which various authors banks, both solvent and insolvent. have argued can be traced to speculative at- Bank panics are short-lived phenomena, tacks on the numeraire (Barry A. Wigmore, historically measured in days or weeks. How- 1987; R. Glenn Donaldson, 1992), illiquidity ever, a panic still can have important long- shocks (Douglas W. Diamond and Philip H. lived costs if it results in the disappearance of Dybvig, 1983; Donaldson, 1993), or shocks solvent banking institutions. That concern is to bank asset values when there is information often invoked to justify the significant expan- asymmetry between bankers and depositors sion of the government safety net for U.S. about the incidence of those shocks (Calomiris banks during the 1930's (Anthony J. Saunders and Gary B. Gorton, 1991; Calomiris and and Berry Wilson, 1994). In this paper, we Charles M. Kahn, 1991; Calomiris and Larry take a close look at one of the clearest exam- Schweikart, 1991; Sudipto Bhattacharya and ples of an asymmetric-induced bank panic dur- Anjan V. Thakor, 1993; George G. Kaufman, ing the Great Depression, and ask whether 1994). In the latter case, when depositors can- solvent banks failed during that panic. not observe whether individual banks are sol- The answer to this question has important public policy implications. Studies of bank panics argue that panics induced by asset shocks and asymmetric information can be * Calomiris: Graduate School of Business, Columbia hard to resolve with monetary policy alone. In University, New York, NY 10027 and National Bureau of contrast, speculative attacks on the numeraire Economic Research; Mason: Bank Research Division, Of- fice of the Comptroller of the Currency, Washington, DC induced by uncertainty about its future value 20219. The authors thank Richard Grossman, Ed Kane, can be stopped by policies that resolve uncer- George Pennacchi, Berry Wilson, two anonymous refer- tainty about monetary policy (for example, by ees, seminar participants at Columbia University, George- a devaluation, as in the United States in March town University, and Universidad Torcuato di Telia, and 1933). Similarly, bank panics that result from conference attendees at the NBER Development of the American Economy meetings, the American Economic shocks to liquidity preference and a limited Association/Cliometric Society meetings, and the Finan- supply of aggregate liquid bank assets can be cial Management Association meetings for helpful resolved by traditional monetary policy in the comments. We are also grateful to Ven Vitale for research form of open-market purchases of securities assistance, and to the National Science Foundation (SBR- 9409768) for funding. Opinions expressed are the au- by the central bank (Bruce Champ et al., thors', not those of the Office of the Comptroller of the 1991). But bank panics caused by asymmetric Currency. information about the condition of banks 863

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 864 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

cannot be resolved by monetary policy. Values ante (and thus more vulnerable to asset price of individual bank loan portfolios (about decline) relative to banks that survived the which uninformed depositors are concerned) panic. We employ data from individual bank are not controllable by monetary policy failure experience, balance sheets, income and (Frederic S. Mishkin, 1991; Calomiris, 1994). expense statements, and stock prices for. fail- To resolve the problem of asymmetric- ing and surviving Chicago banks before and information-induced runs one must limit the during the panic. We analyze characteristics of risk depositors face as the result of asymmetric failing and surviving banks to determine information, and thus remove the incentive for whether the banks that failed during the panic depositors to demand immediate withdrawal. were similar ex ante to those that survived the This can either be done privately or publicly. panic. We find that panic failures were weaker Privately, banks can either act individually to than panic survivors, and argue that panic fail- reassure depositors or agree temporarily to ures can be attributed to asset value decline of stand behind each other's liabilities. So long failed banks rather than to depositor confusion as depositors are confident that the coalition of about the value of bank assets. mutually insuring banks is solvent collec- While depositors did confuse panic survi- tively, that collective action can bring the vors with panic failures, the failure of solvent panic to an end without resort to suspension of banks did not result from that confusion. One convertibility. Alternatively, the government reason such failures were avoided may be that can provide insurance of deposits, either in the solvent banks knew each other's condition bet- form of a commitment to pay depositors, or by ter than depositors, and had the incentive and lending cash to banks against their illiquid as- the ability to help each other avoid failure dur- sets at a subsidized rate. ing the crisis. Private cooperation by the Chi- This paper addresses the empirical question cago clearing house banks appears to have of whether private bank actions to stem been instrumental in preventing the failure of asymmetric-information runs are adequate, or at least one solvent bank during the panic. whether government deposit insurance is Section I provides historical background for needed. Specifically, we ask whether private the Chicago panic to support our use of the institutions can prevent the failure of solvent June crisis as an example of an asymmetric- banks during a bank panic. We examine that information-induced bank panic, and our iden- question in the context of the banking crises tification of the panic event window. Section of the Great Depression. The example we fo- II presents our empirical analysis of the char- cus on is the Chicago panic of June 1932. We acteristics of panic failures, panic survivors, choose this example for three reasons. First, and banks that failed outside the panic win- we argue it is a quintessential example of an dow. Section III concludes with a summary of asymmetric-information-induced panic. Sec- our findings and a discussion of their impor- ond, this was one of the most publicized tance and limitations. examples of a run on banks during the banking crises of the early 1930's, which coincided I. The June 1932 Banking Crisis in Chicago with the federal government's decisions to es- tablish the public safety net for banks. Third, Our discussion of the Chicago banking cri- by focusing on a particular location and epi- sis establishes five "facts" that support its use sode, we are able to clearly identify the for testing the value and limitations of private origins of the panic and to control for the ef- cooperation during asymmetric-information- fects of location, time, and macroeconomic induced panics. Together, these five facts es- environment—factors that might otherwise tablish that the Chicago panic resulted from complicate our analysis. location-specific asset shocks that were rele- Our strategy in the paper is as follows. We vant for bank portfolios; that it was a true use a variety of measures to judge whether asymmetric-information-induced panic in that banks that failed during the Chicago panic all banks (ex post solvent and ex post insol- were likely to have been solvent. We investi- vent) experienced heavy withdrawals and gate whether they were predictably weaker ex stock price declines during the panic; that (at

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 865

Number of Chicago Bank Failures, Daily

FIGURE 1. NUMBER OF FAILED OR SUSPENDED BANKS (DAILY AND MONTHLY), - Notes: Bank suspensions are from the Federal Reserve Bulletin, various issues. Bank failures consist of receiverships and voluntary liquidations, and come from the Annual Report of the Comptroller of the Currency, and the Statement Showing Total Resources and Liabilities of Illinois State Banks at the Close of Business, Superintendent of Banking of the State of Illinois, various issues. In order to accommodate the logarithmic scale on the right-handside , 0.1 is substituted for observed values of zero in the monthly data.

least in one case) banks were willing to sup- high at the state, Federal Reserve District, or port each other against the uninformed runs of national level in comparison to previous depositors; and that Chicago bank failures that months. In contrast, Chicago experienced a se- occurred outside the panic window did not co- vere concentration of failures during the week incide with similar (panic) events. of the panic. Of the 49 bank failures in the state of Illinois during that month, 40 took place in A. Was the June Crisis in Chicago Chicago, and 26 of these failed in the week of a Unique Event Nationally? -27 {Commercial and Financial Chronicle, July 2, 1932 p. 71 ).2 As Figure 1 shows, mid-to-late June of 1932 Deposit outflows indicate a similar pattern. witnessed concentration of bank failures in As shown in Figure 2, Chicago banks saw an Chicago, whether measured by the number or unusually large decline in their deposits during total assets of failed banks.1 The number of bank failures in June 1932 was not particularly

2 The reported * 'failure dates'' in the Commercial and Financial Chronicle of June 20— correspond to 1 Asset plots are available from the authors upon failure dates of - reported by state and request. national bank regulators.

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 866 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

Chi Total Deposits (3MA)

NY Total Note: There were 40 failures in Deposits Chicago during June, 24 of which (3MA) occurred in the panic week preceding

FIGURE 2. PERCENT CHANGE IN THE THREE-PERIOD MOVING AVERAGE OF TOTAL DEPOSITS, WEEKLY-REPORTING BANKS IN CHICAGO AND NEW YORK CITY, JUNE 1931-DECEMBER 1932

late June, and banks in other areas of the coun- These reports emphasized that long lines of in- try (notably New York City, the financial cen- dividual depositors formed at banks. Some ter) did not share in that precipitous decline. banks that were reported to have experienced large withdrawals (including First Chicago and B. Solvent and Insolvent Banks Both Continental) were able to withstand their runs Suffered During the Panic and remain open, while other banks (including one Loop bank—the Chicago Bank of Com- Contemporary chroniclers and economic merce), were forced to close. historians have pointed to the June banking Initially (before ), bank distress was crisis in Chicago as an important example of limited to a few banks, but this soon spread how a systemwide attack by depositors on and was associated with a dramatic decline in banks can produce pressure on solvent and in- aggregate deposits in Chicago banks. The dra- solvent banks alike. In 1932, the crisis re- matic withdrawals from downtown banks be- ceived national, as well as local, attention in gan on June 22 and reached their peak on the press. Contemporary reports clearly indi- Friday, . F. Cyril James (1938 p. 1034) cate that depositors ran ex post solvent as well distinguishes the panic in late June from pre- as ex post insolvent banks en masse. vious periods of banking distress in Chicago: The Commercial and Financial Chronicle (July 2, 1932 pp. 70-71) provided a detailed [previous] runs ... were directed against account of the runs on Chicago banks, and spe- particular banks that were known to be cifically noted that even healthy banks (includ- enfeebled; this one was directed against ing, for example, First Chicago) were affected. the whole Chicago money market and

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 867

the First National group, in the center of meant that the banks were forced to bear in- the battle, still had more than a hundred creased risk on their bond portfolios as the and twenty-five million dollars of cash flow of coupons was interrupted. Second, Chi- resources available, even though it had cago banks were called upon to purchase illiq- paid out fifty millions since Tuesday uid tax warrants to help keep the municipal night. In the case of earlier runs, the crowds had been drawn from a particular government afloat. Third, city workers were locality or a special group: this time peo- forced to draw down their bank deposits to pay ple from all parts of the city seemed to normal living expenses, thus reducing bank re- converge on the Loop in hysterical fear serves and increasing the proportion of (risky) and anxiety. loans and securities in bank portfolios. Not surprisingly, Chicago bankers saw the viabil- Bank stock prices fell during the panic, al- ity of the banking system and the financial though not all Loop banks experienced large problems of the city as closely related. A del- declines. Central Republic, First Chicago, and egation of Chicago city officials and citizens Continental were among the Loop bank survi- visiting Congress to request federal govern- vors whose stock prices declined the most. First ment assistance for the city in June 1932 Chicago stock fell from 150 (bid) at the close included many prominent bankers. The dele- on to a low of 131 (bid) at the close gation's request for $80 million in aid was re- on June 24. Continental's stock fell from 70 buffed by Congress on June 22 (Chicago (bid) on June 18 to a low of 60 (bid) on June Tribune, , 1932 p. 1). 27. Central Republic saw its stock price fall The municipal revenue crisis was sympto- from 52 (bid, June 18 close) to 47 (bid) on matic of the deep contraction in local asset , and then its stock plummeted to a price prices and economic activity. Among the of 4 (bid) on Monday, June 27. Afterwards, its many victims was the Insull utility empire in stock price, and those of the other surviving Chicago, whose stock and debt were widely Loop banks, rebounded rapidly. The stock of held by institutions and individual investors. the only Loop bank to fail during the panic, the Chicago utility companies grew dramatically Bank of Commerce, was trading at 9 (bid) on during the 1920's in anticipation of growing June 18-June 24 for a $20 par value. On June demand and were caught short by the sudden 25 its price ceased to be reported. decline in the local economy. The prices of Commonwealth Edison (ComEd) and Insull C. Local Adverse Economic News stocks and bonds show precipitous declines in Precipitated the Panic the first six months of 1932. From January through March, Insull stock declined by 76 Contemporary discussion of the crisis em- percent, Insull bonds fell by 62 percent, and phasized the adverse local economic news that ComEd stock lost 25 percent of its value. From had precipitated it. The panic did not reflect ex- March through early June the declines accel- ogenous liquidity-demand shocks. Rather than erated. Insull debt lost 98 percent of its value, withdrawing deposits to spend them, depositors Insull common stock lost 89 percent, and In- often redeposited those withdrawals in then- sull preferred lost 47 percent. postal savings accounts. James (1938) argues Chicago's economic problems were re- that the panic was triggered by several factors, flected in several local financial disasters and all of which combined in rapid succession to cases of bank fraud (an activity traditionally undermine depositors' faith in the value of Chi- more pronounced in bad times than in good) cago banks' assets. The bad news included fall- that made front-page news in Chicago day af- ing prices for local utility stocks and other ter day just prior to the crisis.3 John Bain, a corporate assets, a well-publicized local case of defendant in the most important bank fraud bank fraud and mismanagement, and a munic- ipal revenue crisis for the city of Chicago (Chi- cago Tribune, , 1932 p. A17). 3 Milton Esbitt (1986) emphasizes the importance of The city government's revenue problem management practices for explaining bank failures in Chi- weakened the banks in three ways. First, it cago in 1931.

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 868 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

case, was a local real estate developer who merce, experienced unusually severe declines owned a chain of banks. On , the 12 in their stock prices, the Bank of Commerce banks in the chain failed to open for business failed while Central Republic survived. Its sur- (James, 1938 p. 1033; Elmus R. Wicker, 1993 vival, however, depended on cooperation by p. 15). Not until June 23, however, did it be- large Loop banks to resolve its distress. come clear just how large the losses from fraud During the crisis, Central Republic was had been in the Bain chain. On that date the nearly taken down by its depositors. As doubts court released its estimate that the value of the about Central Republic's solvency grew and banks' assets was roughly $3.5 million, com- deposit withdrawal pressure mounted, the pared to total deposits of $ 13 million (Chicago bank's management prepared to close the bank Tribune, June 24, 1932 p. 9). voluntarily to avoid the risk of its being closed An important additional piece of evidence by bank depositors. Solvent banks that had lost that the panic was induced by fears of bank depositors' confidence had an incentive to insolvency, rather than by exogenous liquidity close preemptively to preserve stock value by demand by depositors, is that much of the avoiding the costs of liquidating bank assets funds withdrawn from banks were redeposited during a run, and the transaction costs asso- immediately in the form of riskless postal sav- ciated with having the bank taken over by a ings. The Commercial and Financial Chroni- regulator. This incentive was particularly cle (July 2, 1932 p. 71) reports that, during strong in 1932, when bank stockholders faced the week of June 20-June 27, "The Postal the threat of double liability on deposits, which Savings Department, which normally has 25 meant that liquidation costs borne by stock- or 30 windows in the Chicago post office ... holders could conceivably exceed the com- increased the number to around 100 ... to ac- plete loss of the bank's capital. commodate deposits." That report noted that Other Chicago banks saw the prospect of Cen- "... about $1,000,000 had been received [in tral Republic's voluntary liquidation as a poten- postal savings on June 27 ], compared to tial disaster for depositor confidence in their $2,000,000 and $3,000,000 a day at [the] peak banks. Bankers clearly believed that depositors [of the crisis], and [compared to] a normal were reacting to fears of bank insolvency and daily average of $200,000." were unable to distinguish between solvent and In summary, by June 23, Chicago bank de- insolvent banks. Prominent bankers from Chi- positors had witnessed, in a matter of only two cago and New York met as a group to devise a weeks, the collapse of some of the largest busi- plan to defend the Central Republic Bank and nesses in their city, several enormously costly Trust Co. from the continuing withdrawals. cases of bank fraud, and a deepening of the Fearing the spillover effects of a decision to liq- municipal financial crisis as the result of the uidate Central Republic, these bankers managed denial of relief to their city government by fed- to persuade General Dawes (its Chair) to con- eral authorities. All of these stories were front- tinue operating by offering an arrangement to page news day after day in the two weeks infuse Central Republic with new liquidity. leading up to the banking crisis, and the news The initial plan for the loan to Central Re- grew progressively worse. In this light, it is not public provided for $10 million in back-up li- surprising that depositors became increasingly quidity from New York and Chicago banks concerned about the ability of banks to pay out and $80 million from the Reconstruction Fi- their deposits, and transferred bank deposits to nance Corporation (RFC), but the final deal riskless postal savings accounts. involved assistance only from Chicago banks and the RFC. The deal that emerged combined D. Interbank Cooperation Helped to liquidity assistance from the RFC with, in es- Preserve Solvent Banks Under Pressure sence, a back-up credit enhancement by the Chicago banks.4 RFC liquidity support for the As already noted, the Loop banks experi- enced severe stock price decline and deposit withdrawals during the crisis. Although two 4 It is important to keep in mind that the RFC was the banks, Central Republic and the Bank of Com- only entity charged with helping avoid the insolvency of

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 869

Chicago banks—like all RFC lending during formation. With the exception of the June this period—was fully collateralized by very panic, no contemporary chronicler or scholar high-quality, liquid assets; credit risk on the of which we are aware has identified any other RFC loan to Central Republic was borne in time interval during 1932 as a panic or banking largest part by the Chicago banks that formed crisis in Chicago. Neither has anyone identi- the lending syndicate.5 Importantly, the RFC fied any nationwide banking panics as having agreed to allow municipal tax warrants—$30 occurred during 1932 (Milton Friedman and million of which had been sold to Loop banks Schwartz, 1963; Wicker, 1980, 1993; Ben S. (Chicago Tribune, June 25, 1932 p. 6)—to Bernanke, 1983; Calomiris, 1993; Clifford qualify as collateral for its loan.6 Once the cri- Thies and Daniel Gerlowski, 1993; Eugene N. sis passed, Central Republic saw its deposit White, 1984). outflows cease and its stock price increase. We have reviewed a variety of facts that Central Republic was a solvent bank saved identify the Chicago bank panic of June 1932 from failure by the collective intervention of as a quintessential example of an asymmetric- other Loop banks. information-induced panic, resulting from lo- cal economic problems that affected bank E. The June 20-27 Crisis asset values. Just as important for our pur- Was a Unique Event poses is the uniqueness of the panic window. The crisis, as reflected in sudden deposit Having argued that the June banking crisis withdrawals and stock price declines, and is an example of a panic induced by asset value widespread coverage in the local and national decline and asymmetric information, we turn press, was brief and was surrounded by times to the question of whether there were other in which bank failures did not coincide with such episodes in Chicago during the first six a panic. months of 1932. That question is relevant for our discussion in Section II, where we will use II. Failures and Survivors During the Panic the absence of panics during that period (out- side the window of the June crisis) as an iden- We now return to our central question of tifying restriction to investigate whether whether the absence of government deposit in- failures during the panic were similar to fail- surance promotes the failure of solvent banks ures outside of the panic. during asymmetric-information-induced pan- Bank failures during 1932, and more gen- ics. Specifically, we investigate whether sol- erally during the Great Depression, for the vent banks were able to survive withdrawal most part have not been identified by histori- pressures (partly via private coordination) ans as resulting from panics or asymmetric in- during the June 1932 Chicago bank panic. To answer that question we make use of the fact that the panic was a unique event during 1932. We assume that outside the panic window (in individual banks. At this time, Federal Reserve Banks did early 1932) banks that failed were actually in- not view the prevention of bank insolvency as their man- solvent. We then use the characteristics of date. This is in sharp contrast to more recent experience during the 1980' s. For a review of the history of Fed lend- those nonpanic failures to evaluate the likely ing policy, see Anna J. Schwartz (1992). solvency or insolvency of banks that failed 5 Joseph R. Mason (1995) argues that prior to its use during the panic. of preferred stock purchases to assist banks, the RFC was If banks that failed during the panic were not effective in stemming bank failures. James (1938 p. just as strong (according to some set of crite- 1044) cites the view, common at the time, that because of the strict collateral requirements on RFC lending, RFC ria) as those that survived during the panic, assistance often increased the credit risk faced by bank that would provide evidence in favor of the depositors. null hypothesis that confusion on the part of 6 Abbreviated bank balance sheets were routinely re- depositors about bank quality produced ran- ported in newspapers following call dates. Thus the Reports of Condition published in the Chicago Tribune dom bank failure. If, on the other hand, banks on July 2 provided further evidence of the soundness of that failed during the panic were weaker than Chicago banks (Chicago Tribune, pp. 18-24). bank survivors, then panic failures cannot be

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 870 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

viewed as purely random. That implies a re- of the market value of equity to the book value jection of the null hypothesis. of equity; (2) the estimated probability of fail- It is not difficult to reject this "strong" ver- ure of banks; (3) the rate of decline in bank sion of the null hypothesis, but that does not deposits; and (4) the interest paid on bank prove that only insolvent banks failed during debts. These various measures of bank risk are the panic. Depositor confusion might have pro- available for different subsets of Chicago duced the failure of some solvent banks, with banks, depending on the availability of data on potentially large social costs, even if on average stock prices and interest paid on deposits. banks that failed during the panic were weaker Stock prices are not available for all Chicago than those that survived it. Thus in our empir- banks, and interest paid is only available for ical work we also address a "weak" form of Fed member banks. The data set for the study the null hypothesis—that panic-induced fail- consists of several components: balance sheet ures were not purely random, but were impor- data, income and expense data, and stock price tantly random. This version of the null data. Balance sheet data from , hypothesis is inherently difficult to reject for- 1931 call reports were collected for all state mally on the basis of ex ante statistical tests of and national banks in Chicago, a total of 123 means, medians, and regression coefficients. banks. Total assets and total deposits were also Combining these with additional evidence, collected for December 31, 1930, to permit however, we argue that the social costs of the calculation of the changes in those variables unwarranted closure of solvent institutions (if during 1931. Balance sheet data for the 22 na- any) must have been very small. tional banks and 11 state banks that were Our empirical discussion divides into two members of the Federal Reserve System come parts. First, in subsection A we present evi- from microfilm of the original Reports of Con- dence that leads to the rejection of the strong dition filed with the Office of the Comptroller form of the null hypothesis. Then we confront of the Currency (OCC) and the Board of Gov- and reject the weak form of the null hypothe- ernors of the Federal Reserve System. State sis, using statistical evidence as well as infor- nonmember bank balance sheet data are from mation from bank examiner reports. the compilation of Statements of State Banks of Illinois issued by the Superintendent of A. Comparisons of Bank Attributes Banking of the State of Illinois. The disaggre- Leading Up to the Panic gated Reports of Condition of member banks facilitated aggregation of balance sheet cate- We divide the Chicago banks in our sample gories to reporting standards comparable with into three groups: panic failures (banks that the Statements of State Banks of Illinois. The failed during the panic, June 20-27), non- stock prices for Chicago banks are end-of- panic failures (banks that failed at other month observations published in the Bank and dates), and survivors (banks that did not fail Quotation Record (of the Commercial and Fi- in the first seven months of 1932). We then nancial Chronicle). Interest payments are compare the ex ante attributes of these three available only for Fed member banks (from groups.7 the Reports of Condition). In our analysis of survivors, panic failures, and nonpanic failures, we focus on four ex 1. Market-to-Book Value Ratios. Figure 3 ante measures of bank condition: (1) the ratio plots the means and 50-percent inclusion ranges for market-to-book value ratios for the three separate groups of Chicago banks (sur- vivors, panic failures, and nonpanic failures). 7 The dates we choose for the panic window are con- To adjust for survivorship bias in plotting sistent with James's (1938) discussion, newspaper ac- these trends, we retain failed banks in our sam- counts of the beginning and end of the panic, and the daily ple after their date of failure, and assume that movements of the stock prices of the ten Loop banks re- ported in the Chicago Tribune, which reached their nadir their stock value after failure is zero. on June 27. The results are robust to reasonable alternative The striking fact illustrated by Figure 3 is definitions of the panic window. that as early as the banks that

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 871

•Panic Failures •Nonpanic Failures Survivors

FIGURE 3. RATIOS OF MARKET-TO-BOOK VALUES OF EQUITY, END-OF-MONTH QUOTES, JANUARY 1931- Notes: Survivorship bias can arise if banks that fail are excluded from subsequent means and standard deviations. We correct for this bias by including failed banks and assuming they had zero market value of equity after the date of failure.

survived the June panic appeared to be a dis- 1931) to compute scores predicting failure tinct group with higher average market-to- during the first seven months of 1932. We book ratios. The banks that failed during the compare the probabilities of failure for the panic generally had slightly higher average ra- three groups of banks (panic failures, non- tios than those that failed at other times, but panic failures, and survivors). Our ex ante throughout the prepanic period (January 1, scores indicate that panic failures and non- 1931-June 20, 1932) the market-to-book panic failures on average were weaker banks value ratios of panic failures were very close than survivors at least as early as the end of on average to those of prepanic failures and 1931. very different from those of panic survivors. We estimate the probability of failure using By , most of the panic failures a logit model of the links between bank char- had market-to-book ratios less than unity. Fig- acteristics (e.g., balance sheet ratios) and bank ure 3 shows that all Chicago banks suffered failure. The danger of using ex post failures to from capital decline during 1931 and 1932, estimate failure risk, of course, is that special and that the banks that failed during the June events with low probabilities may have influ- panic reached and maintained unusually low enced actual failure experience during the panic market-to-book value ratios long before the in ways that were unpredictable ex ante and panic. possibly unrelated to underlying insolvency. For example, if a panic were a common shock 2. Failure Predictions. Next, we use ex ante to all banks, then the level of reserves might do observable characteristics of Chicago banks an excellent job of forecasting panic failures (based on bank data reported in December even if the banks that failed during the panic

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 872 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

did not have higher ex ante probabilities of fail- composition. Bank liability composition is a ing. Thus using ex post panic failures to con- useful signal of weakness because—as White struct a model of ex ante bank weakness may (1984), Wheelock (1992), Calomiris and bias toward identifying panic failures as inher- Wheelock (1995), and Mason (1995) ently weak when in fact they were not. To avoid argue—reliance on high-interest, borrowed (or at least minimize) this problem, we report funds was an undesirable necessity only suf- logit failure forecasts constructed from both in- fered by higher-risk banks (see also our dis- sample and out-of-sample estimation.8 In the cussion of debt composition below). While out-of-sample forecasts, we exclude banks that not all variables prove significant in our logits, failed during the panic from the sample when we retain measures of the basic concepts that we estimate the coefficients relating bank char- previous studies have found to be important acteristics to the probability of failure. This even if they do not prove statistically signifi- constrains the panic failures to be predicted us- cant in our sample. Excluding them would not ing model parameters that were constructed to affect our results. We also experimented with explain nonpanic failures, and thus prevents including two other variables (not reported in special unpredictable events during the panic Table 1): the ratio of book net worth to assets from influencing predictions of failure. and the percent changes in deposits or assets Our logit results are reported in Table 1. We of banks from to December include the following variables (all measured 1931. In neither case did the regressors add at year-end 1931) in our specification: size predictive power to our models. (log of total assets), the reserve-to-demand The results in Table 1 are quite similar for deposit ratio (where reserve assets are defined the in-sample and out-of-sample specifications, as cash and government securities), the real which is consistent with the view that failures estate loan share (defined as the ratio of loans during panics were similar events to nonpanic on real estate to total loans), the ratio of real failures. The variable coefficients that are sig- estate owned to illiquid assets (which mainly nificant are of the expected signs. Banks with includes repossessed real estate collateral, and higher reserve ratios, higher ratios of retained excludes bank premises), the ratio of last earnings to net worth, and lower proportions of year's retained earnings to net worth, and the long-term debt were less likely to fail. long-term debt ratio (bills payable plus redis- Table 2 reports the mean and median pre- counts plus time deposits, divided by total dicted failure probabilities for the logit models assets). by category of bank (panic failure, nonpanic This combination of variables also forms the failure, and survivor), and the significance lev- basis of the logit models of White (1984), els for tests of differences across categories in David C. Wheelock (1992), Calomiris and means and medians. These results indicate that Wheelock (1995), and Mason (1995), all of the banks that failed during the panic were less which are used to forecast bank failures for the risky ex ante than banks that failed outside the 1920's and 1930's. This specification typically panic and more risky than survivors. Compar- is interpreted as capturing measures of each of isons using predicted values from in-sample the following fundamental bank characteris- and out-of-sample regressions are similar. By tics: bank size, asset liquidity, exposure to real construction, the in-sample results show less of estate market risk, nonperforming loans (real a difference between panic and nonpanic fail- estate owned), recent bank performance (re- ures. Also by construction, out-of-sample fore- tained earnings/net worth), and bank liability casts tend to have lower probabilities of failure. Our results are consistent with the notions that panic failures were much weaker banks than panic survivors, and that failures during the 8 We also estimated logit models for prepanic failures panic were a continuation of the same process only (excluding the failures that occurred during or after that underlay other failures. the panic). The results were essentially identical to those we report below for nonpanic failures (which include fail- ures that occurred after the panic), and so we do not report 3. Deposit Withdrawals and Debt Composi- them here. tion. If panic failures had been relatively weak

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 873

TABLE 1—LOGIT MODEL RESULTS

Out-of-sample In-sample Dependent variable of all models: Bank failure N[0, 1] used for significance levels Number of observations 86 114 Number of panic failures 0 28 Number of nonpanic failures 18 18 Log-likelihood -18.95 -51.37 Restricted (slopes = 0) log-likelihood -44.12 -76.88 Chi-squared statistic (k — 1 df) 50.34 51.03 Significance level 3.38E-09 2.28E-09 Variable name Coefficient Coefficient Standard Standard error error Constant 3.31 1.49 7.40 3.73 Bank size (log of total assets) -0.75* -0.20 0.54 0.25 Ratio of reserves to demand deposits -4.47*** -2.97*** 1.81 0.83 Real estate loan share 2.10 -1.39 2.78 1.71 Ratio of other real estate owned to illiquid assets 13.89 2.09 12.16 8.25 Ratio of net earnings to net worth -25.33*** -15.20*** 70.27 5.30 Long-term debt 23.31*** 12.05*** 6.48 2.64

* Statistical significance at = 0.10. ** Statistical significance at = 0.05. *** Statistical significance at = 0.01.

institutions for months prior to the panic, then pared to 55 percent for nonpanic failures, and they should have experienced larger rates of 33 percent for survivors. depositor withdrawal before the panic. It is not The higher rate of decline in deposits for possible to obtain comparable records of de- panic failures and nonpanic failures during posit accounts for failed banks after the De- 1931 is reflected in the debt compositions of cember 1931 call, but one can ask whether those banks. Detailed data on the composition panic failures experienced relatively large de- of bank liabilities are available for all banks in posit withdrawals from December 1930 to De- our sample, either from Federal Reserve or cember 1931. Table 3 reports data on the rate state call reports. Table 3 presents data on the of decline of deposits over that year. Clearly, liability composition of banks as of December panic failures and nonpanic failures experi- 1931. enced much larger withdrawals than survivors Interestingly, the shares of the various debt in 1931. Panic survivors experienced an av- categories vary systematically across the three erage decline in deposits of 41 percent, com- groups of banks. In particular, panic failures

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 874 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

TABLE 2—MEANS AND MEDIANS OF FAILURE PROBABILITIES, BY CLASS OF BANK

In-sample logit Out-of-sample logit Mean Median Mean Median Nonpanic failures Score 0.753 0.796 0.669 0.780 Standard error 0.042 0.033 0.068 0.053 Number 18 18 18 18 Panic failures Score 0.556 0.625 0.269 0.166 Standard error 0.043 0.059 0.055 0.106 Number 28 28 28 28 Survivors Score 0.248 0.187 0.088 0.005 Standard error 0.030 0.055 0.021 0.015 Number 68 68 68 68 t-statistics for tests of differences Panic, survivors 5.66*** 4.67*** 3.75*** 2.25*** Panic, nonpanic 3.11*** 2.19** 4.57*** 4.42*** Nonpanic, survivors 8.13*** 5.61*** 10.78*** 19.32***

* Significant at =0.10. ** Significant at = 0.05. *** Significant at = 0.01.

(like nonpanic failures) tend to rely much example, in referring to the trouble at the Hyde more on borrowed money (defined as bills Park-Kenwood National Bank, the examiner payable and rediscounts). As we have noted wrote that: "... practically all of the bank's above, the standard interpretation of this bonds and securities are pledged [as collateral finding—which is consistent with observed for borrowed money] and the bank is now a differences in deposit withdrawal rates across heavy borrower, the Chief Examiner advising categories during 1931 reported in these that the total borrowings on January 28 tables—is that when demandable debt is with- amounted to $684,000 due to the heavy de- drawn from risky banks, those banks are cline in deposits." forced to rely on high-cost borrowed money, In summary, panic failures and nonpanic fail- typically collateralized by liquid bank assets. ures experienced significantly larger withdraw- As noted above, other studies have found that als of deposits long before the panic. Consistent the share of borrowed money is a reliable pre- with this deposit withdrawal pressure, panic dictor of bank failure during the 1920's and nonpanic failures alike experienced fun- and 1930's (White, 1984; Wheelock, 1992; damental debt financing reallocations charac- Calomiris and Wheelock, 1995; Mason, teristic of troubled institutions. 1995). Moreover, examiners from the Office of the Comptroller of the Currency used a re- 4. Interest Rates on Debt. Interest rates on liance on borrowed money as a clear indica- debt should indicate debtholders' perceptions tion that a bank was having trouble. For of the risk of bank failure. If panic failures and

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 875

TABLE 3—DEPOSIT AND INTEREST RATE COMPOSITION, BY CLASS OF BANK

Change Change in in Interest total total Demand Due to Time Borrowed on total deposits assets deposits banks deposits money debt Survivors Mean -0.3251 -0.2145 0.5098 0.0301 0.4600 0.0197 0.0062 Standard error 0.0363 0.0491 0.0226 0.0078 0.0245 0.0070 0.0005 Number of obs. 62 63 68 68 68 68 18 Panic failures Mean -0.4115 -0.3397 0.4911 0.0216 0.4873 0.0831 0.0093 Standard error 0.0595 0.0299 0.0314 0.0058 0.0315 0.0159 0.0009 Number of obs. 28 28 28 28 28 28 11 Nonpanic failures Mean -0.5514 -0.3979 0.3835 0.0053 0.6113 0.1872 0.0116 Standard error 0.0316 0.0242 0.0336 0.0029 0.0325 0.0301 0.0014 Number of obs. 18 18 18 18 18 18 3 Tests of differences between means (t-statistics) Nonpanic, panic 1.777** 1.381* 2.263*** 2.131** 2.629*** 3.338*** 1.161 Panic, survivor 1.287* 1.637* 0.461 0.667 0.630 4.249*** 3.386*** Nonpanic, survivor 3.245*** 1.971** 2.672*** 1.618* 2.99*** 8.192*** 4.201***

Notes: Deposits are presented as a proportion of total deposits, equal to demand deposits, interbank deposits, time deposits, and bills payable and rediscounts. Interest is reported as interest expense as a proportion of the relevant deposit category, i.e., demand deposit interest expense/demand deposits. Interest is calculated as the amount of interest paid over the last six months as a proportion of the total in each category of debt as of December 31, 1931. Changes in total assets and deposits are from December 31, 1930 to December 31, 1931. * Significant at =0.10. ** Significant at = 0.05. *** Significant at =0.01.

nonpanic failures were more likely to fail ex ence. It is important to keep in mind that our ante they should have been forced to pay reported interest rate differences capture the higher interest on their debt prior to the June experience of only a small sample of banks, panic. For a small sample of Chicago banks and are measured with error because we divide (31) that were Fed members, we have data on interest flows over a six-month period by end- the interest paid during the last six months of of-year balances. Therefore, these data on each of the categories of debt dis- not provide an entirely accurate picture of in- cussed above (individual demand deposits, terest rates paid as of . bank deposits, time deposits, and borrowed In the column entitled interest paid on total money). We report the aggregate amounts of debt we compute the means for each of the these in Table 3 as a fraction of the respective three categories of banks of the ratio of total outstanding debts shown on the December 31, interest paid relative to total debt. We find that 1931 balance sheets. The banks are grouped, panic failures and nonpanic failures paid sig- as before, according to their failure experi- nificantly higher interest rates on debt than

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 876 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

survivors. Panic failures paid an overall inter- survivors implies that the failure process was est cost that was 50 percent higher than that of less discriminating during the panic—that is, survivors, and nonpanic failures paid nearly that the traits of panic failures reflect a mix of double the interest rate paid by survivors.9 solvent and insolvent institutions. In summary, we find that panic failures and But such an assumption is surely incorrect. nonpanic failures paid higher interest rates on In Section I we presented evidence that local their debt than survivors. The higher interest asset values (and the value of bank portfolios) paid by those two classes of banks reflected declined dramatically in the first half of 1932. their relative reliance on high-cost funds (bor- This implies that the failure threshold for rowed money and time deposits) rather than banks was shifting over that period. In an en- higher interest costs on demand debt. vironment of persistently declining asset prices, the first banks to fail (nonpanic fail- B. The Role of Declining Asset Values ures) will appear measurably weaker than in Chicago Bank Failures banks that fail only after asset values have fallen much more (panic failures). To control The fact that panic failures appear to have for changes in the probability of failure within been stronger institutions on average than non- our period, we estimate a survival duration panic failures, as measured by their publicly model of bank failure. This model supports the available financial data as of the end of 1931, argument that declining fundamentals can ex- has several possible explanations. One possibil- plain the quality difference between (early) ity is the weak form of the null hypothesis (that nonpanic failures and (late) panic failures. a significant number of solvent banks failed dur- Our survival duration model is similar to our ing the panic). If one believed that bank char- logit model except that it forecasts the length acteristics (as measured in December 1931) of time the bank will survive (measured in accurately reflect unchanging cross-sectional days after December 31,1931), and allows for differences in bank condition throughout the pe- changes in the underlying transition probabil- riod January-July 1932, then panic failures con- ities during our period, i.e., the conditional sequently appear excessive. That is, under the probabilities of failure on any given day, via assumption of unchanging bank condition, the a logistic hazard function. This hazard func- fact that panic failures' characteristics lie be- tion effectively separates the effects of tween those of nonpanic failures and those of changes in the probability of failure across in- dividuals from shifts in the baseline probabil- ity of failure associated with diminishing fundamental bank asset prices (Guido W. 9 We also examined the breakdown of interest paid ac- Imbens, 1994 p. 703). The implied baseline cording to each category of debt. Differences in total de- probability of failure estimated in our model posit risk show up in withdrawals of demand deposits from banks, in changes in relative weights attached to var- increases at a decreasing rate from January to ious types of debt, and in overall debt costs, but not in June, and declines during July. A technical demand deposit interest rate differences. Other research Appendix, available from the authors upon re- examining links between bank-failure risk and demand de- quest, provides a formal discussion of our sur- posit interest rates during the Great Depression has also failed to find a positive relationship between demand de- vival duration model. posit interest rates and bank-failure risk. George J. The results of our survival duration model Benston (1964) analyzed banks during the period 1929- are reported in Table 4. The results are quali- 1935 and found no significant positive relationship be- tatively similar to those for the logit model in tween demand deposit interest rates and failure risk. As in our sample, he sometimes found a negative (and insignif- Table 1, although, of course, coefficients in the icant) relationship between the two. One explanation for two models are of opposite sign. As shown in these findings is provided by Gary B. Gorton and George Table 4, our survival duration model is capable G. Pennacchi (1990), who argue that some bank deposi- of approximating the gaps in time between tors may be very unwilling to accept increased risk on their accounts. Risk-intolerant depositors may prefer to adjust nonpanic failures and panic failures. Table 5 to changes in bank riskiness via changes in the quantity illustrates that the model overpredicts survival of balances they hold with a bank rather than changes in duration on average for both panic failures and the interest paid on those balances. nonpanic failures. That is, using the same scor-

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 877

TABLE 4—SURVIVAL MODEL RESULTS

Out-of-sample In-sample Dependent variable: Log of time elapsed (in days) from December 31, 1931 Number of observations 86 114 Number of panic failures 0 28 Number of nonpanic failures 18 18 Log-likelihood -39.32 -84.36 Restricted (slopes = 0) log-likelihood -67.08 -117.30 Chi-squared statistic (k - 1 df) 55.51 65.88 Significance level 3.65E-10 2.85E-12 Coefficient Coefficient Variable name Standard error Standard error Constant 1.77 4.53*** 5.21 1.76 Bank size (log of total assets) 0.53 0.13 0.42 0.12 Ratio of reserves to demand deposits 2.69*** 1.44*** 0.98 0.39 Real estate loan share -0.61 0.58 1.62 0.72 Ratio of other real estate owned to illiquid assets —4.95 0.04 7.29 3.33 Ratio of net earnings to net worth 10.52** 7.41*** 6.07 2.40 Long-term debt -12.15*** -5.80*** 3.63 1.10

* Statistical significance at = 0.10. ** Statistical significance at = 0.05. *** Statistical significance at =0.01.

ing model (based on December 1931 charac- were unwarranted and nonpanic failures were teristics), and allowing for time variation in warranted. In other words, when one accounts the hazard function, we estimate mean sur- for the time-varying probability of failure for vival duration for nonpanic failures of 192 all banks, the model does as well estimating days, compared to 349 days for panic failures, cross-sectional hazards during the panic as while the actual respective survival means prior to the panic. were 107 and 177 days. But the model accu- rately estimates the relative health of panic and C. Further Evidence the Panic Entailed nonpanic failures. Our model predicts that Low Social Costs nonpanic failures survive (on average) 60 per- cent as long as panic failures, and in fact they The duration survival model results are averaged 55 percent of the survival time of consistent with the view that only observa- panic failures. Thus our model does not un- bly insolvent banks failed during the panic. derpredict panic failures relative to nonpanic But these findings do not constitute a formal failures, as one would expect if panic failures rejection of the weak form of the null

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 878 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

TABLE 5—MEANS AND MEDIANS OF DURATION PREDICTIONS (IN DAYS FROM DECEMBER 31, 1931), BY CLASS OF BANK

In-sample duration Actual duration Mean Median Mean Median Nonpanic failures Score 192 168 107 131 Standard error 24 25 NA NA Number 18 18 18 18 Panic failures Score 349 253 177 177 Standard error 49 38 NA NA Number 28 28 28 28 Survivors Score 1,482 688 NA NA Standard error 308 99 NA NA Number 68 68 68 68 t-statistics for tests of differences Panic, survivors 2.35*** 2.78*** NA NA Panic, nonpanic 2.46*** 1.66** NA NA Nonpanic, survivors 2.15** 2.69*** NA NA

* Significant at =0.10. ** Significant at = 0.05. *** Significant at = 0.01.

hypothesis—that some solvent banks failed Table 6 presents data on the distributions of during the panic, and that the social costs logit scores for the three groups of banks using from these failures were important. To in- in-sample and out-of-sample estimation. Note vestigate that possibility, we take a closer that none of the panic failures has an out-of- look at panic failures, particularly at "out- sample score that is as low (that is, as good) as liers' ' that appear (on the basis of observable the top quartile of survivors. The minimum traits in December 1931) to have been (best) out-of-sample score of the panic failures healthy institutions. Examination reports on is 0.00059, and the cutoff for the lowest (best) the condition of these outliers reveal deep quartile of survivors is 0.00025. Only six panic problems in these institutions prior to the failures had out-of-sample logit scores that were panic, which were publicly known but not lower than or equal to the median of survivors. captured by 1931 balance sheet ratios. In Were these six panic failures examples of many cases, bank fraud and accounting ir- banks that were solvent but allowed to fail by regularities explain why banks that failed their fellow bankers? If so, were the social during the panic appear stronger statistically costs of those failures high? It is easier to an- (on the basis of 1931 accounting data) than swer the second question. These six banks col- they did to contemporaries, for whom their lectively represented a trivial proportion of the problems were common knowledge by mid- bank assets of Chicago (1.2 percent of total 1932. assets as of December 1931), and while it is

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 879

TABLE 6—DISTRIBUTIONAL ANALYSIS OF LOGIT SCORES FOR EX POST PANIC FAILURES AND SURVIVORS

Percent of survivors Percent of survivors estimated below estimated below In-sample p(fail), that probability Out-of-sample p(fail), that probability panic failures (in-sample logit) panic failures (out-of-sample logit) Minimum 0.048 0.311 0.001 0.326 25th percentile 0.448 0.782 0.038 0.697 Median 0.626 0.910 0.175 0.832 75th percentile 0.701 0.932 0.469 0.929 Maximum 0.905 0.988 0.906 1.000 Percent of panic failures Percent of panic failures estimated below estimated below In-sample p(fail), that probability Out-of-sample p(fail), that probability survivors (in-sample logit) survivors (out-of-sample logit) Minimum 0.000 0.000 0.000 0.000 25th percentile 0.040 0.000 0.000 0.000 Median 0.162 0.038 0.006 0.139 75th percentile 0.403 0.222 0.074 0.300 Maximum 0.943 1.000 0.821 0.918

conceivable that some of them were solvent Comptroller's office. That examination re- banks, the social costs of their demise cannot vealed that the bank had experienced large un- have been very large. accounted losses that left its capital impaired, To answer the first question we explored the and placed it in violation of its charter. The specific circumstances of these six panic fail- Deputy Comptroller wrote that "the officers ures and their financial condition prior to fail- and directors have been operating the bank ure. We searched the records of the OCC for along unsound lines." In particular, the Dep- any relevant examinations and correspondence uty Comptroller criticized the bank's manage- with respect to the two national banks included ment for allowing a large loan to the bank's in the group of six (South Ashland National Director/Manager that produced an enormous Bank and Standard National Bank). We were loss for the bank. The examiner, in his May 5 able to locate information about both of these letter to the OCC, wrote: "This bank is now banks. Prior to the panic, both of these banks under the incompetent management of Direc- had experienced large loan losses and were un- tor James G. Hodgkinson, who dominates the der investigation by the U.S. Attorney General policies; he is absolutely broke and the manner and the OCC for fraud. in which he has furthered his own interests is most reprehensible. A report on his operations The records we discovered for South Ash- land clearly indicate that this bank was on the has been made to the United States District verge of failure at least two months prior to Attorney." The details of the examination re- the panic, and possibly earlier. While the bank veal fraudulent activities, including check kit- was closed on June 24 and placed in the hands ing by Hodgkinson. of a receiver on June 27, South Ashland's Vice President, Guy Brown, had written to the OCC Standard National Bank was also involved as early as to announce that the bank in a case of fraud. Its Vice President, who was had decided on May 25 to liquidate itself in also the Vice President of another bank that response to an April 27 examination by the failed during the panic (People's National Bank and Trust Co.) confessed to appropriating

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 880 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

bank funds to finance his speculation in the (2) National Bank of Woodlawn (Unpub- stock market. The individual and the two lished examination report, April 25, banks were all under investigation by the 1932): "The report of an examination ... OCC and the U.S. Attorney General as early completed April 25 ... shows a bond de- as . preciation ... which greatly exceeds the South Ashland and Standard are interesting bank's entire capital, surplus, and undi- examples of banks whose accounts as of De- vided profits ...." cember 1931 do not provide pictures of their (3) Bowmanville National Bank (Unpub- true position prior to the panic. While their lished examination report, Letter of April scores in our models are very strong, their 6,1932): "The report of the examination condition according to the examiners was ex- ... completed February 25 shows such an tremely weak. It is possible that some or all unsatisfactory condition it is requested of the four state bank outliers may be expli- that you hold a meeting with the directors cable in similar terms. After all, if the strength or a committee thereof and ascertain of a six-month-old balance sheet were enough whether or not something further can be to conclusively indicate a bank's strength, done to strengthen the bank, after which asymmetric-information panics could never this office would be advised fully of the occur. conclusions reached ... the solvency of The Comptroller's examination reports indi- [the bank] is questioned in view of the cate that information about fraud and risk taking doubtful and loss items and the bond by banks that failed during the panic surfaced depreciation." between December 1931 and . (4) Midland National Bank (Unpublished ex- Clearly, investors in bank stock detected special amination report, April 27, 1932): "This problems in the banks that failed during the little institution is struggling along prob- panic in those same months. As Figure 1 shows, ably as best as could be expected under the market-to-book value ratios for panic failures the circumstances ... Loss of $1,000,000 fell sharply from January through April 1932. in deposits within a year has just about The market-to-book ratio of surviving banks did taken away earning capacity." "The re- not decline nearly as precipitously. port ... shows a bond depreciation of The wealth of OCC file material we discov- $162,004 and losses of $16,041 in loans, ered led us to search for the examination rec- which consume the surplus fund, undi- ords of the other national banks in our sample vided profits and reserve for contingen- that failed during the June panic. It is difficult cies and impair the bank's capital to the to quantify the statements of examiners (to extent of $67,414." convert them to scores), but it is easy to sum- (5) Hyde Park-Kenwood National Bank (Un- marize their content. We were able to locate published examination report, February information for all but one of the other national 10,1932): "The report of an examination banks that failed during the panic. In every of your bank completed ... case for which we have records, the bank ex- shows an exceedingly unsatisfactory con- aminers had indicated extreme problems at the dition and that you are confronted with a bank at least as early as the end of April 1932. serious situation. This conclusion is based Following are excerpts from (or synopses of) upon the slow and doubtful assets ... the examiner's remarks about each of these shown in the report; the fact that practi- banks: cally all of the bank's bonds and securities are pledged and the bank is now a heavy (1) Jackson Park National Bank (Unpub- borrower, the Chief Examiner advising lished examination report, April 27, that the total borrowings on January 28 1932): "... the condition of this institu- amounted to $684,000 due to the heavy tion remains highly unsatisfactory from decline in deposits." every angle. It will be noted criticized as- (6) Ravenswood National Bank (Unpub- sets have increased materially since last lished examination report, April 18, examination." 1932): "The report of an examination of

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 881

your bank completed April 18 ... dis- probabilities, the opinions of bank examiners, closes several unsatisfactory conditions debt composition, and interest rates. as set out by the examiner throughout the We conclude that failures during the panic report and itemized on page 11, the cor- reflected the relative weakness of failing banks rection of which should be effected as in the face of a common asset value shock rapidly as possible. The report shows a rather than contagion. The panic was precipi- high percentage of the bank's loans to be tated by exogenous asset-price decline, and the in very unsatisfactory condition ...." By banks that failed during the panic were among May 9, accounting fraud was also dis- the weakest banks in the city. While asym- covered to have taken place by a former metric information between depositors and employee. banks precipitated a general run on banks, our evidence suggests that this asymmetric- These qualitative statements about national information problem did not produce failures banks that failed during the panic reinforce the of solvent banks. evidence from Tables 2 and 3 that panic fail- If the risk of solvent banks failing during an ures were among the weakest banks in the sys- asymmetric-information panic is not high (as tem at the time of the panic.10 the evidence from the Chicago panic sug- gests), that could have important implications III. Conclusion for bank regulatory policy. Deposit insurance and government assistance to banks since the We have compared the attributes of banks Depression have been motivated in part by the that failed during the Chicago panic of June perception that bank failures during the De- 1932 to those of banks that failed at other pression were a consequence of contagion, times in early 1932, and those of banks that rather than the insolvency of individual banks. survived the period using a variety of stan- If private interbank cooperation, buttressed by dards of comparison. Comparisons of the liquidity assistance from the monetary author- market-to-book value of equity, the estimated ity (like the assistance provided by the RFC probability of failure or duration of survival, to the Chicago clearinghouse), are adequate to the rates of withdrawal of debt during 1931, preserve systemic stability, then a far less am- and the interest rates paid on borrowed money bitious federal safety net might be desirable lead to the same conclusion: failures of banks (Calomiris, 1990). during the panic reflected the continuation of Our findings lend support to James's (1938) the same process that produced failures before account of the role of interbank cooperation in the panic. The special attributes of failing mitigating the costs of the banking crisis. The banks are distinguishable months before the limited duration and costs of contagion may panic and were reflected in stock prices, failure have reflected the cooperative intervention by the Chicago clearinghouse, which used its liq- uid assets to protect at least one solvent bank from unwarranted attack until the runs by un- 10 The Chicago Bank of Commerce, the largest bank to informed depositors subsided. Absent such co- fail during the panic and the only Chicago Loop bank to operation, the failure experience during the fail, had an estimated probability of failure of 0.00572 in the out-of-sample logit, and an estimated probability of panic of June 1932 could have been very dif- failure of 0.448 in the in-sample logit. Four panic failures ferent. Our evidence suggests, somewhat con- had lower out-of-sample estimated probabilities of failure trary to the portrayals in Friedman and than the Chicago Bank of Commerce, and seven panic Schwartz (1963) and Bernanke (1983), that failures had lower in-sample estimated probabilities of failure. The panic failures with lower estimated failure clearinghouses continued to serve an impor- probabilities (for which we have examination reports) tant function during the Great Depression, and were viewed as severely troubled banks by examiners. did not see the Fed or the RFC as ''reliev [ing] While it is not possible to determine whether the Bank of them of the responsibility of fighting runs" Commerce was solvent or insolvent during the panic using its logit or survival scores alone, we are able to say that (Bernanke, 1983 p. 260). How far can one its scores were not unusual relative to panic failures that generalize from these conclusions? Was the were perceived by examiners as troubled institutions. Chicago panic of June 1932 representative of

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 882 THE AMERICAN ECONOMIC REVIEW DECEMBER 1997

other banking panics during the Great Depres- American Economic Review, June 1991, sion? Because panics and waves of bank fail- 81(3), pp. 497-513. ure were scattered across time and location Calomiris, Charles W. and Schweikart, Larry. during the Great Depression, we believe an- "The Panic of 1857: Origins, Transmission, swering that question will require analysis of and Containment." Journal of Economic other local panics, using detailed bank-level History, December 1991, 57(4), pp. 807- data similar to those we have analyzed for the 34. Chicago panic. Defining and analyzing those Calomiris, Charles W. and Wheelock, David C. events is an important area for future research "The Failures of Large Southern Banks on the causes of bank failures during the During the Great Depression." Working pa- Depression. per, Columbia University, November 1995. Champ, Bruce; Smith, Bruce D. and Williamson, REFERENCES Stephen D. "Currency Elasticity and Bank- ing Panics: Theory and Evidence." Work- Bank and Quotation Record (of the Commercial ing paper, University of Western Ontario, and Financial Chronicle). New York: Dana, 1991. various issues. Chicago Tribune. Various issues, 1932. Benston, George J. "Interest Payments on De- Commercial and Financial Chronicle. New mand Deposits and Bank Investment Be- York: Dana, July 2, 1932, 135(1), p. 71. havior." Journal of Political Economy, Diamond, Douglas W. and Dybvig, Philip H. October 1964, 72(5), pp. 431-49. "Bank Runs, Deposit Insurance, and Li- Bernanke, Ben S. ' 'Nonmonetary Effects of the quidity." Journal of Political Economy, Financial Crisis in the Propagation of the June 1983, 97(3), pp. 401-19. Great Depression." American Economic Donaldson, R. Glenn. "Sources of Panics: Evi- Review, June 1983, 73(3), pp. 257-76. dence from the Weekly Data." Journal of Bhattacharya, Sudipto and Thakor, Anjan V. Monetary Economics, November 1992, "Contemporary Banking Theory." Journal 30(2), pp. 277-305. of Financial Intermediation, October 1993, "Financing Banking Crises: Lessons 3(1), pp. 2-50. from the Panic of 1907." Journal of Mon- Calomiris, Charles W. "Is Deposit Insurance etary Economics, February 1993, 31(1), Necessary?" Journal of Economic History, pp. 69-95. June 1990, 50(2), pp. 283-96. Esbitt, Milton.' 'Bank Portfolios and Bank Fail- "Regulation, Industrial Structure, ures During the Great Depression: Chi- and Instability in U.S. Banking: An cago." Journal of Economic History, June Historical Perspective," in Michael 1986, 46(2), pp. 455-62. Klausner and Lawrence J. White, eds., Federal Reserve Bulletin. Washington, DC: Structural change in banking. Home- Board of Governors of the Federal Reserve wood, IL: Business One Irwin, 1993, pp. System, various issues. 19-116. Friedman, Milton and Schwartz, Anna J. A Mon- .. "Is the Discount Window Necessary? etary History of the United States, 1867- A Penn Central Perspective." Review, Fed- 1960. Princeton, NJ: Princeton University eral Reserve Bank of St. Louis, May/June Press, 1963. 1994, pp. 31-55. Gorton, Gary B. and Pennacchi, George G. ' 'Fi- Calomiris, Charles W. and Gorton, Gary B. nancial Intermediation and Liquidity Crea- "The Origins of Banking Panics: Models, tion." Journal of Finance, December 1990, Facts, and Bank Regulation," in R. Glenn 45(1), pp. 49-72. Hubbard, ed., Financial markets and finan- Imbens, Guido W. "Transition Models in a cial crises. Chicago: University of Chicago Non-Stationary Environment." Review of Press, 1991, pp. 109-73. Economics and Statistics, November 1994, Calomiris, Charles W. and Kahn, Charles M. 96(4), pp. 703-20. ' 'The Role of Demandable Debt in Struc- James, F. Cyril. The growth of Chicago banks, turing Optimal Banking Arrangements." volume 11. New York: Harper, 1938.

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis VOL. 87 NO. 5 CALOMIRIS AND MASON: BANK FAILURES DURING THE DEPRESSION 883

Kaufman, George G. "Bank Contagion: A Re- Superintendent of Banking of the State of Illinois. view of the Theory and Evidence." Journal Statement showing total resources and lia- of Financial Services Research, April 1994, bilities of Illinois state banks at the close of 8(2), pp. 123-50. business. Springfield, IL: Journal Printing Mason, Joseph R. "Determinants and Effects of Company, various issues. RFC Lending to Banks During the Great De- Thies, Clifford and Gerlowski, Daniel. "Bank pression." Working paper, Office of the Capital and Bank Failure, 1921 -1932: Test- Comptroller of the Currency, November 1995. ing the White Hypothesis." Journal of Eco- Mishkin, Frederic S. "Anatomy of a Financial nomic History, December 1993, 53(4), pp. Crisis." National Bureau of Economic Re- 908-14. search (Cambridge, MA) Working Paper Wheelock, David C. "Deposit Insurance and No. 3934, December 1991. Bank Failures: New Evidence from the Office of the Comptroller of the Currency. Un- 1920s." Economic Inquiry, July 1992, published examination reports and letters 30(3), pp. 530-43. pertaining to national banks in Chicago, White, Eugene N. "A Reinterpretation of the 1931 and 1932. Banking Crisis of 1930." Journal of Eco- Annual report. Washington, DC: nomic History, March 1984, 44(1), pp. U.S. Government Printing Office, various 119-38. issues. Wicker, Elmus R. "A Reconsideration of the Reports of Condition of (Federal Reserve Member) Causes of the Banking Panic of 1930." Banks. Unpublished records, various calls. Journal of Economic History, September Saunders, Anthony J. and Wilson, Berry. "Con- 1980, 40(3), pp. 571-83. tagious Bank Runs: Evidence from the "A Reconstruction of the Gold and 1929-1933 Period." Working paper, New Banking Crises in 1931." Working paper, York University, 1994. Indiana University, October 1993. Schwartz, Anna J. "The Misuse of the Fed's Wigmore, Barry A. "Was the Bank Holiday of Discount Window." Review, Federal Re- 1933 Caused by a Run on the Dollar?" serve Bank of St. Louis, September/Octo- Journal of Economic History, September ber, 1992, 74(5), pp. 58-69. 1987, 47(3), pp. 739-56.

Digitized for FRASER http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis