Why do contagion effects vary among bank failures?

Journal of Banking & Finance 25 (2001) 657±680 www.elsevier.com/locate/econbase

Why do contagion e€ects vary among bank failures? Aigbe Akhigbe a, Je€ Madura

b,*

a b

Frederick W. Moyer Chair in Finance, The University of Akron, Akron, OH 44325, USA Department of Finance, College of Business, Florida Atlantic University, 777 Glades Road, Boca Raton, FL 33431, USA Received 25 August 1998; accepted 13 January 2000

Abstract Many of the previous studies on contagion e€ects in the banking industry focused on the failure of a large bank to determine whether the adverse e€ects spread to other banks. Yet, little is known whether other publicized bank failures cause contagion effects, and why the e€ects may vary among bank failures. Given the changes in the banking environment over time, contagion e€ects could be conditioned on the characteristics of the failing bank and of the banking environment at that time. We assess 99 publicized bank failures over the 1980±1996 period, and ®nd that contagion e€ects exist in general for the surviving rivals of the failed bank. The degree of contagion e€ects varies over time (among bank failures), and is stronger when the failed bank is a multibank holding company, when the failed bank is publicly held, when the failed bank is relatively large, when the rivals are relatively small, and when the rivals have relatively low capital levels. The contagion e€ects are less pronounced in the period following the passage of FIRREA. Furthermore, the total risk-shifts of surviving rival banks in response to the announcement of a failed bank are inversely related to their capital level, and total risk-shifts of rival banks are less pronounced for failures occurring just after the passage of FIRREA. The results suggest that a bankÕs exposure to possible contagion e€ects due to a bank failure can be partially controlled by a bankÕs managerial policies and by regulatory policies. Ó 2001 Elsevier Science B.V. All rights reserved.

*

Corresponding author. Tel.: +1-561-368-2260. E-mail address: [email protected] (J. Madura).

0378-4266/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 8 - 4 2 6 6 ( 0 0 ) 0 0 0 9 2 - 3

658

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

JEL classi®cation: G21; G14 Keywords: Bank failures; Contagion e€ects; Total risk-shifts

1. Introduction The term contagion e€ect has been described and interpreted in di€erent ways within the banking industry (see Aharony and Swary, 1983, 1996). It can refer speci®cally to fear of a bank run (see Diamond and Dybvig, 1983). Conversely, it can refer more generally to any transmission of information across banks, as a given amount of information pertaining to one bank may be contagious to other banks that may be exposed in a similar manner as the one bank. The general de®nition broadens the focus of contagion to any source of information that can a€ect bank values. Using the more general de®nition, any transmission of adverse e€ects can be detected by measuring the valuation e€ects of banks in response to an event pertaining to a single bank. The valuation e€ects could be attributed to the fear of a bank run (Ôbank runÕ e€ect) or to information about the bankÕs asset quality that is revealed by an event (informational e€ect). We use the more general de®nition to explain the variation in contagion e€ects of publicized bank failures in the US during the 1980±1996 period. While most of these publicized bank failures are unlikely to cause bank runs, they may elicit some informational e€ects that result in the revaluation of shares of other banks. Since the type of information that is transmitted may vary among bank failures, the assessment of the distribution of contagion e€ects may determine how the contagion e€ects are conditioned upon the type of information transmitted. By testing the sources of contagion e€ects in the banking industry, this study o€ers useful inferences for the banks that are attempting to insulate themselves from external shocks, and bank regulators that are attempting to protect the banking system. Research on bank failures has determined that failures of large well-known banks can cause contagion e€ects. However, previous implicit too-big-to-fail policies of regulators limited the number of large bank failures. Furthermore, the few large bank failures that have occurred do not o€er valuable inferences about whether other bank failures cause any contagion e€ects, the degree to which those e€ects are spread, or why the e€ects vary among failures. Our objective is to ®ll this research void by assessing the variation in contagion e€ects among bank failures. By focusing on this issue, we are able to address the following questions. First, do publicized bank failures cause contagion e€ects? Second, do the contagion e€ects vary among the bank failures that have occurred over time? Third, what factors in¯uence the degree of contagion

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

659

e€ects? Fourth, what factors in¯uence the degree of risk-shifts among the surviving banks in response to a bank failure? Our analysis of 99 publicized bank failures over the 1980±1996 suggests that publicized bank failures generally produce contagion e€ects on the surviving rivals of the failed bank. Contagion e€ects vary among bank failures, and are driven by information content pertaining to the failing bank, the characteristics of the banks considered being rivals, and bank regulation. In addition, the total risk-shifts of the surviving rival banks are conditioned on the capital levels of rivals, and the regulatory environment. The paper is organized as follows: Section 2 summarizes the literature on contagion e€ects. Section 3 describes the sample selection and methodology. Section 4 presents the empirical results. Finally, Section 5 o€ers conclusions and implications. 2. Review of literature There is a strong evidence that contagion e€ects can occur within an industry. Lang and Stulz (1992) found that bankruptcy announcements had negative e€ects on rivals in some industries. Fenn and Cole (1994) determined that contagion e€ects resulted from failures of life insurance companies. Some studies, such as those by Pettway (1976), Lamy and Thompson (1986), and Swary (1986), focused on an individual bank failure and determined negative valuation e€ects of other bank rivals as a result. Several other studies o€er evidence that information about a single bank can be transmitted throughout the banking industry. Cornell and Shapiro (1986); Bruner and Simms (1987), and others have documented contagion e€ects of the international debt crisis. Gay et al. (1991) found that bank failures in Hong Kong cause a decline in stock returns of rival banks in Hong Kong. Aharony and Swary (1996) assessed southwestern banks in response to ®ve large bank failures and found evidence of contagion e€ects. Docking et al. (1997) found that loan-loss reserve announcements have negative e€ects on rival banks. To the extent that the failures of banks signal new information about conditions in the banking industry, contagion e€ects may not only result from the failures of the largest banks, but also from any publicized bank failure. Since information about loan performance in the banking industry is limited, investors may use the information content of publicized bank failures when valuing shares of other banks. Our primary objective is to explain the cross-sectional variation in contagion e€ects among bank failures. While it is clear that some bank failures cause contagion e€ects, it is not clear why contagion e€ects have been more pronounced in some periods than others. To the extent that conditions in the

660

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

banking industry have changed, the disparate contagion e€ects could be attributed to the disparate banking industry conditions at the time of the failures. For example, the regulatory environment can in¯uence the transparency of a bankÕs ®nancial conditions. When regulators are slow to close weak banks, the market may fear that many existing banks are distressed, but continue their operations until their ®nancial distress is blatant. Under these circumstances, contagion e€ects may be stronger. Conversely, under conditions in which regulators expedite the closing of distressed banks, a bankÕs ®nancial conditions are more transparent, and contagion e€ects may be tempered. 3. Sample selection and methodology A sample of bank failure announcements is compiled by searching the ÔFederal Deposit Insurance CorporationÕ section of the Wall Street Journal Index from 1980 through 1996. While there are numerous sources of news, the sample selection based on announcements in the Wall Street Journal ensures that the news is disseminated across investors throughout the US We de®ne a bank failure announcement as the report in the Wall Street Journal of an actual bank failure or a critical event that is related to the failure of the bank. When the failure was preceded by a Wall Street Journal announcement about the bank that signaled impending failure, the earlier announcement was used as the relevant announcement date, since our goal was to measure bank industry e€ects in response to new information signaling a bank failure. Thus, a subsequent announcement for the same bank is less likely to signal any new information about the industry. A total of 99 bank failures was assessed. The sample of bank failures includes independent banks, bank holding companies and the subsidiaries of bank holding companies. Dates and brief descriptions of the bank failures and critical events related to each of the 99 bank failures are provided in Appendix A. The bank failures that quali®ed for the sample selection process are classi®ed by year in Table 1. There were more bank failures in the late 1980s than in the early 1980s or in the 1990s. However, these results are at least partially attributed to an unusually strong economy in those years. Panel A of Table 2 provides a summary of the characteristics of the 99 failing banks. The mean and median total assets of the failed banks are $1476.71 million and $40.37 million, respectively. Thus, most of the failed banks are small banks. The mean book-value capital-to-assets ratio is 3.10%, while the median is 4.05%. Note that 50 (51%) of the 99 failed banks are independent banks. The other failures include 26 one-bank holding companies, 13 multibank holding companies, and 10 subsidiaries of multibank holding companies. All failing banks and bank holding companies in our sample have subsidiaries in the same state that they are headquartered. Nine failed banks

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

661

Table 1 Distribution of a sample of 99 bank failures for the period 1980±1996a Year

Number

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996

3 3 2 10 7 10 35 6 10 7 1 2 0 1 0 1 1

Total

99

a

This table presents the distribution of a sample of 99 bank failures. The sample of bank failures was developed over the periods from 1980 to 1996 by searching the ``Federal Deposit Insurance Corporation'' section of the Wall Street Journal Index. We de®ne a bank failure announcement as the report in the Wall Street Journal of an actual bank failure or a critical event that is related to the failure of the bank. The sample of bank failures includes independent banks, bank holding companies and the subsidiaries of bank holding companies, publicly traded banks, and privately held banks.

are publicly traded while 90 are not. Finally, eight banks failed in the period following the passage of DIDMCA (1980±1982), 78 bank failures are reported for the Garn-St. Germain period (1983±1988), 10 failures are reported for the period after FIRREA (1989±1991), and only three bank failures occurred after FDICIA (1992±1996). Most studies on intra-industry e€ects do not restrict their rivals to be in a geographic region within the US Yet, given the geographic restrictions that had historically been enforced in the banking industry, many publicly traded banks had focused their operations in a single state. While some banks have spread their operations throughout some regions in the US, some still focus most of their operations within a single state. We de®ne rivals in two ways. First, we de®ne rivals on a national level, as all publicly traded banks in the US at the time of each bank's failure. Second, we de®ne rivals at the state level, as all publicly traded banks that are headquartered in or have subsidiaries in the same state as the failed bank. That is, the analysis is conducted by de®ning a separate market for every state. Even within a state, the sample of rivals changes over time due to bank consolidations, initial public

662

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

Table 2 Summary statistics for the failed banks and their corresponding rival portfolios during the 1980± 1996 perioda Description Panel A: Summary statistics for the 99 failed banks Total assets ($ million) Mean Median Book value of capital to total assets (%) Mean Median Type of failed bank (#) Independent bank One-bank holding company Multibank holding company Subsidiary of a multibank holding company Ownership status of failed bank (#) Publicly traded Not publicly traded Bank regulation (#) DIDMCA (1980±1982) Garn St. Germain (1983±1988) FIRREA (1989±1991) FDICIA (1992±1996) Panel B: Summary statistics for rivals of the 99 failed banks Total assets of rival portfolios ($ million) Mean Median Book-value of capital-to-total assets (%) Mean Median Mean number of rivals per failure Median number of rivals per failure Minimum number of rivals per failure Maximum number of rivals per failure Number of di€erent states represented

Statistic

1476.71 40.37 3.1 4.05 50 26 13 10 9 90 8 78 10 3

4694.63 3068.06 5.71 5.88 8.23 6 1 35 26

a

This table presents the summary statistics for the failed banks and their corresponding rival banks. The sample of bank failures includes independent banks, bank holding companies and subsidiaries of bank holding companies, publicly traded banks and privately held banks. We de®ne rivals at the state level, as all publicly traded banks that are headquartered in or have subsidiaries in the same state as the failed bank.

o€erings, and failures. Thus, the composition of the sample of rivals can vary among the bank failures analyzed in this study, either because failures occur in di€erent states, or because of changes in market structure within each state over time. We use the CRSP daily returns ®les to compile a list of all publicly traded banks in the US at the time of each bank's failure, as rival banks. In

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

663

addition, each rival bank had to be listed as a bank or a bank holding company in the Report of Condition and Income Database on the Federal Reserve Bank of ChicagoÕs web site. The sample of rival banks was screened to exclude the failed bank, and other confounding events during the examination window. 1 To be included in the sample of rivals de®ned at the state level, a bank must have either its headquarters or one or more subsidiaries based in the respective state. This means that a bank can be considered a rival in more than one state. 2 The state in which the bank headquarters and bank subsidiaries are located are obtained from the Report of Condition and Income Database. Descriptive statistics for the sample of rival banks that are located in the same state as the failed banks are disclosed in Panel B of Table 2. The rival portfolios comprise a total of 815 banks. The mean and median total assets of rival banks are $4694.63 million and $3068.06 million, respectively. The average number of rival banks per failure is 8.23, and the minimum, median and maximum rival banks per failure are 1, 6 and 35, respectively. The diversity of the banks in the sample is substantiated in that the failed banks and their corresponding rivals are located in 26 di€erent states. The event study methodology is used to measure the average daily abnormal returns in response to bank failure announcements. The event date, t0 , is de®ned as the date of the ®rst failure-related event for the failing bank. 3 We follow the methodology employed by Lang and Stulz (1992) when computing the abnormal returns to rival banks. For each announcement, we form an equally weighted portfolio of all rival banks that are headquartered in or have subsidiaries in the same state as the failed bank, and are publicly traded at the time of the announcement. The procedure of creating an equally weighted portfolio accounts for potential cross-sectional correlation of returns in the industry. The returns of the rival portfolio are used to estimate the market model parameters from the period t 220 to t 20 relative to the event date: Rpt ˆ ap ‡ bp Rmt ‡ ept ; where Rpt is the return on rival portfolio p over day t, Rmt the return on the CRSP equally weighted market index on day t, ap estimated intercept term for portfolio p, bp estimated beta coecient for the rival portfolio p, and ept is the 1 Confounding events include earnings surprises, dividend changes, security o€erings, stock repurchases, and other forms of corporate restructuring. 2 For example, NationsBank (a multi-BHC), which was headquartered in Charlotte, NC and had several large subsidiaries in Texas for much of the sample period would be classi®ed as a rival of First Republic Bank, TX. 3 See Appendix A for a listing of the critical events surrounding bank failures.

664

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

error term of the market model for rival portfolio p on day t. 4 The abnormal return of each rival portfolio p and for each day t in the event period t 1 ±t‡1 is computed as ARpt ˆ Rpt

…ap ‡ bp Rmt †;

where ARpt is the abnormal return of the rival portfolio and Rpt is the daily return of the rival portfolio. Cumulative daily abnormal returns (CARs) over two di€erent time intervals are obtained including t0 ±t‡1 and t 1 ±t‡1 . We then compute the equally weighted average of the rival portfolio daily ARs and CARs. Following the methodology of Mikkelson and Partch (1988), the zstatistics are computed and used to test for statistical signi®cance of standardized average ARs and CARs of the rival portfolios. Computational details of the z-statistic used appear in Appendix B. 4. Empirical results 4.1. Abnormal returns to rival portfolios The bank industry e€ects of bank failures are disclosed in aggregate in Panel A of Table 3. For this initial analysis, all publicly traded banks for which share price information is available are classi®ed as rivals. The mean intra-industry e€ect, based on the three-day CAR is )0.12%. This e€ect is statistically signi®cant at the 0.05 level, and supports the hypothesis that a bank failure announcement signals contagion e€ects for all publicly traded banks. When classifying rivals according to the state of the failing bank (Panel B), the mean three-day cumulative abnormal return (CAR 1;0;‡1 ) of rivals in response to the bank failures is )1.49%, which is signi®cant at the 0.01 level. 5 Given a mean market value of $426.99 million per rival bank across the sample of 99 bank failures, these results imply a mean market decline of $52.36 million in the banking industry per bank failure. 6 Furthermore, the proportion of bank failures in which bank rival portfolios experienced a negative CAR 1;0;‡1

4 Similar results are obtained when the market model parameters are estimated using a postannouncement estimation period from t‡20 to t‡220 , and when the CRSP value-weighted market index is used. 5 In one instance, in which the failures of two subsidiaries (Citizens Bank of Texas and Citizens Bank ± Houston) of the same multi-BHC were reported separately on the same date, we re-compute the rival response by treating both events as one, and ®nd qualitatively similar results. 6 The average market loss of $52.36 million in the banking industry per bank failure is obtained as the product of the average three-day abnormal return to the rival banks … 0:0149†, the average market value per rival bank ($426.99 million), and the mean number of rivals per failure (8.23).

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

665

Table 3 Cumulative abnormal returns to rival banks in response to bank failure announcementsa Number of events

Event period

CAR (%)

z-Statistic

% Pos

Panel A: CARs for all rival banks (all publicly traded banks in the US at the time of each bank failure) 99 )1 )0.03 )1.19 49 99 0 )0.02 )1.06 52 99 +1 )0.07 )2.28 40 46 99 [0,+1] )0.09 )2.35 99 [)1,0,+1] )0.12 )2.44 45 Panel B: CARs for rival banks matched by location (all publicly traded banks that are headquartered in or have subsidiaries in the same state as the failed bank) 99 )1 )0.36 )1.03 43 37 99 0 )0.54 )2.83 99 1 )0.59 )1.91 45 99 [0,+1] )1.13 )3.28 32 99 [)1,0,+1] )1.49 )3.17 31 a This table presents the abnormal returns to rival bank portfolios in 99 bank failure announcements. Rival portfolios contain all rival banks that were publicly traded at the time, grouped into portfolios by the event. The sample period is 1980±1996. Abnormal returns are calculated as the di€erences between the actual returns and expected returns. Expected returns are generated from the market model parameters, estimated with daily returns from the period t 220 to t 20 , where t0 is the event date. The z-statistic tests the null hypothesis that the average abnormal return equals zero. The percentage of positive CARs of the respective rival portfolios is shown in the last column. * Signi®cant at the 0.05% level. ** Signi®cant at the 0.10% level.

on average was signi®cantly less than the hypothesized level of 0.50 under conditions in which there was no e€ect. To determine whether there is a di€erence in response to bank failures for rival portfolios de®ned on the national level versus the state level, a t-test of di€erence in mean CARs is applied. We focus on the three-day [)1,0,+1] event window since rival portfolio price responses were strongest in this window for both de®nitions of rival banks. The t-statistic is 3.48 (statistically signi®cant at the 0.01% level), which suggests a stronger negative price response for those rivals that were headquartered in or have subsidiaries in the same state as the failing bank. Consequently, we focus our cross-sectional analysis on those rivals that were located in the same state since those rival banks were a€ected by the bank failures to a greater degree. 4.2. Distribution of rival portfolio e€ects The distribution of valuation e€ects of bank rival portfolios classi®ed according to the same state as the failing bank is shown in Table 4. While some failures had a larger negative impact on rivals than others, this table shows that

666

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

Table 4 The distribution of rival bank three-day abnormal returns associated with 99 bank failuresa Interval for CARs

Number of CAR [)1,0,+1]

2% < CAR 1% < CAR < 2% 0% < CAR < 1% )1% < CAR < 0% )2% < CAR < )1% CAR < )2%

4 10 17 35 9 24

Total

99

a

This table presents the distribution of rival bank three-day abnormal returns associated with 99 bank failures over the sample period 1980±1996. Rival portfolios contain all rival banks that were publicly traded at the time, grouped into portfolios by the event. Abnormal returns are calculated as the di€erence between the actual returns and expected returns. Expected returns are generated from the market model parameters, estimated with daily returns from the period t 220 to t 20 , where t0 is the event date. Daily abnormal returns are summed over the three-day interval t 1 to t‡1 to obtain CAR[)1,0,+1].

many bank failures have some degree of contagion e€ects on rival banks. Thus, contagion e€ects within the state are not restricted to the largest bank failures. 4.3. Factors hypothesized to explain variation in rival bank portfolios The distribution of the valuation e€ects of bank rivals suggests that the information content implied about the rivals varies among bank failure announcements. A cross-sectional analysis is conducted to determine why valuation e€ects of bank rival portfolios vary among bank failures. The information content is hypothesized to vary because of di€erences in the failures, in the characteristics of the rivals, or in the regulation of banks, as explained below. Bank holding company status. A multibank holding company that owns multiple subsidiaries normally has its operations more spread throughout a given state. Therefore, the contagion e€ects of the failure of a multibank holding company with multiple subsidiaries are expected to be more widespread within the respective state. Phase of ®nancial distress for the failing bank. The valuation e€ects of a bank failure could be conditioned on the degree of signal about impending failure. For some failed banks, there was no public announcement of severe ®nancial problems; the ®rst public announcement was that the bank was declared insolvent. If the market is already aware of a bankÕs ®nancial problems before it fails, the announcement of insolvency may not contain much additional information. The capital-to-asset ratio of the failing bank in the year-end preceding the bank failure announcement is used as a proxy to represent the phase of ®nancial distress. Banks that appear to be weaker by the time of the

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

667

announcement are expected to emit less pronounced intra-industry e€ects at the time of their announced failures. Size of the failing bank. The failure of a larger bank is expected to have a more pronounced adverse impact on rival banks, because it receives more publicity and may create more fear by investors. The size of the failing bank is measured as the natural logarithm of the book value of its total assets. Ownership status of failing bank. A publicly held bank is typically more wellknown to the investment community than one that is privately held. In addition, investors may perceive that a failing publicly traded bank has more similar characteristics to other publicly traded banks in the region. Thus, the publicity of a publicly traded bank may emit a stronger signal to investors who focus their attention on publicly traded ®rms. An alternative argument is that the announced failure of a publicly traded bank will contain less information, because investors are more aware of the condition of publicly traded banks. Thus, contradictory hypotheses exist, and can only be resolved empirically. A dummy variable is used to distinguish the failure of a publicly held bank from that of a privately held bank. Size of rival banks. Since the sample of rivals varied with the state of the failing bank, the characteristics of the rival bank portfolio varied among failures. In particular, the size of the rival bank portfolios varies among bank failures, since the size of the rival banks is dependent on the state of the bank that failed. Contagion e€ects of bank failures are expected to be more pronounced when the rival portfolio is composed of relatively small banks, because these banks tend to be less diversi®ed and may be more exposed to external shocks. The median size of rival banks is determined for each bank failure, in which bank size is measured by the logarithm of the book value of the failed bankÕs total assets. Capital levels of rival banks. The capital levels of rival bank portfolios also vary among bank failures. Contagion e€ects of bank failures are expected to be more pronounced on rival bank portfolios that have lower capital levels, because those rival portfolios are less able to withstand ®nancial distress. The capital level is measured as the book value of the capital divided by total assets in the year-end prior to the failure announcement. Bank regulation. The intra-industry e€ects may be in¯uenced by the regulatory regime. Four regimes are considered: (1) 1980±1982, following the Depository Institutions Deregulation and Monetary Control Act (DIDMCA), (2) 1983±1988, following the Garn-St. Germain Act, (3) 1989±1991, following the Financial Institutions Reform, Recovery, and Enforcement Act, and (4) the 1992±1996, following the FDIC Improvement Act. The DIDMCA created a more competitive banking environment by removing interest rate ceilings on deposits and allowing more ¯exibility for savings institutions. The Garn-St. Germain Act also facilitated competition by allowing more ¯exibility for ®nancial institutions to cross state lines when

668

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

acquiring a failed institution, and allowed for the creation of money market deposit accounts. The FIRREA of August 1989, required a higher minimum level of capital for savings institutions, and required that they phase out their holdings of junk bonds. In November 1991, the FDIC Improvement Act was passed. This act expedited the correcting of de®cient capital ratios by banks, and expedited the closing of troubled banks. Therefore, the act may ensure that banks are in better ®nancial shape to withstand adverse conditions in the banking industry. In general, the environment during the ®rst two regimes re¯ected an increase in competition, while the environment during the last two regimes re¯ected a focus on safeguards against excessive risk. Thus, it is hypothesized that the valuation e€ects of a bank failure would emit stronger contagion e€ects during the ®rst two regimes than during the last two regimes. Dummy variables are used to identify the speci®c regulatory period in which the bank failure occurred. Model. The variables that are hypothesized to a€ect the cross-sectional variation in valuation e€ects among bank failures are contained within the following model: RBCARj ˆ k0 ‡ k1 MBHCj ‡ k2 PHASEj ‡ k3 BANKSIZEj ‡ k4 PUBLICj ‡ k5 RIVCAPj ‡ k6 RIVSIZEj ‡ k7 GARNj ‡ k8 FIRREAj ‡ k9 FDICIAj ‡ lj ; where RBCAR ˆ three-day ()1,0,+1) rival banks' portfolio CAR at the announcement of the bank failure j; MBHC ˆ dummy variable which takes a value of 1 if the failed bank is a multibank holding company, and zero otherwise; PHASE ˆ the book-value of capital-to-assets ratio of the failed bank; BANKSIZE ˆ the natural log of the book value of total assets of the failed bank; PUBLIC ˆ dummy variable which takes a value of one if the failed bank is publicly traded, and zero otherwise; RIVCAP ˆ the median bookvalue of capital-to-assets ratio of rival banks; RIVSIZE ˆ the natural log of the median book value of total assets of rival banks; GARN ˆ 1 if the bank failure occurred between 1983±1988, and zero otherwise; FIRREA ˆ 1 if the bank failure occurred between 1989±1991, and zero otherwise; FDICIA ˆ 1 if bank failure occurred between 1992±1996, and zero otherwise; k1 ; . . . ; k9 ˆ parameters to be estimated; and lj ˆ error term. The regression model was tested for heteroskedasticity using the White's (1980) test, and none was detected. Thus, we applied ordinary least squares 7 We also perform tests of the regression parameter estimates using WhiteÕs (1980) heteroskedasticity-consistent covariance matrix. The results are qualitatively similar to those obtained using the OLS. Hence the results for the OLS are reported. The complete results are available from the authors.

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

669

(OLS) regression analysis to test the model. 7 Like most multivariate models, there is some degree of collinearity between the independent variables speci®ed in the multivariate model speci®ed above. Table 5 shows that the MBHC variable is correlated with the BANKSIZE and PUBLIC variables; in addition, the PUBLIC variable is correlated with the BANKSIZE variable. The PHASE variable is correlated with the GARN variable; the RIVCAP variable is correlated with the FIRREA and FDICIA variables; the RIVSIZE variable is correlated with the FDICIA variable; and the GARN variable is correlated with the FIRREA and FDICIA variables. The correlation coecients between these pairs of variables, which range between 0.23 and 0.76 (in absolute values) and are signi®cantly di€erent from zero, could bias the estimated t-statistics in the cross-sectional model. Given the potential bias on signi®cance tests of coecients due to multicollinearity, we orthogonalize the cross-sectional model to remove the collinearity between MBHC, BANKSIZE, and PUBLIC. The MBHC variable is regressed on BANKSIZE and PUBLIC, and the error terms from this Table 5 Correlation matrix of regression variablesa MBHC PHASE BANKSIZE PUBLIC RIVCAP RIVSIZE GARN FIRREA FDICIA

1.00 0.06

1.00

0.76

0.10

1.00

0.62

0.06

0.65

1.00

)0.13

0.19

)0.07

)0.02

1.00

0.01

0.13

0.07

)0.02

)0.18

1.00

0.05 0.08

0.27 )0.19

0.03 0.15

0.02 0.10

0.17 )0.46

0.09 0.03

)0.07

)0.03

0.04

)0.06

MBHC

PHASE

BANKSIZE

PUBLIC

0.23 )0.30 RIVCAP

RIVSIZE

1.00 )0.62

1.00

)0.35

)0.05

1.00

GARN

FIRREA

FDICIA

a The variables are de®ned as follows: MBHC ˆ 1 if the failed bank is a multibank holding company, and zero otherwise; PHASE ˆ the book value of capital to total assets ratio of the failed bank; BANKSIZE ˆ the natural log of the book value of total assets of the failed bank; PUBLIC ˆ dummy variable which takes a value of one if the failed bank is publicly traded, and zero otherwise; RIVCAP ˆ the median book-value of capital-to-total assets ratio of rival banks; RIVSIZE ˆ the natural log of the median book-value total assets of rival banks; GARN ˆ 1 if the bank failure occurred between 1983±1988, and zero otherwise; FIRREA ˆ 1 if the bank failure occurred between 1989±1991 period, and zero otherwise; and FDICIA ˆ 1 if bank failure occurred between 1992±1996 period, and zero otherwise. * Signi®cant at the 5% level.

670

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

Table 6 OLS regression results explaining the three-day rival banks' portfolio CARsa Variable

Coecient

Intercept MBHC PHASE BANKSIZE PUBLIC RIVCAP RIVSIZE GARN FIRREA FDICIA

)0.1267 )0.0690 )0.0105 )0.0055 )0.0284 0.7503 0.0101 )0.0099 0.0286 0.0025

Sample F-Value R2 Adj R2

99 6.282 0.4023 0.3383

t-Statistic 

)3.38 )3.58 )0.20 )1.79 )2.80 4.35 2.55 )0.85 1.73 0.11

VIF 0.00 1.61 1.14 1.70 1.02 1.40 1.18 2.30 2.25 1.59

a

RBCARj ˆ k0 ‡ k1 MBHCj ‡ k2 PHASEj ‡ k3 BANKSIZEj ‡ k4 PUBLICj ‡ k5 RIVCAPj ‡ k6 RIVSIZEj ‡ k7 GARNj ‡ k8 FIRREAj ‡ k9 FDICIAj ‡ lj : The variables are de®ned as follows: RBCAR ˆ three-day ()1,0,+1) rival banks' portfolio CAR at the announcement of the bank failure j; MBHC ˆ 1 if failed bank is a multibank holding company, and zero otherwise; PHASE ˆ the book-value of capital-to-total assets ratio of the failed bank; BANKSIZE ˆ the natural log of the book-value of total assets of the failed bank; PUBLIC ˆ dummy variable which takes a value of one if the failed bank is publicly traded, and zero otherwise; RIVCAP ˆ the median book-value of capital-to-total assets ratio of rival banks; RIVSIZE ˆ the natural log of the median book-value total assets of rival banks; GARN ˆ 1 if the bank failure occurred between 1983±1988, and zero otherwise; FIRREA ˆ 1 if the bank failure occurred between 1989±1991 period, and zero otherwise; and FDICIA ˆ 1 if bank failure occurred between 1992±1996 period, and zero otherwise. The VIF statistics are reported to demonstrate that multicolinearity is not in¯uencing the coecients. * Signi®cant at the 5% level. ** Signi®cant at the 10% level.

regression are used in place of the original values of MBHC to remove the collinearity between MBHC and the other two variables. A similar procedure is applied to BANKSIZE, whereby the error terms replaced the original values. Results. Results of the cross-sectional analysis are disclosed in Table 6. The variance in¯ation factors (VIFs) are reported to assess the extent of muticollinearity. The VIFs ranged from 1.02 to 2.30, indicating multicollinearity is not in¯uencing the coecients. 8 The cross-sectional model employed to test each 8

Some studies use VIF > 5 as an indication of severe multicollinearity (see Marquardt and Snee, 1975). Furthermore, univariate regressions show the magnitude and signi®cance of the coecients are stable.

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

671

of the variables identi®ed above explains about 33% of the variation in the contagion e€ects (based on the adjusted R2 statistic) in response to bank failures. The model tested is signi®cant based on the F-statistic. The MBHC variable is inversely related to RBCAR and signi®cant, which suggests that the announced failure of a multibank holding company elicits more pronounced contagion e€ects than the announced failure of either an independent bank, a one-bank holding company or a subsidiary of a multibank holding company. The coecient of the BANKSIZE variable is negative and signi®cant, which suggests that contagion e€ects are more pronounced when the failed banks are relatively large. Also, the PUBLIC variable is inversely related to RBCAR and signi®cant, which supports the hypothesis that the contagion e€ects are stronger when the failed bank is publicly held. The coecient for the RIVSIZE variable is positive and signi®cant, which supports the hypothesis that rival portfolios containing larger rival banks experience less pronounced contagion e€ects due to bank failures. The coecient of the RIVCAP variable is also positive and signi®cant, which supports the hypothesis that the rival portfolios containing better capitalized rival banks experience less pronounced contagion e€ects due to bank failures. Regarding the regulatory regime variables, the FIRREA variable is positive and signi®cant, re¯ecting less contagion e€ects following that act. The other regulatory regime variables are not signi®cant. Recall from the distribution of the rival bank portfolio CARs that there are 14 events in which the rival portfolios experienced an average abnormal return of greater than 1%. We attempt to determine whether the rival banks associated with those bank failures exhibit characteristics that are distinctly di€erent from rival banks associated with other bank failures. The mean level of the characteristics used in the cross-sectional analysis is disclosed for the rival bank portfolios in which abnormal return exceeds 1% versus the rival bank portfolios that experienced a negative abnormal return in response to bank failures. Results are shown in Table 7. Notice that the rival bank portfolios experiencing an abnormal return of more than 1% contain banks that are signi®cantly larger and have more capital than the rival banks of the other rival portfolios. This evidence corroborates the results derived for RIVSIZE in the multivariate model, and suggests that rival banks that are larger and better capitalized are more insulated from contagion e€ects. 4.4. Impact of bank failures on the total risk of rival portfolios To determine whether bank failures have an e€ect on rival bank portfolioÕs total risk, the corresponding rival portfolioÕs volatility of returns for each of the 99 bank failures is measured over the estimation period

672

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

Table 7 Comparing rival bank portfolios with ARs greater than one percent to rival bank portfolios having negative ARsa Variable

AR >1% (#14)

AR <0% (#68)

t-Statistic for di€erence in means

MBHC (#obs) PHASE (%) BANKSIZE ($ million) PUBLIC (#obs) RIVCAP (%) RIVSIZE ($ million) DIDMCA (#obs) GARN (# obs) FIRREA (#obs) FDICIA (#obs)

1 4.73 1076.85 2 6.26 7611.99 0 10 3 1

12 3.12 1969.97 7 5.2 4710.32 5 56 6 1

± 1.32 0.70 ± 1.70 2.03 ± ± ± ±

a The variables are de®ned as follows: MBHC ˆ 1 if the failed bank is a multibank holding company, and zero otherwise; PHASE ˆ the book-value of capital-to-total assets ratio of the failed bank; BANKSIZE ˆ the natural log of the book value of total assets of the failed bank; PUBLIC ˆ dummy variable which takes a value of one if the failed bank is publicly traded, and zero otherwise; RIVCAP ˆ the median book-value of capital-to-total assets ratio of rival banks; RIVSIZE ˆ the natural log of the median book-value total assets of rival banks; DIDMCA ˆ 1 if the bank failure occurred between 1980±1982 period, and zero otherwise; GARN ˆ 1 if the bank failure occurred between 1983±1988 period, and zero otherwise; FIRREA ˆ 1 if the bank failure occurred between 1989±1991 period, and zero otherwise; and FDICIA ˆ 1 if the bank failure occurred between 1992±1996 period, and zero otherwise. ** Signi®cant at the 5% level. * Signi®cant at the 10% level.

(pre-event period) and in a post-event period. 9 The pre-event period rival portfolioÕs total risk is estimated over the 200 days ending 20 days prior to the bank failure. The post-event period rival portfolioÕs total risk is estimated over the 200 days beginning from 20 days after the bank failure announcement. To control for possible shifts in bank volatility that are simply due to shifts in market volatility, we use a ratio, stock return volatility of a bank rival portfolio divided by the volatility of other publicly traded banks. The results are shown in Table 8. The mean shift in the total risk of rival portfolios in response to bank failures is 0.25, which is statistically signi®cant. Therefore, there is a signi®cant increase in the total risk of the rival portfolios as a result of bank failures.

9

Some studies on the e€ects of deregulation on bank risk used beta as a measure of risk (for example, see Sundaram et al., 1992). We also perform all subsequent tests by using beta as the risk measure. Since the results are qualitatively similar, we do not report them.

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

673

Table 8 Mean and median of rival bank portfolios pre- and post-failure total risksa Actual values Pre-failure total risk Post-failure total risk Change in total risk T-test for total risk change (p-value) Sign rank test for total risk change (p-value)

Adjusted values

Mean

Median

Mean

Median

2.03% 2.71% 0.68% 4.50 (0.0001) ±

1.73% 1.84% 0.29% ±

0.96 1.21 0.25 4.12 (0.0001) ±

0.82 0.83 0.10 ±

1237 (0.0001)

1024 (0.0002)

a This table presents the rival portfolioÕs total risks (volatility of returns) in 99 bank failure announcements. The pre-event period rival portfolioÕs total risk is estimated over the 200 days ending 20 days prior to the bank failure. The post-event period rival portfolioÕs total risk is estimated over the 200 days beginning from 20 days after the bank failure announcement. To control for possible shifts in bank volatility that are due to shifts in market volatility, we adjust the volatility of the rival bank portfolio by dividing the actual volatility of rival bank portfolio by the industry return volatility (using all non-rival banks in the sample). Volatility is measured by the standard deviation of returns.

4.5. Distribution of total risk-shifts among rival portfolios The distribution of total risk-shifts on rival bank portfolios in response to bank failures is disclosed in Table 9. Since the total risk-shifts varied among bank failures, we attempt to determine whether the sources of the valuation e€ects are also sources of total risk-shifts. The variation in bank portfolio total risk-shifts may be attributed to the same factors that were associated with the magnitude of the contagion e€ects. While the total risk-shift of a rival portfolio is distinctly di€erent from the valuation e€ects of a rival portfolio, both types of e€ects could be in¯uenced by the same factors. 4.6. Cross-sectional analysis of rival portfolio total risk-shifts Based on the hypotheses presented earlier, we attempt to determine whether the magnitude of a total risk-shift experienced by a rival bank portfolio in response to a bank failure is conditioned upon information about the failing bank, the characteristics of the rival portfolio, and bank regulation. Results of this analysis would suggest whether the shift in total risk in the banking environment surrounding a failed bank could be at least partially controlled by the surviving banks or by bank regulators. The cross-sectional model used to explain the variation in the total riskshifts among the rival portfolios is applied. The change in total risk (i.e., postevent total risk minus pre-event total risk) is used as the dependent variable:

674

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

Table 9 The distribution of adjusted rival bank total risk-shifts (CHVAR) associated with 99 bank failuresa Interval for CHVAR

# of CHVARs

1.0 < CHVAR 0.5 < CHVAR < 1.0 0.0 < CHVAR < 0.5 )0.5 < CHVAR < 0.0 )1.0 < CHVAR < )0.5 CHVAR < )1.0

9 10 43 35 2 0

Total

99

a

This table presents the distribution of adjusted rival bank total risk-shifts (CHVAR) in response to 99 bank failures over the sample period 1980±1996. The adjusted rival bank total risk-shift is measured by the change in adjusted volatilities of the pre-event and post-event periods. The adjusted volatility of the rival bank portfolio is obtained by dividing the actual volatility of rival bank portfolio by the industry return volatility (using all non-rival banks in the sample). Volatility is measured by the standard deviation of returns.

CHVARj ˆ k0 ‡ k1 MBHCj ‡ k2 PHASEj ‡ k3 BANKSIZEj ‡ k4 PUBLICj ‡ k5 RIVCAPj ‡ k6 RIVSIZEj ‡ k7 GARNj ‡ k8 FIRREAj ‡ k9 FDICIAj ‡ lj : Since the variables are similar to those used in the previous cross-sectional analysis, the same collinearity problems exist in this multivariate model, and therefore the model is orthogonalized as explained for the previous cross-sectional model. The model is then applied to assess the cross-sectional variation in rival portfolio total risk-shifts due to bank failures. Results. Results of the cross-sectional analysis of total risk-shifts are disclosed in Table 10. The model is signi®cant based on the F-statistic. Two of the independent variables are signi®cantly related to the risk-shift of rival banks following bank failures. The PUBLIC variable is positive and signi®cant, which supports the hypothesis that the bank rival portfolios were more susceptible to an increase in risk when the failed bank is publicly held. The CAPITAL variable is negative and signi®cant, which supports the hypothesis that rival bank portfolios that are more adequately capitalized (on average) experience less pronounced shifts in total risk in response to bank failures. 5. Conclusions By addressing the cross-sectional variation in bank contagion e€ects, the following questions are addressed. First, do bank failures generally transmit contagion e€ects? Second, why do contagion e€ects on the surviving rival bank portfolios vary among bank failures? Third, do bank failures generally cause a

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

675

Table 10 OLS regression results explaining adjusted total-risk shifts (CHVAR) of rival banksa Variable

Coecient

t-Statistic

VIF

Intercept MBHC PHASE BANKSIZE PUBLIC RIVCAP RIVSIZE GARN FIRREA FDICIA

0.0027 0.0045 )0.0140 )0.0002 0.0162 )0.2458 0.0015 0.0028 0.0089 0.0056

0.17 0.54 )0.61 )0.15 3.66 )3.38 0.86 0.55 1.24 0.61

0.00 1.61 1.14 1.70 1.02 1.40 1.18 2.30 2.25 1.59

Sample F-Value R2 Adj R2

99 4.654 0.3327 0.2612

a

CHVARj ˆ k0 ‡ k1 MBHCj ‡ k2 PHASEj ‡ k3 BANKSIZEj ‡ k4 PUBLICj ‡ k5 RIVCAPj ‡ k6 RIVSIZEj ‡ k7 GARNj ‡ k8 FIRREAj ‡ k9 FDICIAj ‡ lj The variables are de®ned as follows: MBHC ˆ 1 if the failed bank is a multibank holding company, and zero otherwise; PHASE ˆ the book-value of capital-to-total assets ratio of the failed bank; BANKSIZE ˆ the natural log of the book value of total assets of the failed bank; PUBLIC ˆ dummy variable which takes a value of one if the failed bank is publicly traded, and zero otherwise; RIVCAP ˆ the median book-value of capital-to-total assets ratio of rival banks; RIVSIZE ˆ the natural log of the median book-value total assets of rival banks; GARN ˆ 1 if the bank failure occurred between 1983±1988, and zero otherwise; FIRREA ˆ 1 if the bank failure occurred between 1989±1991 period, and zero otherwise; and FDICIA ˆ 1 if the bank failure occurred between 1992±1996 period, and zero otherwise. The VIF statistics are reported to demonstrate that multicolinearity is not in¯uencing the coecients. * Signi®cant at the 5% level.

shift in the total risks of the surviving rival banks? Fourth, why do total riskshifts among the surviving rival bank portfolios vary among bank failures? Our analysis of a sample of 99 bank failures suggests that bank failures generally result in contagion e€ects on the surviving rivals of the failing bank. In addition, contagion e€ects vary among the failures, and are conditioned on various variables speci®c to the failing bank and rival characteristics at the time of the failure. The announcements of bank failures transmit stronger contagion e€ects when the failing bank is a multibank holding company, relatively large, and is publicly held. In addition, the contagion e€ects are more pronounced when the sizes of the corresponding rival banks are relatively small, and when the capital levels of corresponding rival banks are relatively low. Bank contagion e€ects are also conditioned by the regulatory environment, as the e€ects are tempered just after FIRREA.

676

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

Based on an analysis of rival bank portfolio total risk-shifts among bank failures, we document a signi®cant increase in risk in response to the sample of bank failures. Moreover, the distribution of total risk-shifts among the bank failures suggests that total risk-shifts vary among failures. A cross-sectional analysis of total risk-shifts ®nds that an increase in risk of the corresponding rival bank portfolio is more pronounced when the failing bank is publicly held. The total risk-shifts have also been attenuated when the rival banks are relatively less capitalized. The results of this analysis suggest that the susceptibility of the bank environment to contagion e€ects and related shifts in bank total risk can be reduced by continuing the enforcement of quick regulatory action on existing problem banks, and by enforcing high capital standards. To the extent that exposure to contagion e€ects is a form of industry risk, and is a concern to regulators, the results here o€er useful implications for preventing excessive exposure. From a managerial perspective, the results suggest that a bank's exposure to contagion e€ects of another bank's failure may be associated with its own characteristics (such as its capital adequacy). Thus, it appears that banks may have some control over their own exposure to contagion e€ects, regardless of the regulatory environment. Acknowledgements We wish to thank two anonymous reviewers of Journal of Banking and Finance for their valuable suggestions. Appendix A Table 11 Failed banks in sample with bank name, location, date of failure related event, failure related event and total assets Name of failed bank

Location of failed bank

Failure related event date

Failure related eventa

Alaska Mutual Bank Alliance Bancorp of Alaska American City Bank American National Bank of Riverton Banc Oklahoma Corp BancTexas Group Incc BancTexas Group Incc BancTexas Group Incc

AK AK TN WY OK TX TX TX

870209 890124 830223 850611 860722 860630 861121 870609

C B A C B B C B

Total assets of failed bankb 819.34 1176.86 23.83 44.07 3075.97 1757.99 1757.99 831.39

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

677

Table 11 (Continued) Name of failed bank

Location of failed bank

Failure related event date

Failure related eventa

Total assets of failed bankb

Bank of Lake Helen Bank of Loretto Bank of Nortonville Bank of San Marino Bank of Sante Fe Bank of Woodson Basin State Bank, Vernal Bucklin State Bank Clarksdale Bank C.H. Butcher Banks Central Bank&Trust of Tulsa Central National Bank Cherokee County Bank of Center Chokio State Bank Citizens Bank Citizens Bank of Winigan Citizens Bank ± Houston Citizens State Bank of Hay®eld Continental Illinois Corp Des Plaines Bank of Des Plaines East Gadsden Bank of Gadsden Empire State Bank Fairview State Bank First Bank&Trust First Bank&Trust Co., Booker First City Bancorp First City Bank of Oklahoma City First National Bank&Trust of Norman First National Bank & Trust of Oklahoma (FNOC) First National Bank in Ri¯e First National Bank of Bandera First National Bank of Borger First National Bank of Irving First National Bank of Jacksonville First National Bank of Midland First National Bank of Sheridan First National Bank of Toms River First National Bank of Vermont First National Bank, Willows First Republic Bank First State Bank&Trust Co First State Bank, Abilene First State Bank, White Cloud First Trust Bank

FL TN IA CA NM TX UT MO MO TN OK NY AL MN TX MO TX MN IL IL AL NY OK TX TX TX OK OK

800111 850905 860502 830411 880422 820302 880216 841015 851121 830526 860911 870914 840606 861110 890210 860310 890210 880128 840521 810316 801231 890731 860905 890210 861219 870910 850624 860602

C C F C L C C F C M C P C F C P C L L C C C C C C L D C

3.03 26.63 19.32 17.9 152.65 3.68 11.98 16.31 6.53 na 74.97 199.69 39.63 15.48 100.13 6.53 40.17 28.01 42097.37 46.50 40.21 32.85 26.36 20.83 109.89 13681.18 99.87 83.27

OK

860714

C

2482.22

CO TX TX TX AL TX WY NJ VT CA TX TX TX MI CA

860825 860425 860620 860425 850708 831017 860718 910523 930129 861121 880222 860527 890221 880216 950303

P C C C C D C D C C B P F C C

19.97 15.54 81.83 38.46 19.10 1811.97 75.45 1646.88 136.69 66.90 27944.00 190.73 58.80 36.42 227.70

678

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

Table 11 (Continued) Name of failed bank

Location of failed bank

Failure related event date

Failure related eventa

Global Bank, Hialeah Golden Paci®c National Bank Hereford State Bank Heritage National Bank Hohenwald Bank&Trust Co. Ina State Bank of Ina Independent National Bank, Covina Inwood State Bank Lago Vista National Bank Landmark National Bank Lawrence County Bank Louisiana Bank&Trust Co, Shreveport Mcorp Medicine Bow State Bank Merchants&Farmers State Bank in Blythe Merchants Bank of Boston Metrobank, Philadelphia PA Metropolitan Bank&Trust Co, Baton Rouge Mineola State Bank National Bancshares Corporation National Bank of Dyersville National Bank of Texas, Austin New Mexico National Bank Norman Bank of Commerce Northwest Commerce Bank of North Bend Peoples Bank of Mercer Permian Bank, Odessa Port City Bank of Houston Rochelle Bank&Trust Co. Sedgwick County Bank, Julesburg South Side Bank of Chicago Southeast Banking Corp. State Bank of Barnum Stewartship Bank of Portland Strong's Bank Sunbelt National Bank Sunshine State Bank Taylor State Bank of Emington Texas American Bancshares Inc. Texas National Bank of College Station Union Bank&Trust, Bartlesville

FL NY CO TX TN IL CA IA TX CO TN LA

880216 850624 840827 860926 820907 830411 861010 850220 861219 861219 840618 890221

C C C C P C C C C C F F

21.74 141.41 61.83 30.53 27.40 19.63 33.40 8.15 19.19 15.52 25.92 270.40

TX WY CA

871023 860811 830222

E P C

21887.00 5.39 5.91

MA PA LA

900521 960202 861110

D F F

350.00 40.37 82.19

IA TX IA TX NM OK OR

850807 880408 860411 860703 860718 861121 810623

C B C C C C C

6.03 2797.07 43.04 58.81 169.34 45.21 7.47

MO TX TX IL CO IL FL MN OR WI TX FL IL TX TX

860411 860718 880129 801013 861106 810316 910411 830211 840611 850614 870209 860527 850315 880408 861121

C C C C P C B P P C P P C B C

10.41 49.89 60.74 7.96 4.43 27.51 13390.40 13.87 7.10 31.10 37.16 108.40 4.98 5170.50 15.69

OK

880722

C

136.17

Total assets of failed bankb

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

679

Table 11 (Continued) Name of failed bank

Location of failed bank

Failure related event date

Failure related eventa

United American Bank of Knoxville United Southern Bank of Nashville Valley State Bank, Baggs West Coast Bank of Los Angeles Western Bank in Midland Western National Bank in Lovell

TN TN WY CA TX WY

830214 830408 861017 840430 860905 830624

F L C P C C

Total assets of failed bankb 760.47 242.27 5.49 174.98 92.98 7.31

a

A ˆ Federal bank examiners audit troubled institution; C ˆ closed by state authorities; D ˆ declared insolvent; F ˆ failure; L ˆ FDIC provides liquidity to counter withdrawals; M ˆ may be closed by regulators; P ˆ FDIC begins or is expected to begin payments to depositors; S ˆ seized assets because bank was in danger of failing; T ˆ FDIC took over; B ˆ bailout plan; SA ˆ seeking assistance; E ˆ reported earnings loss made failure possible. b Measured at the year-end prior to the failure or failure-related event. c Failures of three di€erent subsidiaries of BancTexas Group were reported on three separate dates.

Appendix B. Computation details of the z-statistic used to test for signi®cance of abnormal returns Following the methodology used by Mikkelson and Partch (1988), the following z-statistic is used to test for statistical signi®cance of standardized abnormal returns and cumulative standardized abnormal returns for the sample observations: 2 3 Pt 2 n X 1 6 7 tˆt1 ARpt  5; Z ˆ p 4 q Pt2 n pˆ1 Var… AR † tˆt1

pt

where ARpt ˆ abnormal return as previously de®ned, t1 ˆ ®rst day of the time interval, t2 ˆ last day of the time interval, n ˆ number of observations, and $ ! % Pt2 ~ t2  m †2 X T …R T2 tˆt1 Rmt 2 ‡ PED Var ARpt ˆ vp T ‡ ; ~ mi R  m †2 ED …R tˆt 1

iˆ1

where Vp2 ˆ residual variance for rival bank portfolio p from the market model over the estimation period, ED ˆ number of days in the estimation period used to estimate the market model, T ˆ number of days in the interval ~ mi ˆ market portfolio return for the ith day of the estimation …t2 ±t1 ‡ 1†, R ~ mt ˆ market portfolio return for day t, R  m ˆ market portfolio avperiod, R erage return over the estimation period. This equation measures the variance of the sum of individual abnormal returns. This method di€ers from the standard formula for the variance of an individual abnormal return because it adjusts for the dependence created by

680

A. Akhigbe, J. Madura / Journal of Banking & Finance 25 (2001) 657±680

cumulating individual abnormal returns calculated using a single set of pa^ derived from the market model. rameter estimates (^ a and b)

References Aharony, J., Swary, I., 1983. Contagion e€ects of bank failures: Evidence from the capital markets. Journal of Business 58, 305±322. Aharony, J., Swary, I., 1996. Additional evidence on the information-based contagion e€ects of bank failures. Journal of Banking and Finance 20, 57±69. Bruner, R., Simms, J., 1987. The international debt crisis and bank security returns in 1982. Journal of Money, Banking, and Credit 19, 46±55. Cornell, B., Shapiro, A., 1986. The reaction of bank stock prices to the international debt crisis. Journal of Banking and Finance 10, 55±73. Diamond, D., Dybvig, P., 1983. Bank runs, deposit insurance, and liquidity. Journal of Political Economy 91, 401±419. Docking, D.S., Hirschey, M., Jones, E., 1997. Information and contagion e€ects of bank loan-loss reserve announcements. Journal of Financial Economics 43, 219±239. Fenn, G.W., Cole, R.A., 1994. Announcements of asset quality problems and contagion e€ects in the life insurance industry. Journal of Financial Economics 35, 181±198. Gay, G.D., Timme, S.G., Yung, K., 1991. Bank failure and contagion e€ects: Evidence from Hong Kong. Journal of Financial Research 14, 153±165. Lamy, R.E., Thompson, G.R., 1986. Penn square, problem loans, and insolvency risk. Journal of Financial Research 9, 103±111. Lang, L.H.P., Stulz, R.M., 1992. Contagion and competitive intra-industry e€ects of bankruptcy announcements. Journal of Financial Economics 32, 45±60. Marquardt, D., Snee, R., 1975. Ridge regression in practice. American Statistician 29, 3±19. Mikkelson, W.H., Partch, M.M., 1988. Withdrawn security o€erings. Journal of Financial and Quantitative Analysis 23, 119±133. Pettway, R., 1976. The e€ects of large bank failures upon investors' risk cognizance in the commercial banking industry. Journal of Financial and Quantitative Analysis 11, 465±477. Sundaram, S., Rangan, N., Davidson III, W.N., 1992. The market valuation e€ects of the ®nancial institutions reform, recovery and enforcement act of 1989. Journal of Banking and Finance 16, 1097±1122. Swary, I., 1986. Stock market reaction to regulatory action in the continental Illinois crisis. Journal of Business 59, 451±473. White, H., 1980. A heteroskedasticity-consistent covariance matrix estimation and a direct test for heteroskedasticity. Econometrica 48, 837±838.