Bank runs, information and contagion in the panic of 1893

Explorations in Economic History 44 (2007) 411–431 www.elsevier.com/locate/eeh

Bank runs, information and contagion in the panic of 1893 Brandon Dupont ¤ Western Washington University, Department of Economics, 315 Parks Hall, Bellingham, WA 98225, USA Received 27 December 2005 Available online 30 June 2006

Abstract Contagious bank runs, which spread to both solvent and insolvent banks, should not occur if bank-speciWc information is provided regularly to the banking public. By mitigating the information asymmetry between banks and depositors, information should restrict runs to insolvent banks. However, oYcial bank statements collected from quarterly reports to local newspapers in Kansas demonstrate that runs did become contagious in the 1893 panic even in an information-rich banking system. Important diVerences between national and non-national banks were also found, which suggests the maturity of the regulatory system may have played an important role in the panic. © 2006 Elsevier Inc. All rights reserved. Keywords: Contagion; Bank panic; Panic of 1893; Bank runs; Bank suspensions

1. Introduction While it occurred well over a century ago, the bank panic of 1893 is an important though often neglected aspect of American economic history and one that also holds important lessons for contemporary policymakers. I argue in this paper that locally contagious bank runs developed in the state of Kansas during the Panic of 1893 despite the existence of a relatively information-rich banking environment. Contagion, according to previous theoretical work, should not develop if bank-speciWc information is provided *

Fax: +1 360 650 6315. E-mail address: [email protected]

0014-4983/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.eeh.2006.05.001

412

B. Dupont / Explorations in Economic History 44 (2007) 411–431

to the public since providing this information allows depositors to discriminate between banks that are at risk of failure and those that are not.1 Weak banks may experience deposit withdrawals and may end up closing but contagious runs that spread indiscriminately from one bank to another should not develop in this environment. The analysis presented here suggests that contagious runs developed in 1893 even though bank-speciWc information was regularly provided to the public. This conclusion is supported by analyses of county- and bank-level deposit Xows and probit regressions on bank suspensions. These results have important implications for structuring policies that are designed to stop or prevent banking instability. What type of information is required in an information-rich system? Theoretically, one would like to have continuous information about speciWc bank customers, loan quality and other investments and the Wnancial standing of the bank’s principal investors. Of course, such real-time perfect information cannot be expected, particularly in 1893. Indeed, such stringent informational requirements are not required—what is required is that the bank depositors have accurate information about the bank’s general solvency just prior to the panic. Kansas depositors could obtain such general solvency information quite easily in 1893 thanks to the General Banking Law of 1891. The 1891 Kansas law required this information to be printed in a local newspaper every 3 months so depositors regularly had easy access to bank-speciWc information. As pointed out later, one of these reports was printed just days before most of the bank runs in 1893. By reducing the asymmetry of information between a bank and its depositors, this information should have prevented contagion. Act 21 of the 1891 Kansas bank law established the Kansas State Bank Commissioner as the oYcial state regulatory agency and created the information-rich system that is central to the following analysis. According to this law, the bank commissioner’s representatives were to visit every bank doing business in the state, except national banks, at least once per year for an examination of the bank’s Wnancial standing.2 The commissioner and his deputy had the power to investigate all people connected with banks when making an investigation.3 The commissioner could also call on all banks, except national banks, at any time for a report of their condition and four such reports were to be made each year. These reports were transmitted to the state commissioner’s oYce and were also required to be published in a local newspaper in the town where the bank operated. The bank commissioner would take charge of insolvent banks until a receiver was appointed and was required, in each even numbered year, to report the “names of owners or principal oYcers, the paid-up capital of each, the number of banks in the state, the name and location of each and the number and date of examinations and reports of and by each.”4

1

As explained later in the paper, the prevailing assumption throughout is that depositors only need basic but regular information about bank quality—essentially, depositors only need to know whether their bank is “weak” or “strong.” 2 The national banks were under the purview of the U.S. Comptroller of the Currency, which had similar oversight abilities. 3 A graduated fee was to be charged for these examinations ranging from $5 for banks of $5000 capital stock to $20 for banks of $50,000 capital stock and over. 4 These created the bank commissioner’s biennial public reports. In the normal course of events, 1893 would not have been among these years but because of the unique events of that year, the commissioner prepared a Special Report of 1893, which was used heavily in this research.

B. Dupont / Explorations in Economic History 44 (2007) 411–431

413

The most important part of the law for understanding the crisis of 1893 was the requirement of quarterly statements from all state and private banks to the State Bank Commissioner and the local newspapers, which mirrored the reporting requirements of the national banks that had been in place since the passage of the National Bank Act in 1863.5 In addition to making individual quarterly bank data available, the newspaper reporting requirement created a relatively information-rich banking system, which is ideal for analyzing the role of information and contagion in banking instability. As previously mentioned, contagious banks runs should not develop in information-rich environments. As explained by Park (1991), “bank failures are contagious due to the lack of bank-speciWc information. Depositors who lack Wnancial information on individual banks make withdrawal decisions based on the condition of the banking system as a whole.” This is not to say that bank balance sheets held all the relevant information about a bank’s risk exposure; however, they do contain reasonable proxies to these more detailed solvency indicators. While therefore not a perfect source of information, these oYcial bank statements did provide basic bank-speciWc information that informed depositors as to the basic risks to which their bank was exposed. 2. Data construction The data that were used to analyze deposit movements and from which the bank solvency and other variables were constructed were taken from the April/May, June/July and October reports of oYcial bank statements that were required in the 1891 bank law and in the National Bank Act.6 While the state of Kansas required quarterly reports of bank condition to be submitted to the state commissioner, these original reports have been lost. As already described, however, the 1891 law also stipulated that reports of condition were to be printed in a local newspaper in the town where the bank was located. Data for national, state and private banks in the state were collected from these newspaper reports. The Wrst step in data collection was to compile lists of all banks operating in the state in 1893 along with the town and county in which they operated. These lists were generated using the Rand McNally Banker’s Directory from January 1893 along with the State Bank Commissioner biennial report of 1892, and the Comptroller of the Currency Annual Report of 1892. Once the list was compiled, a separate list of the newspapers in the towns in which banks were located was generated from a database at the Kansas State Historical Society. The newspaper records were then examined for oYcial bank statements in each town. Since the bank panic in 1893 was of relatively short duration over the summer of that year, I examined only three of the four call dates for that year. The relevant dates, which capture the entirety of the crisis, were April 5, June 20 and October 3. The national banks, which did not report to the state commissioner but which had been making similar reports to the local newspapers since 1864, had slightly diVerent call dates of May 4, July 12 and October 10. The newspapers typically printed the oYcial bank statements from two to four weeks after the call date.

5

The law had some teeth to it. Section 15 held that any oYcer, agent or clerk of any bank that willfully made false reports faced a Wne of up to $1000 and/or imprisonment in the county jail not to exceed one year. 6 April and June for the state and private bank reports to the State Bank Commissioner and May and July for the national bank reports to the U.S. Comptroller of the Currency.

414

B. Dupont / Explorations in Economic History 44 (2007) 411–431

Data for 265 state and 144 private banks were collected representing 94.3 and 88.9 % of the total number of banks in the state, respectively. The data also include 130 national banks, or 91.5 % of all national banks in the state. Although it was possible to collect data for the vast majority of banks in the state, banks were sometimes missing from newspaper reports either because the newspapers themselves were unavailable for the necessary dates or did not report for some other reason but the dataset clearly captures the large majority of banks that were operating in the state at the time. These reports of statement were combined with call reports from October 3 that were provided in the Kansas State Bank Commissioner Special Report to form a complete picture of bank-speciWc balance sheet variables for Kansas banks from April to October 1893, which spans the period of the bank runs. Using the October 3 Special Report is particularly useful because it included a list of the banks that suspended in 1893 along with their balance sheets at the time of closure. This is important in making sure that the data do not omit banks that were closed during the crisis because the newspaper reports alone do not report on most closed banks. This amounts to an issue of timing since the newspaper reports appeared about 2–4 weeks after the oYcial call dates. For example, while one call date for reports was June 20, the newspapers did not publish the bank statements until mid-July. However, most of the banks that suspended during the panic did so in mid-July. Since they were closed at around the time that their oYcial statements would normally have been required at the newspapers, the local newspaper reports often did not print the statements for the suspended banks but these statements were recovered from the October Special Report. Therefore, the degree to which the data are censored is minimal. 3. Analysis of quarterly newspaper data The quarterly bank statements collected in this research have a number of advantages over the publicly available biennial reports of the state commissioner’s oYce. First, two of these quarterly reports were due on April 5 and June 20, which are nearly ideal dates for studying the 1893 banking crisis. The most severe bank runs in 1893 occurred in mid-July, so having detailed data from April 5 and June 20 allows us to examine the condition of the state and private banks just prior to the major banking disruptions. This is a signiWcant advantage over relying on the published biennial reports, which only allow us to observe bank conditions about 1 year prior to the panic. Second, being able to track bank deposits from April to June to October allows us to determine which banks faced bank runs. While many of the state and private banks saw large deposit reductions over this period, a few actually had deposits increase. Finally, the data on deposit changes provided a unique way to empirically identify contagion, which is an issue of considerable importance in formulating eVective policy responses to developing bank crises and to establishing mechanisms that will be more likely to prevent such crises. Nearly all of the empirical literature on banking panics takes as the object of explanation an outcome, which is the bank suspension or failure, rather than the fundamental cause, which is the bank run. Yet panics are by deWnition composed of a large number of runs on individual banks. Explaining bank closures is important but understanding the factors behind those closures—the bank runs themselves—is arguably even more important. The banks can be generally grouped into one of three categories. First, there were banks that were subjected to runs and were forced to close. There were also banks that were subjected to runs but remained open. Finally, there were banks that were not targeted by

B. Dupont / Explorations in Economic History 44 (2007) 411–431

415

Table 1 Bank type by deposit change and suspension Number of Observations

Percent of total banks of this type

Change in deposits %

Median $

Change in deposits/ assets

All banks Deposits decrease, bank non-suspending Deposits decrease, bank suspending Deposits increase, bank non-suspending

420 41 77

78.1% 7.6% 14.3%

¡35.0% ¡37.3% 41.2%

$(10,609.00) $(10,400.00) $2760.00

¡9.4% ¡12.3% 7.0%

National banks Deposits decrease, bank non-suspending Deposits decrease, bank suspending Deposits increase, bank non-suspending

102 9 19

78.5% 6.9% 14.6%

¡0.304 ¡0.368 0.212

($24,379.00) ($29,457.00) $4003.00

¡7.4% ¡11.4% 7.9%

State banks Deposits decrease, bank non-suspending Deposits decrease, bank suspending Deposits increase, bank non-suspending

203 20 41

76.9% 7.6% 15.5%

¡0.347 ¡36.30% 49.10%

($7985.00) ($9441.00) $2980.00

¡8.9% ¡12.4% 6.7%

Private banks Deposits decrease, bank non-suspending Deposits decrease, bank suspending Deposits increase, bank non-suspending

115 10 17

81.0% 7.0% 12.0%

¡39.50% ¡48.40% 44.40%

$(10,079.00) $(11,857.00) $2092.00

¡12.0% ¡13.8% 6.6%

depositors and made it through the period without facing runs—some of these actually had deposit increases. As summarized in Table 1, nearly all banks in the state experienced at least some deposit reduction.7 In the pooled dataset, 85.5% had reduced deposits from April to October 1893. Of the 265 state banks, 224 (84.5%) had deposit levels decline. A similar picture emerges for the private banks: 125 of the 144 private banks, or 87.4%, saw deposits shrink. Finally, 85.4% of national banks had reduced deposits. For most banks the measure of deposit change is percent change in individual deposits from the April to October reports, which allows us to capture the entire period of the crisis. For the banks that suspended in the summer of 1893, however, we are without reports from the June/July call dates because most of the banks that suspended were still closed at the time those regular reports were due to the state commissioner’s oYce and the local newspapers. There are a few exceptions to this.8 For the suspended banks, the deposit change variable is taken from the April call date to the date of closure, which is based on data from the 1893 Special Report. While individual dates will therefore vary in a few cases, we end up with a deposit change measure that captures movements in deposits from some point prior to the bank runs to some point after the runs whenever possible. The diVerences between deposit reductions for national banks that were forced to suspend and those that remained open are large and are statistically distinguishable at the 5% level across banks. Interestingly enough, the diVerence in deposit reductions at both state

7 Note that national banks held about 52% of all deposits in the state whereas state banks had 33 percent and private banks had almost 15%. 8 The statements from the June 20 call date for state and private banks and the July 12 call date for national banks were generally printed in local newspapers sometime in mid- to late-July. In 1893, however, most banks that closed did so in this same part of July so did not submit statements to the newspapers.

416

B. Dupont / Explorations in Economic History 44 (2007) 411–431

and private banks that remained open and those that closed is small and not statistically diVerent at either the 5% or 1% level.9 The period in which the bank runs and subsequent suspensions occurred was also the time of year during which deposits were normally withdrawn by farmers who needed cash for harvesting. As noted in previous work (Miron, 1986; Kemmerer, 1910), this currency drain has been responsible for many bank runs. Kansas banks did not typically lose deposits from July to September, at least back to the beginning of bank reports in 1891; however, bank deposit growth did usually slow in this period.10 For example, Kansas state banks actually had a nearly 12 % increase in deposits from July to October 1892 with a net addition of only four new banks. Average deposits per state bank therefore increased slightly over this period. Private banks in the state did see a drain on deposits in 1892 from March to June of about six % but then had a resurgence from June to September. Kansas national banks had a 6.5 % increase in deposits from July to September 1892 and a comparable increase in 1891. At a cursory level, the deposit changes suggest there may have been diVerent reasons for closures at national banks and those at state and private banks. The national banks seem to have been exposed to something that looks more like an information-based run on the weakest banks. By contrast, the state and private banks closed not because of targeted information-driven runs by depositors but because of the contagion that was impacting local banking markets. All that is required for contagion to develop is a possibility of failure at one institution in the area so state and private banks may have closed as they observed runs at nearby banks that they anticipated moving rapidly into their own banks. This could have happened even before severe deposit losses. These results are intriguing but not conclusive since they are based on simple comparisons of means across suspending and surviving banks. To more adequately examine the issue, we turn to an examination of deposit Xows within local banking markets in the next section. 3.1. Using deposit movements to analyze contagion in bank panics Contagion can occur as bank depositors reassess the viability of other banks when they observe either suspension or bank runs at a nearby bank. One failure, or the possibility of failure at one institution, may be thought to reveal information about other potential failures even if no actual link exists between the two institutions. Empirically testing for contagion in Wnancial and banking markets is diYcult but a number of papers have taken interesting approaches to this issue although most have been from an aggregate rather than an individual-bank level.11 Kaufman (1994) reviewed the existing evidence for contagion and cited a 1929 American Bankers Association report, which found that 30% of bank closings were followed by the closing of at least one other neighboring bank within 10 days. Friedman and Schwartz (1963) argued in favor of 9

The t-values for a diVerence in percent deposit changes across suspending and surviving banks at national, state and private banks are ¤¤¤2.793, ¤¤1.230 and ¤1.348. ¤¤¤ SigniWcant at 1% level; ¤¤ at 5% level; ¤ at 10% level. 10 It should be noted that this is based on quarterly deposit data from the State Bank Commissioner and the Comptroller of the Currency. There could be unobservable currency drains that occurred within the quarter. 11 This is important because, as Saunders and Wilson (1996) pointed out, aggregated data are unlikely to reveal whether a particular contraction in deposits was due to informed depositors.

B. Dupont / Explorations in Economic History 44 (2007) 411–431

417

contagion when they wrote, “In October 1930, the monetary character of the contraction changed dramatically ,ƒ, a contagion of fear spread among depositors, starting from agricultural areas, which had experienced the heaviest impact of bank failures in the twenties.”12 Wicker (1996), however, disputed this conclusion by arguing that the 1930 bank failures were preceded by the closure of Caldwell and Company in Nashville, which controlled the largest chain of banks in the Southern U.S. Saunders and Wilson (1996) analyzed the determinants of deposit withdrawal rates at failing banks and found evidence of contagion from 1930 to 1932 but ruled out contagion in the panics of 1929 and 1933. They also argued that in non-contagious runs, informed depositors switch deposits (usually locally) from insolvent banks to solvent banks. However, in a contagious run, uninformed depositors withdraw from both solvent and insolvent banks. Calomiris and Mason’s (1997) work on contagion during the 1930 bank runs in Chicago is one of the few that looked at the issue from a bank-speciWc level. They studied the 1932 Chicago bank panic to determine whether private bank action can eVectively stem asymmetric-information bank runs or whether deposit insurance was required. They found that most of the bank withdrawals were redeposited in riskless postal savings accounts, which is evidence against the contagion hypothesis.13 They also found that interbank cooperation helped preserve the solvent banks that were under pressure from large withdrawals. They compared bank failures both during and outside the period of the bank panic and found that failures of banks during the panic reXected the continuation of the same process that led to failures prior to the panic. It followed that failures during the panic reXected the weakness of failing banks in the face of some common asset shock rather than contagion. Consistent with the previous literature, if bank runs were information driven and not due to contagion, then we should observe withdrawals from insolvent banks that are then re-deposited in solvent institutions. Bank solvency measures should exert a large inXuence on bank deposit movements over the course of the crisis if it is driven by information and not by contagion. Evidence that most withdrawn deposits were not re-deposited in stronger institutions would work against the information-driven bank panic hypothesis and in favor of contagion. The data that were collected from the state of Kansas allow for an empirical examination of this redepositing issue. If the bank runs were contagious then we should not see signiWcant deposit movements from weak to strong banks in local markets over the course of the summer of 1893.14 Regardless of fundamental strength, depositors would have withdrawn from all banks in a contagious panic. In a non-contagious information-based event, depositors should have withdrawn only from banks that were exposed to an economic shock or that may otherwise have been unhealthy. It follows that if we are observing contagious bank runs, we should not see substantial diVerences in characteristics of suspending versus non-suspending banks. Those that suspend should look a lot like those that remain

12

Friedman and Schwarz, A Monetary History of the United States, p. 308. Postal savings accounts were created in 1910 and were designed as depositories for the savings of low-income groups. The federal government designated speciWc post oYces as postal savings oYces and interest of 2% per year was to be paid on these deposits. 14 Local markets here are considered to be within a county. This is a reasonable assumption given the nature of late 19th century banks in the western states—they developed in response to local agricultural credit needs and generally served relatively small areas. Interstate branch banking was prohibited and a number of states prohibited lending to out-of-state residents. 13

418

B. Dupont / Explorations in Economic History 44 (2007) 411–431

open in a contagious event. This is in contrast to a non-contagious event in which weak banks or those exposed to negative shocks are targeted in which case we see substantial diVerences in balance sheet characteristics between suspending and surviving institutions. 3.2. Evidence from deposit changes There were 45 counties out of 101 in the dataset where one or more banks had deposit increases. Yet in 40 of these 45 counties, the net change in deposits for all banks in the county was negative and the changes were nearly always large negative ones. The median deposit decrease at banks with falling deposits was ¡$10,609 compared to a median deposit increase of only $2750 at those where deposits increased. Even though the size of the deposit increases at the banks that had higher deposits was quite small in nearly all cases, we must still consider them as possibilities for information-based redepositing by bank customers if deposits were increasing at banks that were considerably stronger than those suVering the withdrawals. Deposits increased from April to October at only 15% of the banks in the state. Only 54% of the banks where deposits increased had a capital ratio that was better than the average for their county.15 This means that 46% of the banks that had deposits increase— potential candidates for having attracted redeposits in an information-driven panic—were actually weaker than the average bank in the same county.16 In other words, we see both deposit increases and stronger-than-average capital ratios at only about 8 % of all banks in the state. This rules out a widespread information-driven redeposit for these banks and is evidence in favor of contagion. There were therefore some cases in which deposits may have been shifted locally from weak to strong banks but these cases are clearly the exception rather than the rule.17 Moreover, the deposit increases were small in nearly every case—and were considerably smaller than the size of the deposit withdrawals at other local banks. It will also be informative to take a look at the eVects of bank strength on deposit Xows at individual banks. If we Wnd that healthier banks, measured by solvency variables and controlling for other factors, attracted deposits while deposits were removed from unhealthy banks then contagion was not the primary means by which the panic spread. This would be evidence of rational behavior based on publicly available bank-speciWc information. On the other hand, contagion would be consistent with a Wnding that deposits did not move into the healthier banks and were in fact withdrawn from both solvent and insolvent banks. In the following empirical models, percent change in bank deposits is regressed on a set of bank- and location-speciWc variables for each bank type to analyze the factors that inXuenced deposit Xows. The variables used in the deposit change regressions are described in Table 2. Included among the regressors is a dummy variable for suspension at a

15

The capital ratio is the ratio of capital, surplus and undivided proWts to total loans and serves as a simple measure of risk. 16 There were 78 banks with deposit increases but two of these are missing capital ratios so these Wgures are based on 76 banks. 17 I do not have evidence on possible interstate shifts but these are unlikely because of the localized nature of 19th century banking and the prohibition on interstate deposits in many states.

Variable

DEPCH STOCKS LOANS DUE CORRSUS DUE*CORRSUE NEIGHBOR CAP RATIO EXCAP LOGASSETS LTDEBT ASSETS/LIAB RES/DEPOSIT CTYPOPD CTYAGVALUES a b c

DeWnition

% Change in deposits (Stocks + Bonds)/Assets Loans/assets Due from other banks/assets D 1 If correspondent suspended Interaction term D 1 If another bank in county suspended Capital ratioa Excess capital D (paid-in ¡ required capital)/assets Log of total assets long-term debt/assets Total assets/(total liabilities ¡ capital) Reserves/individual deposits County population density D people/square mileb % Change in county agricultural valuesc

National

State

Private

Pooled

Mean

Std. Dev.

Mean

Std. Dev.

Mean

Std. Dev.

Mean

Std. Dev.

¡0.233 0.018 0.598 0.022 0.169 0.003 0.300 0.672 $25,891 5.359 0.057 1.516 0.197 33.955 ¡0.357

0.257 0.029 0.099 0.037 0.376 0.016 0.460 0.222 $89,065 0.281 0.084 0.260 0.109 44.147 0.379

¡0.217 0.015 0.657 0.124 0.208 0.024 0.253 0.653 $22,942 4.807 0.129 1.626 0.255 31.648 ¡0.372

0.489 0.042 0.144 0.103 0.406 0.061 0.435 0.596 $31,446 0.350 0.124 0.779 0.187 51.502 0.393

¡0.322 0.011 0.575 0.179 0.324 0.049 0.239 0.658 $10,989 4.691 0.123 1.459 0.263 24.150 ¡0.382

0.305 0.033 0.169 0.134 0.470 0.105 0.428 0.641 $14,965 0.339 0.129 0.459 0.333 15.005 0.389

¡0.249 0.015 0.621 0.114 0.229 0.026 0.261 0.659 $20,448 4.910 0.110 1.555 0.243 30.224 ¡0.371

0.400 0.037 0.146 0.116 0.421 0.071 0.439 0.543 $49,550 0.420 0.121 0.613 0.227 42.997 0.388

(Capital + Surplus + Undivided ProWts)/(Loans + Overdrafts). 100£ for regressions. Weighted average valuations for corn, oats and wheat % change from 1892–93 from Ks Agriculture Commissioner.

B. Dupont / Explorations in Economic History 44 (2007) 411–431

Table 2 Variable deWnitions and summary statistics

419

420

B. Dupont / Explorations in Economic History 44 (2007) 411–431

neighboring bank (NEIGHBOR), which is designed to indicate whether contagion spread from one bank to another within local banking markets in the 1893 panic. To examine the possible transmission of the disturbance through connections to the reserve system, the amount due from other banks as a percentage of total assets (DUE) and a dummy variable for suspension at a bank’s correspondent (CORRSUS), news of which usually reached local markets via newspaper reports, were included. Bank correspondents were identiWed from the 1892 Rand McNally Banker’s Directory to construct the correspondent suspension variable. This information was then matched with bank closure reports from the September 23, 1893 edition of Bradstreet’s magazine. Any interaction eVects are measured using the variable (DUE*CORRSUS). Other variables were designed to capture bank exposures to potential real economic shocks that may have led to runs and, ultimately, to suspensions. These include exposure to stock market Xuctuations in a variable that measures stock and bond holdings relative to total assets (STOCKS) and loans on personal collateral and real estate relative to total assets (LOANS).18 A simple measure of bank solvency is also required for trying to determine whether deposits were shifted away from insolvent banks and into solvent banks or whether contagion was a factor in the panic. Since capital adequacy is currently used to trigger supervisory action under the Prompt Corrective Action measures provided of the 1991 Federal Insurance Deposit Corporation Improvement Act (FIDCIA), it is a reasonable measure of bank solvency. The capital adequacy measured used here is the capital ratio (CAPRATIO), which is deWned as (Capital + Surplus + Undivided proWts)/Loans on Personal Collateral.19 It is also possible that depositors considered capital relative to what was required so the models include a measure of excess capital, which is measured as paid-in capital minus required capital divided by total assets, in the variable (EXCAP). A measure of the bank’s long-term debt relative to total assets (LTDEBT) is used as an indicator of the proportion of assets tied up in illiquid long-term commitments. Long-term debt is deWned as the sum of rediscounted notes, bills payable and time deposits. Bank size is controlled for by using the log of total assets (LOGASSETS) and the bank’s asset to liability ratio is also included as a control in the regressions (ASSETS/LIAB). Liquidity is captured with a measure of banks reserves relative to deposits (RES/DEPOSIT).20 County-level controls include population density, which measures the number of people per square mile in the county (POPD). The county-level population Wgures were drawn from the 1890 Census while the Kansas State Agricultural Commission’s biennial reports provided data on total area in square miles in each county. Given that banks in Kansas were primarily engaged in lending for agricultural ventures, it is also reasonable to control for diVerences in agricultural valuation changes across counties. This was done by constructing an index of agricultural values for each county (AGVALUES), which is based on dollar valuations of corn, wheat and oats that were prepared by the Kansas State Agricultural Commissioner. The index was constructed by weighting the value of production of 18

Recall that the 1863 National Bank Act prohibited lending on real estate by national banks. This is obviously a simple version of a capital ratio. However, bank supervisors historically relied on this simple capital/asset measures as a rule of thumb for general bank safety. Indeed, a paper by Estrella et al. (2000) found that a simple measure of the capital ratio similar to that used in this paper performs about as well as a more complex risk-weighted ratio. 20 I also tried related variables including a measure of bank specie relative to assets but Wnd little change. 19

B. Dupont / Explorations in Economic History 44 (2007) 411–431

421

each crop type by the share of that crop produced in each county. The result is an index that provides a simple way of capturing valuation changes for all of the crops produced on a county-level. 3.3. Regression models Three separate models of deposit changes were estimated using these data. The Wrst used ordinary least-squares (OLS) and White standard errors to correct for heteroskedasticity. The OLS residuals were found to be highly skewed so the models were then re-estimated using two separate models that are more robust to inXuential outliers: quantile regression (Koenker and Bassett, 1978) and the M-estimator proposed by Huber (1964), which uses iteratively weighted least-squares to place smaller weights on the less well-behaved observations. The results of these three estimations over the pooled sample and for each bank type are summarized in Table 3. There is a positive eVect of capital ratios on deposit changes in the robust regressions. The estimated coeYcients for the pooled data suggest that increasing the capital ratio by 10% points, say from the mean value of 65%–75%, corresponds to an increase of only between 1.9 and 2.3% points in the deposit change variable.21 This suggests that, while there was a positive eVect of capital ratios on deposit changes, deposits were not moving in a large way into the banks with higher capital ratios as we would expect to Wnd if this was an information-driven event. Such an event would suggest large Xows of withdrawn deposits into stronger banks. There are, however, considerable diVerences across bank types. National bank deposits were much more strongly aVected by higher capital ratios with a coeYcient of 0.86 in the median regression and 0.66 in the OLS and M-estimator models. The state and private banks had much smaller coeYcients. This suggests that depositors may have responded by moving deposits to banks that had higher capital ratios— but only in any signiWcant way for the national banks. This is consistent with the previous deposit change analysis results, which showed that the national banks were an exception to the contagion theory. County agricultural value changes were also statistically signiWcant for both the state and private banks but were smaller in magnitude and not statistically signiWcant for national banks. This result is intuitive given the prohibition on real estate lending for the national banks. Notice also that most of the coeYcient estimates are consistent across the regression models although the M-estimator coeYcients did produce a greater number of statistically signiWcant results. Other measures of bank-speciWc conditions had relatively minor eVects on deposit movements. For example, the asset-to-liability ratio is another measure of bank strength but the estimated coeYcient was only between 0.14 and 0.19. To generate even a relatively small increase (or, as the case may be, a smaller decrease) in individual deposits would have required increases in assets relative to liabilities that are implausibly large. These data do not suggest that depositors were relying on information about bank assets relative to liabilities to judge the soundness of particular institutions. 21 Note that the standard deviation for the capital ratios reported in Table 2 is picking up wide variation across diVerent banks but the point to consider here is that capital ratios within an individual bank did not vary much over time. Therefore, for a particular bank, the likelihood that a capital ratio would change even by 5 or 10% points in such a short period of time is unlikely even though capital ratios could easily vary by that amount across diVerent banks.

¡0.577 (0.338) 0.254 (0.135) 0.173 (0.171) 0.036 (0.039) ¡0.159 (0.244) 0.018 (0.026) 0.232 (0.032)b ¡ 0.473 (0.122)b 0.209 (0.047)b 0.211 (0.105)a 0.193 (0.015)b ¡0.112 (0.045)a ¡0.047 (0.029) 0.151 (0.032)b 0.129 (0.058)a 0.086 (0.059) ¡1.881 (0.278)b 511 0.12

Median

¡0.397 (0.331) 0.338 (0.137)a 0.401 (0.168)a 0.060 (0.037) ¡ 0.317 (0.228) 0.030 (0.025) 0.194 (0.031)b ¡0.360 (0.122)b 0.165 (0.046)b 0.179 (0.099) 0.190 (0.026)b ¡0.075 (0.088) ¡ 0.034 (0.028) 0.144 (0.030)b 0.051 (0.056) 0.017 (0.057) ¡1.693 (0.280)b 510

Huber M ¡0.620 (0.839) 0.994 (0.377)b 0.680 (0.653) 0.007 (0.072) 1.596 (1.563) 0.043 (0.048) 0.658 (0.227)b ¡0.708 (0.397) 0.223 (0.162) 0.093 (0.261) ¡0.214 (0.175) 0.046 (0.221) 0.085 (0.064) 0.041 (0.062)

¡2.183(1.056)a 119

¡3.068(0.824)b 119 0.14

¡1.884 (1.066) 119 0.20

Huber M

0.198 (0.701) 1.287 (0.342)b 0.557 (0.597) 0.021 (0.061) 1.447 (0.876) 0.058 (0.043) 0.858 (0.208)b ¡0.971 (0.323)b 0.368 (0.127)b 0.039 (0.223) ¡0.344(0.149)a ¡0.135 (0.199) 0.071 (0.031)a ¡0.026 (0.056)

Median

¡1.461 (0.785) 1.159 (0.347)b 1.022 (0.572) 0.124 (0.117) 0.334 (1.301) 0.107 (0.063) 0.667 (0.240)b ¡0.820 (0.374)a 0.161 (0.168) 0.053 (0.314) ¡0.238 (0.206) ¡0.081 (0.271) 0.108 (0.046)a 0.062 (0.062)

OLS

¡0.696 (0.421) 0.231 (0.202) 0.236 (0.297) 0.060 (0.056) ¡0.506 (0.281) 0.021 (0.029) 0.203 (0.041)b ¡ 0.345 (0.215) 0.108 (0.064) 0.157 (0.118) 0.138 (0.042)b 0.171 (0.244) ¡0.034 (0.029) 0.149 (0.035)b 0.072 (0.079) ¡0.021 (0.075) ¡1.258 (0.424)b 511 0.20

OLS

Robust standard errors in parentheses. a SigniWcant at 5%. b SigniWcant at 1%.

STOCKS LOANS DUE CORRSUS DUE*CORRSUS NEIGHBOR CAPRATIO EXCAP LOGASSETS LTDEBT ASSETS/LIAB RES/DEPOSIT CTYPOPD CTYAGVALUES STATE PRIVATE CONSTANT Observations R2

National

Pooled

Deposit change regressions—dependent variable is percent change in deposits

Table 3

¡0.961 (0.455)a 255 0.26

¡0.383 (0.657) 0.099 (0.340) ¡0.168 (0.425) ¡0.062 (0.081) ¡0.033 (0.527) 0.003 (0.046) 0.218 (0.043)b ¡0.591 (0.368) 0.081 (0.079) 0.129 (0.172) 0.142 (0.036)b 0.565 (0.679) ¡0.033 (0.025) 0.141 (0.053)b

OLS

State

¡1.513 (0.382)b 255 0.15

¡0.420 (0.521) 0.222 (0.263) ¡0.005 (0.328) ¡0.059 (0.080) 0.162 (0.537) 0.039 (0.044) 0.256 (0.028)b ¡0.443 (0.175)a 0.171 (0.067)a 0.228 (0.162) 0.193 (0.016)b ¡0.235 (0.141) ¡0.038 (0.036) 0.158 (0.052)b

Median

¡1.276 (0.330)b 254

¡0.279 (0.439) 0.217 (0.222) 0.189 (0.278) ¡0.021 (0.064) ¡0.048 (0.438) 0.057 (0.035) 0.229 (0.040)b ¡0.403 (0.157)a 0.122 (0.058)a 0.184 (0.131) 0.203 (0.029)b ¡0.352 (0.120)b ¡0.059 (0.035) 0.172 (0.042)b

Huber M

¡1.384(0.674)a 137 0.30

¡0.875 (0.551) 0.471 (0.272) 0.571 (0.270)a 0.154 (0.084) ¡0.738 (0.344)a ¡ 0.047 (0.053) 0.172 (0.047)b 0.088 (0.360) 0.143 (0.122) 0.025 (0.215) 0.020 (0.159) 0.140 (0.177) ¡0.257 (0.192) 0.261 (0.070)b

OLS

Private

¡0.985 (0.971) 137 0.19

¡0.621 (1.068) 0.252 (0.360) 0.342 (0.418) 0.162 (0.116) ¡0.502 (0.533) ¡0.046 (0.081) 0.135 (0.070) 0.183 (0.510) 0.053 (0.176) 0.124 (0.307) 0.081 (0.226) 0.000 (0.237) ¡0.033 (0.296) 0.134 (0.114)

Median

¡1.885 (0.718)b 136

¡0.397 (0.754) 0.373 (0.237) 0.555 (0.283) 0.140 (0.073) ¡0.384 (0.341) ¡0.027 (0.049) 0.090 (0.062) ¡0.186 (0.325) 0.158 (0.116) 0.181 (0.191) 0.312 (0.172) 0.230 (0.193) ¡0.025 (0.180) 0.166 (0.072)a

Huber M

422 B. Dupont / Explorations in Economic History 44 (2007) 411–431

B. Dupont / Explorations in Economic History 44 (2007) 411–431

423

Bank size, as measured by the logarithm of total assets (LOGASSETS) exerted a positive eVect on deposit changes. An increase in the logarithm of total assets from, for example, 4.8–4.9 (an increase in total assets of about 25.9% from $63,096 to $79,433)22, other things constant, increased the deposit change variable by between 1.0 and 2.0% points in the pooled data. Thus, the eVect was rather small—a relatively large increase in assets would have been required to generate a small improvement in deposits. Bank reserves as a share of total deposits (RES/DEPOSIT) is a measure of the bank’s ability to meet withdrawals so lower reserves relative to deposits may have been an indication that the bank would be unable to meet the demands of all its depositors. The robust regressions suggest that reserve ratios played practically no role in deposit movements for national and private banks so depositors were not running only on banks that were known to have liquidity problems. Note that most of these coeYcients were statistically insigniWcant. Banks that were more highly exposed to real economic shocks should be more susceptible to failure and therefore more likely to see large deposit withdrawals.23 This is measured in the regressions with the bank’s exposure to stock market shocks (STOCKS) and the shocks to loan collateral and real estate with the LOANS variable. While most banks had relatively low exposure to stocks—the mean value for stocks as a fraction of total assets is only 1.5%—there were a few cases in which banks had considerably higher exposure. There was somewhat less variation in the stock exposure for the national banks but a higher overall average value of 1.8% (but a maximum value of only 13.1%). The state banks averaged 1.5% with considerable variation of up to 27.1%. Private banks had the smallest exposure to stocks, averaging only 1.1% of total assets. Exposure to stock market Xuctuations is negatively related to the deposit change measure in all three regression models and across all bank types. The eVect of stock exposure on deposit changes was most pronounced for the national banks. For the national banks, a 1% point increase in stocks as a percentage of assets would have driven the deposit change variable lower by over 1% point in the OLS model but the coeYcients vary considerably in the robust models. While the estimated eVects of stock exposures could have been large, the role of stocks in the bank runs seems somewhat limited because of the low balance sheet exposures for most banks. It may, however, have played a role at some more highly exposed banks. A comparable increase in stock exposure for the state and private banks would have generated smaller reductions in deposits than at the national banks but the eVects were still quite large. The second possible exposure to negative shocks for the banks was through their direct loans on personal collateral and real estate. If underlying collateral values were negatively aVected by collapsing land or crop values, for example, then banks that had a higher ratio of loans to assets (LOANS) may have been in trouble. The state banks had the highest percentage of assets in loans on personal collateral at 65.7 % compared to 59.8% for the national banks and 57.5% for private banks. The banks that had higher fractions of total assets dedicated to loans on personal collateral were also more likely to have less severe deposit reductions in the 1893 crisis. This Wnding is consistent across the three regressions and all bank types although the magnitude is larger for national banks. One possibility is that banks that were more active lenders also had a larger number of connections with the 22

This uses log base 10 so the antilog of log(4.8) D 63,096 and the antilog of log(4.9) D 79,433. This is based on the asymmetric information theory of bank panics as described in Calomiris and Gorton (1991). 23

424

B. Dupont / Explorations in Economic History 44 (2007) 411–431

depositor community—those who obtained loans with a local bank generally held deposits with that same bank. However, this does not imply that those depositors would have been increasing their deposits or that they would have pulled out smaller amounts of deposits. Another possibility is that for depositors who lived in towns with only one or two banks, having debt to the banks creates a redemption value for deposits which would reduce the depositor’s desire to convert deposits into currency.24 It is possible that disturbances were transmitted through the pyramid reserve system that characterized 19th century banking.25 Connections to the reserve system are measured with a variable for the fraction of assets that were due from other banks (DUE) and a separate dummy variable that measured the eVects of suspension at a bank’s correspondent (CORRSUS). Correspondents were other banks at which commercial banks frequently held part of their deposits so suspensions at the correspondent bank could have caused either actual or perceived disruptions to bank liquidity. The estimated coeYcients across all three regression models indicate only small and mostly statistically insigniWcant eVects from correspondent suspensions. The amount due from other banks as a fraction of total assets (DUE) had generally larger eVects but most estimated coeYcients for this variable are positive and statistically insigniWcant. Runs do not seem to have materialized only at the banks that were more closely tied to the reserve system. The interaction eVects for amount due from correspondents with correspondent suspension (DUE*CORRSUS) had large coeYcients at the national banks but they also carry high standard errors so are not very precisely measured. Exploring the possibility of locally contagious bank runs in the 1893 panic is central to the analysis and one way to capture contagion is with a dummy variable for suspension at another bank in the same county (NEIGHBOR). As noted earlier, contagious bank runs should be characterized by withdrawals from all banks and not just from those weaker banks that are more likely to fail. Therefore, if contagion was at work in the bank panic we would expect to Wnd little evidence of re-depositing at other banks. In other words, we would not expect to see deposits increasing as a result of suspensions at neighboring banks. Indeed, we see little evidence that this happened. The estimated coeYcients for national banks are positive across all three models but are quite small, particularly in the robust regressions. The extent to which deposits increased in response to suspensions at neighboring banks was therefore positive but very small for national banks. The state and private banks had similarly small coeYcients although the coeYcients for private banks were all negative, but close to zero.26 One important Wnding of this analysis is that bank type matters. Previous research, which often focuses on national banks and extrapolates from that to all bank types, may have taken too narrow a focus. More importantly, these deposit change regressions suggest 24

Thanks to an anonymous referee for pointing out this possibility. The National Bank Act of 1863 created the reserve pyramid system in which national banks were required to hold minimum reserves against their deposits and these reserve requirements varied according to the domicile of the bank. New York banks, which were designated as central reserve city banks, were required to hold all their required reserves of 25% against deposits in their vaults as lawful money. After 1887, banks in the reserve cities of Chicago and St. Louis were also brought under this 25% reserve requirement. Banks in other cities were allowed to carry half of their 25% reserves with other banks as interbank balances. The remaining banks were allowed to keep three-Wfths of their 15% required reserves as balances in other banks. 26 A Variance InXation Test conWrmed that multicollinearity is not a major problem in the speciWcations reported. 25

B. Dupont / Explorations in Economic History 44 (2007) 411–431

425

that measures of individual bank strength such as the capital ratio and asset-to-liability ratios did not have a large inXuence over deposit movements with the exception of the national banks. Capital ratios did have a larger positive eVect on national banks, which again suggests that they were somewhat diVerent. If there is evidence here against the contagion thesis, it is within the national bank system. Why would evidence for contagion only be found for the non-national banks in the sample? One possibility, though a diYcult one to empirically verify, is that information was better diVused about the national banks. The national banks had been publishing their oYcial statements of condition to local newspapers since 1864, as required in Section 34 of the National Bank Act. By contrast, the state and private banks had only begun reporting oYcial statements in late 1891. Knowledge about the health of the older more well-established national banks that had been reporting publicly for nearly three decades at the time of the panic may have been more readily accessible. This suggests that the maturity of the regulatory regime may play an important role in contagion eVects during banking crises. Another possibility is that depositors were diVerentiating between national and nonnational banks but not between state and private banks. As Kaufman (1994) explained, if depositors were unable to diVerentiate among banks (for example, between state and private banks) they are unable to tell whether or not Wnancial problems at one bank apply to their own. Given diVerentiation along the lines of national versus non-national, they may have been able to make that distinction thus explaining why national banks stand apart. 4. Bank suspensions in the 1893 panic Ultimately, the withdrawals led to closures at some banks. In total, Kansas experienced 41 closures, most of which occurred in July 1893. Fig. 1 indicates the location of the bank suspensions across the state in 1893. As was previously established, contagious bank runs should not distinguish between strong and weak banks. Therefore, evidence supporting contagion would be found if the banks that were forced to close were not clearly weaker

Fig. 1. Location of suspended banks in Kansas.

426

B. Dupont / Explorations in Economic History 44 (2007) 411–431

Table 4 Surviving versus suspending bank characteristics Paid-in capital

Assets/ liabilities

All banks(Pooled) Surviving Mean Standard deviation Number of observations

$38,581 $59,425 496

1.518 0.05 496

0.624 0.146 496

0.108 0.119 496

Suspending Mean Standard deviation Number of observations

$43,215 $52,710 41

1.559 0.028 41

0.585 0.148 41

¡ 0.484

¡ 0.410

1.633

Test of diVerences in means (t-statistics)

Loans/ assets

LT debt/ assets

Capital ratio

Individual deposits

% Change in deposits

0.657 0.554 496

$55,500 $83,603 496

¡23.84% 1.84% 496

0.138 0.134 41

0.680 0.404 41

$58,859 $90,556 41

¡37.30% 23.50% 41

¡ 1.545

¡ 0.256

¡ 0.246

2.077a

Statistically signiWcant at a1% level.

than those that remained open. On the other hand, information-based runs would be suggested if we observe that the suspending banks were materially weaker than those that survived. In this case, depositors would have based withdrawals on fundamental bank weakness. While there were diVerences between the suspending and surviving banks, those diVerences were not statistically diVerent as reXected in t-tests for diVerences in means that are summarized in Table 4.27 The suspending banks were actually larger in terms of deposits and capital and, in fact, had slightly higher capital ratios. Long-term debt relative to total assets was higher at the banks that suspended although the diVerence is not statistically signiWcant. On this basis, it does not appear that depositors were using the information they had to target identiWably weaker or smaller banks. While not reXected in this table, location also appears to have been important and the location-speciWc shocks that aVected banks in those local markets may have prompted bank runs in the uncertain environment generated by the business closures and broader macroeconomic instability of the 1890s that has been documented elsewhere (Mishkin, 1991; Calomiris and Gorton, 1991). Another possible source of disturbance is through connections to the stock market and the suspending banks had higher proportions of assets in stocks and bonds, although none had very high overall exposures.28 These Wgures are based on the larger pooled sample, which could be obscuring important diVerences across bank type. DiVerences between national and non-national banks have already been established, so Table 5 summarizes results of the same tests by bank type. There are again noticeable diVerences between national and non-national banks. While capital ratios were higher at surviving national banks they are not statistically diVerent from those at suspending banks; however, most of the other measures analyzed were diVerent at suspending institutions. Long-term debt relative to total assets was nearly twice as

27

Wilcoxon–Mann–Whitney nonparametric tests yield similar results. There is also the possibility of transmission of stock market disturbances to the country banks via the reserve system and the call loan market in New York. 28

B. Dupont / Explorations in Economic History 44 (2007) 411–431

427

Table 5 Surviving versus suspending bank characteristics by bank type Paid-in capital

Assets/ Loans/ liabilities assets

LT debt/ Capital assets ratio

Individual deposits

% Change in deposits

National banks Surviving Mean Standard deviation Number of observations

$84,398 $8824 121

1.505 0.023 121

0.594 0.009 121

0.054 0.007 121

0.673 0.021 121

$118,943 $11,973 121

¡22.30% 2.33% 121

Suspending Mean Standard deviation Number of observations

$103,333 $26,101 9

1.674 0.105 9

0.657 0.029 9

0.101 0.042 9

0.663 0.057 9

$130,074 $48,280 9

¡36.75% 7.91% 9

0.133

¡0.243

1.638a

t-Statistics for diVerences in means ¡0.571

¡1.908a

¡1.857a ¡1.619a

State banks Surviving Mean Standard deviation Number of observations

$28,420 $2030 244

1.636 0.052 244

0.663 0.009 244

0.129 0.008 244

0.656 0.039 244

$37,319 $3270 244

¡20.64% 3.23% 244

Suspending Mean Standard deviation Number of observations

$24,443 $4766 21

1.511 0.076 21

0.585 0.03 21

0.140 0.031 21

0.613 0.068 21

$43,111 $15,096 21

¡34.17% 4.79% 21

0.709

2.39b

¡0.420

0.313

¡0.484

t-Statistics for diVerences in means 0.563

1.218

Private banks Surviving Mean Standard deviation Number of observations

$15,189 $1138 131

1.403 0.075 131

0.580 0.015 131

0.119 0.011 131

0.645 0.056 131

$30,764 $2378 131

¡31.21% 2.68% 131

Suspending Mean Standard deviation Number of observations t-Statistics for diVerences in means

$29,864 $9842 11 ¡3.059b

1.463 0.041 11 ¡0.418

0.527 0.057 11 0.998

0.165 0.040 11 ¡1.120

0.820 0.192 11 ¡0.872

$30,655 $6660 11 0.013

¡43.71% 8.10% 11 1.310

Statistically signiWcant at a5% level; b1% level.

high at the national banks that suspended. Loans as a fraction of assets were also markedly higher at suspending national banks but not at state and private banks, which suggests that exposure to shocks that aVected collateral values could have played a role. It is striking that the state and private banks do not show statistically signiWcant diVerences in most of these bank-speciWc measures including percent change in deposits. While the suspending banks at all bank types had more severe deposit reductions, the means were not statistically diVerent at non-suspending banks. The number of suspending banks, when broken down by type, was quite small but this still supports the notion that contagion was a factor—particularly in the state and private bank runs. The information-based theory holds that depositors with bank-speciWc information should distinguish between healthy and unhealthy banks and run on the latter. Therefore, the banks that suspend in a bank panic should be those that were already on shaky ground; yet, in the data analyzed here, it

428

B. Dupont / Explorations in Economic History 44 (2007) 411–431

is clear that this was not the case. Depositors were not withdrawing from and closing only banks that were already fundamentally weak. 4.1. Explaining bank suspensions: A probit analysis Exploring the causes of bank suspensions in a multivariate setting is desirable so that we can analyze the eVects of the variables in question while holding other factors constant. The probability of suspension was therefore analyzed in a probit regression framework using data from the April and May call reports.29 In these models, the dummy variable for bank suspension (SUSPEND), which takes a value of 1 if the bank suspended and 0 otherwise, is regressed on the set of variables described earlier including measures of connection to the reserve system, exposure to negative shocks and bank-and location-speciWc factors. The analysis on deposit shifts within local banking markets suggested that in the large majority of cases, there was evidence suggesting contagion was a mechanism by which the bank runs were propagated. Here, we address this issue using the dummy variable that was constructed to identify cases in which neighboring banks, deWned as other banks in the same county, suspended (NEIGHBOR). The probit regressions therefore allow for a determination of whether there was an impact of a neighbor’s suspension on the probability of suspension at a given bank. The models described in Table 6 allow us to measure the marginal eVect of a neighboring suspension on this probability while controlling for other bank- and location-speciWc factors. There are four separate models reported in Table 6 over the pooled sample. Models (1) and (2) diVer only in that Model (1) is run without the NEIGHBOR variable while the model summarized in column (2) includes that variable. Note that NEIGHBOR enters the model with a relatively large and statistically signiWcant positive coeYcient. Model (2), which includes the NEIGHBOR variable, also has a considerably better overall Wt. The coeYcient in Model (2) suggests that having a neighboring bank suspend increases the probability of suspension by 0.077 even while holding other factors constant. The previous analysis consistently showed diVerences in terms of contagion for national banks, so Model (3) includes interactions for state and private banks with the dummy variable NEIGHBOR that will pick up separate contagion eVects for these bank types. Model (4) is identical to Model (3) except that it combines the state and private bank dummies into a single dummy variable called NONNATIONAL. This model also includes an interaction term for NONNATIONAL and NEIGHBOR. Regressions (3) and (4) support the previous analysis, which suggested that contagion was not a factor for national banks but was for state and private banks since the NEIGHBOR coeYcient drops to near zero when the STATE and PRIVATE dummies are interacted with NEIGHBOR. There is some evidence that bank exposure to negative shocks also increased the probability of suspension. For example, the measure of stocks relative to total assets is rather large and positive (but not statistically signiWcant), which supports the earlier results on this variable from the deposit change models.

29

Reported coeYcients are the marginal eVects.

B. Dupont / Explorations in Economic History 44 (2007) 411–431

429

Table 6 Probit regressions on probability of suspension

STOCKS LOANS DUE CORRSUS DUE*CORRSUS

(1)

(2)

(3)

(4)

0.189 (0.185) ¡0.109 (0.074) 0.027 (0.103) 0.068 (0.041) 0.000 (0.131)

0.008 (0.025) 0.098 (0.106) ¡0.047 (0.038) 0.171 (0.064)b ¡0.061 (0.050) ¡0.092 (0.075) 0.030 (0.014)a ¡0.053 (0.020)b

0.104 (0.139) ¡0.068 (0.061) 0.023 (0.084) 0.039 (0.029) 0.038 (0.101) 0.077 (0.025)b 0.016 (0.018) 0.056 (0.089) ¡0.034 (0.032) 0.134 (0.056)a ¡0.050 (0.042) ¡0.093 (0.061) 0.016 (0.011) ¡0.041 (0.019)a

0.123 (0.136) ¡0.064 (0.059) 0.029 (0.085) 0.045 (0.030) 0.026 (0.100) 0.009 (0.028) 0.016 (0.018) 0.055 (0.086) ¡0.035 (0.032) 0.131 (0.055)a ¡0.048 (0.040) ¡0.093 (0.060) 0.017 (0.011) ¡0.041 (0.019)a

¡0.020 (0.050) ¡0.026 (0.037)

¡0.010 (0.041) ¡0.015 (0.032)

0.114 (0.139) ¡0.068 (0.062) 0.029 (0.087) 0.046 (0.031) 0.026 (0.101) 0.009 (0.029) 0.016 (0.018) 0.049 (0.086) ¡0.034 (0.032) 0.131 (0.056)a ¡0.049 (0.040) ¡ 0.097 (0.061) 0.015 (0.011) ¡ 0.041 (0.019)a 0.093 (0.087) 0.100 (0.110) ¡0.036 (0.048) ¡0.032 (0.026)

NEIGHBOR CAPRATIO EXCAP LOGASSETS LTDEBT ASSETS/LIAB RES/DEPOSIT CTYPOPD CTYAGVALUES STATE*NEIGHBOR PRIVATE*NEIGHBOR STATE PRIVATE NONNATIONAL NONNATIONAL*NEIGHBOR Observations Pseudo R2

511 0.17

511 0.24

511 0.25

¡0.057 (0.085) 0.085 (0.073) 511 0.25

Standard errors in parentheses. Note: Reported coeYcients are marginal eVects. a SigniWcant at 5%. b SigniWcant at 1%.

Unlike in the deposit change regressions, the probit models suggest some role for correspondent suspensions, which supports the notion that connections to the reserve system may have transmitted shocks to the country banks. Suspension at a correspondent bank

430

B. Dupont / Explorations in Economic History 44 (2007) 411–431

increased the probability of suspension in the sample of Kansas banks by 0.04, other things constant.30 Contagion seems to have been a factor not only locally but also long-distance via the correspondent system. Long-term debt levels also appear to have been important which may have been an informational element used by depositors to target certain banks but is more likely a function of the rediscounts that comprise a part of the debt measure used here. Banks that were in Wnancial diYculty often resorted to rediscounting notes with their correspondent banks so that higher redis count levels are associated with higher suspension probabilities. 5. Conclusion This paper began with a straightforward objective: to determine if contagion characterized the bank runs in Kansas during the 1893 panic. The state of Kansas is of particular interest in understanding the issue of bank run contagion because of its transparent and information rich banking system in the late 19th century. The main story to emerge is that contagious bank runs developed at state and private banks in 1893 despite the availability of bank-speciWc information to the public. Theoretically, contagion should not be a problem in such banking environments because information allows bank customers to target only the banks that are weak and/or exposed to negative economic shocks. Park (1991, p. 273) summarized the main conclusion of the information-driven hypothesis with respect to contagious bank runs this way: “ƒthe fundamental problem is the lack of bank-speciWc information, ƒ, depositors who are informed of the Wnancial strength of individual banks would make withdrawal decisions based primarily on the soundness of the bank of their concernƒa better informed public is less likely to panic.” On the contrary, the research presented here suggests that a betterinformed public is not necessarily less likely to panic. Reducing the information asymmetry between banks and the banking public may be a necessary condition for eliminating this type of instability in bank markets but it clearly is not suYcient. Policies that focus only on transparency and information availability alone may be inadequate. This research has also demonstrated that bank type was an important factor, possibly because of diVerences in the maturity of the national and state regulatory systems. There is evidence that national banks were subjected to information-driven runs and were not generally part of a contagion-driven event. On the other hand, the state and private banks were not targeted by information-driven runs but experienced runs that grew out of a contagion of fear among depositors. One possible reason for this is that the banking public had more knowledge about the national banks, which had been reporting statements of condition to the local newspapers since the 1863 National Bank Act whereas the state and private banks only began reporting to the newspapers in 1891. Whatever the reason, it is clear that there were important distinctions between national and non-national banks and other research that is conducted on bank runs of this period should carefully consider these distinctions before drawing conclusions about the banking system as a whole based only on an analysis of the national banks.

30

While not reported here, I have found that this eVect does not vary much across bank types.

B. Dupont / Explorations in Economic History 44 (2007) 411–431

431

Acknowledgments The author thanks Joshua Rosenbloom and Tom Weiss for valuable comments on the paper. The author also thanks the Kansas State Historical Society for assistance with data collection. References Calomiris, C.W., Gorton, G., 1991. The Origins of Banking Panics: Models, Facts, and Bank Regulation. In: Hubbard, R.G. (Ed.), Financial Markets and Financial Crises. University of Chicago Press, Chicago, pp. 109–173. Calomiris, C.W., Mason, J.R., 1997. Contagion and bank failures during the Great Depression: the Chicago banking panic of june 1932. American Economic Review 87, 863–883. Estrella, A., Park, S., Peristiani, S., 2000. Capital ratios as predictors of bank failure. Federal Reserve Bank of New York Economic Policy Review, 33–52. Friedman, M., Schwartz, A.J., 1963. A Monetary History of the United States, 1867–1960. Princeton University Press, Princeton. Huber, P.J., 1964. Robust estimation of a location parameter. Annals of Mathematical Statistics 35, 73–101. Kaufman, G.G., 1994. Bank contagion: a review of the theory and evidence. Journal of Financial Services Research 8, 123–150. Kemmerer, E.W., 1910. Seasonal Variations in the Relative Demand for Money and Capital in the United States. National Monetary Commission, S.Doc.588, 61st Congress, 2nd session. Koenker, R., Bassett, G., 1978. Regression quantiles. Econometrica 46, 33–50. Miron, J., 1986. Financial panics, the seasonality of the nominal interest rate, and the founding of the fed. The American Economic Review 76, 125–140. Mishkin, F.S., 1991. Asymmetric Information and Financial Crises: A Historical Perspective. In: Hubbard, R.G. (Ed.), Financial Markets and Financial Crises. University of Chicago Press, Chicago, pp. 69–108. Park, S., 1991. Bank failure contagion in historical perspective. Journal of Monetary Economics 28, 271–286. Rand McNally & Company, 1893. The Bankers’ Directory and List of Bank Attorneys. Rand McNally & Company, Chicago. Saunders, A., Wilson, B., 1996. Contagious bank runs: evidence from the 1929–1933 period. Journal of Financial Intermediation 5, 409–423. U.S. Comptroller of the Currency, various years. Annual Report. U.S. Government Printing OYce, Washington D.C. Wicker, E., 1996. Banking Panics of the Great Depression. Cambridge University Press, Cambridge.