Corporate control, bank risk taking, and the health of the banking industry q

Transcription

1 Journal of Banking & Finance 24 (2000) 1383± Corporate control, bank risk taking, and the health of the banking industry q Ronald C. Anderson a,1, Donald R. Fraser b, * a Department of Finance and Real Estate, Kogod College of Business Administration, American University, 4400 Massachusetts Avenue, NW, Washington, DC , USA b Department of Finance, Lowry Mays College and Graduate School of Business, Texas A&M University, College Station, TX , USA Received 17 April 1998; accepted 13 July 1999 Abstract We present evidence that managerial shareholdings are an important determinant of bank risk-taking. Managerial shareholdings are positively related to total and rm speci c risk in the late 1980s when banking was relatively less regulated and when the industry was under considerable nancial stress. However, following legislation in 1989 and 1991 designed to reduce risk-taking and also re ecting substantial improvements in bank franchise value, managerial shareholdings and total and rm speci c risk became negatively related in the early 1990s. In contrast, systematic risk was unrelated to managerial ownership in both periods. Ó 2000 Elsevier Science B.V. All rights reserved. Keywords: Bank risk taking; Ownership structure; Franchise value JEL classi cation: G21; G28; G30 q The paper has bene ted from the suggestions of J. Bizjak, J. Byrd, D. Dubofsky, D. Ellis, S. Lee, and A. Mahajan. Naturally, all remaining errors are the responsibility of the authors. * Corresponding author. Tel.: ; fax: addresses: Ron.Anderson@mciworld.com (R.C. Anderson), d-fraser@tamu.edu (D.R. Fraser). 1 Tel.: ; fax: /00/$ - see front matter Ó 2000 Elsevier Science B.V. All rights reserved. PII: S (99)

2 1384 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383± Introduction The nance literature abounds with attempts to quantify and explain risk taking behavior at commercial banks. Following Merton (1977) much of this research focuses on the incentives created by the xed-rate deposit insurance system for banks to increase the amount of risk in their asset and liability portfolio (an incentive system referred to as the moral hazard problem). More recently, Keeley (1990, p. 1183) nds evidence to support the hypothesis that ``increases in competition caused bank charter values to decline, which in turn caused banks to increase default risk through increases in asset risk and reductions in capital.'' While there are other explanations of risk-taking focusing on bank (or shareholder) behavior, there is little explicit consideration of the role of bank management in these decisions. Yet it is management rather than shareholders who make the portfolio decisions that determine the risk structure of a bank. Identifying the link between managerial risk preferences and share ownership is a complex task. Conceptually, it might be expected that managers with small ownership stakes in their banks would behave in a risk averse rather than value maximizing way, as they seek to protect the value of their rm-speci c human capital. As their shareholdings increase, however, they have more incentives to engage in risk taking, especially under a xed rate deposit insurance system that existed prior to 1993 and in a period such as the late 1980s when charter or franchise values were diminished. At some substantial levels of shareholdings, however, managers may become entrenched and no longer make value maximizing decisions. To the extent that managerial entrenchment becomes important, the risk/share ownership relationship may thus become curvilinear. We provide evidence on the managerial shareholdings-risk relationship from bank stock price data drawn from two distinct time periods with quite distinct regulatory regimes and bank charter values. We conjecture that a positive relationship should exist in the 1987±1989 period as the xed rate deposits system, lax regulation, and low charter-values encouraged managers to take on additional risk. In contrast, in the latter period, 1992±1994, management risk taking was constrained by additional regulations, including risk-adjusted deposit insurance premiums and also by substantially increased bank charter values. Our evidence is generally consistent with these arguments. We nd that: 1. Total and rm speci c risk are positively and signi cantly related to managerial holdings between 1987 and Total and rm speci c risk are negatively and signi cantly related to managerial holdings between 1992 and Systematic risk is unrelated to ownership in both periods. Existing evidence on the relationship between managerial shareholdings and risk is inconclusive, perhaps re ecting di erences in risk measures, di erent

3 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383± time periods in the analysis, and di erent approaches to analyzing the relationships. Saunders et al. (1990), using capital market measures of risk, provide support for the hypothesis that banks in which managers hold a large fraction of the stock exhibit signi cantly higher risk-taking behavior over the 1978± 1988 period, and that the incremental risk was more prevalent following deregulation. Demsetz et al. (1997) also nd a positive (and nonlinear) relationship between market risk measures and managerial shareholdings over the 1991±1995 period, though only at low franchise value banks. In contrast, Chen et al. (1998) nd a negative and nonlinear relationship between managerial ownership and bank risk over the 1988±1993 period. Gorton and Rosen (1995) focus on the risk/managerial shareholdings relationship as it is a ected by both entrenchment and by shifts in the economic conditions of the banking industry. They argue that ``bad'' managers take excessive risk in periods when the industry is unhealthy (as re ected in low franchise values). Their model predicts a nonlinear relationship between managerial shareholdings and risk, with risk initially increasing with shareholdings and then declining as entrenchment becomes a signi cant motivating factor. Their results, drawn from the 1980s period of low franchise values, are generally consistent with their model. While their model implies that risk will decline in the 1990s as bank franchise values increase, the time period encompassed by their study did not permit analysis of this issue. Our sample period and model does, however, allow us to address this important issue. Section 1 describes the sample and the variables used in the analysis, while Section 2 presents the empirical model. Section 3 discusses the results and Section 4 concludes the paper. 2. Sample and variables description 2.1. The sample Our sample consists of a panel of 150 banks covering the period from 1987 through The banks were initially selected from the November 1988 Compact Disclosure CD-ROM. Information is gathered on all banks listed in standard industry classi cation of national commercial banks (6021), state commercial banks (6022), and commercial banks not classi ed elsewhere (6029). The search of the Compact Disclosure database yielded a total of 604 individual banks. Return information from the Center for Research in Security Prices (CRSP) daily tapes is used to calculate the market based measures of risk. CRSP covers 423 of the original 604 banks taken from Compact Disclosure. The remaining 181 banks for which we could not obtain CRSP data are banks that delisted, traded on regional exchanges, or were ``pink sheet'' rms. Once we identify

4 1386 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383±1398 those rms carried by the CRSP database, the remaining years of the time series are collected from Compact Disclosure. Because many of the rms in the remaining sample are traded infrequently, we follow Esty (1997) and discard 156 banks whose equity went untraded for 75 days or more during a year. The discarded banks have, on average, 129 days per year where no trades took place. 2 Because we want to analyze the relation between bank risk and managerial holdings under di erent conditions we restrict the sample to the same set of banks during both periods. As a consequence, we require that each bank exist over the 8 year sample period, a constraint that requires us to drop 117 rms from the sample, resulting in a nal sample of 150 banks with 8 years of data per bank or 1200 bank-years 3 (this sample selection procedure does, of course, eliminate failed banks as well as acquired banks) Variable description Consistent with prior studies, we de ne management holdings as the aggregate percentage of shares held by all o cers and directors of the bank as reported in Compact Disclosure. This ownership measure is the year-end holdings of o cers and directors for the preceding year as reported in the bankõs proxy statement. Anderson and Lee (1997) demonstrate that Compact Disclosure is a reliable surrogate for the data provided in corporate proxy statements. However, in their analysis, they show that Compact Disclosure only reports the largest class of shares for rms with multiple equity classes. As a consequence, we verify the Compact Disclosure information with that of the annual proxy statement for years 1987, 1989, 1992, and We use a single-index (market returns) market model to estimate the returngenerating process for bank stocks. 5 From the model, we estimate three dif- 2 The typical trading year consists of 250±255 trading days. Thus, the rms deleted from the sample trade on about 50% of the available days. For several of the banks, the number of days where trading did not occur exceeds 200 days per year. 3 The average number of years that the discarded banks are in the sample is 3.2 years with a range of 1±7 years. Over 64% of these banks are newly listed after 1991 on the CRSP tapes. The remaining banks delisted prior to 1991 due to failures, mergers or acquisitions. 4 We replicated the analysis reported later in the paper for samples that included a number of banks that failed during our analysis periods (principally 1987±1989). While these results are less powerful, the classi cations based upon statistical signi cance do not change. 5 In a prior version of this paper, we used a two-index model incorporating both market returns and interest rates. The results are not signi cantly or materially di erent from the single index model and, as a result, we use the simpler model. In our two index model, We did not correct for autocorrelation of our interest rate series since Flannery and James (1984) and Unal and Kane (1994) demonstrate that such a correction does not materially e ect the calculations. Also, we do not orthogonalize the interest rate and market returns series in view of evidence provided by Giliberto (1985) that such adjustments can bias results.

5 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383± ferent measures of rm risk: Total risk is the standard deviation of the bankõs daily returns; systematic risk is the component of risk that captures the in uence of underlying economic and nancial conditions that a ect the entire banking industry. Firm speci c risk measures the stock price volatility that is unique to an individual bank and is related to the nature of its loan, investment, deposit, and capital structure. Firm speci c risk is measured as the standard deviation of the residuals of the market model for each rm. Systematic risk is measured as the di erence between total and rm speci c risk. 6 More speci cally, the risk measures are calculated as follows: Total risk ˆ 1 X xi x 2 ˆ r 1=2 ; n where x is the daily return of bank I. Firm specific risk ˆ 1 X ei e 2 ˆ e 1=2 ; n where e's are the residuals of the single-index market model. 1 2 Total risk ˆ a i b i market returns e i : Market returns are proxied by the CRSP equal-weighted returns. Systemic risk ˆ Total risk Firm specific risk: 3 Since the market experienced substantial volatility in October 1987, we calculate our risk measures with and without the last 15-days of October The regressions yield statistically similar results, therefore we include the October 1987 ``crash'' in all of our calculations. We also estimate the equations with weekly returns. These results (not reported here) are not materially different from the daily returns results. Demsetz and Strahan (1997) argue that frequency of trading is a proxy for the speed with which new information is captured in stock price and that this variable should be correlated with the variances of a bankõs assets, liabilities, and o -balance sheet portfolios. We measure frequency of trading as the average daily volume of shares traded for each bank divided by the total number of shares outstanding for the bank. 6 We calculate our risk measures using banks daily returns and as a result, may introduce some measurement error into our rm speci c and systematic risk dependent variables. Kmenta (1986) notes that any errors present in the rst-stage regression model (determining the dependent variable) are also introduced into the second-stage model. He indicates however, that measurement errors from the rst-stage regression are ``merged'' into the disturbance term in the second-state regression. Consequently, the second-stage regression with the merged disturbance term satis es all of the basic assumptions of ordinary least squares techniques and any measurement error in our dependent risk measures are not of signi cant importance.

6 1388 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383±1398 Size may play an important role in determining risk levels at banking organizations because larger banks are more capable of diversifying risk, both geographically and by industry, than small banks. Moreover, larger banks have greater access to capital markets and thus more exibility to adjust to unexpected liquidity and capital shortfalls. We might expect larger banks to have lower risk levels than smaller banks if their portfolios are merely larger replications of those at small banks. However, if large banks di er from small banks in the composition of their asset base or portfolios (holding greater amounts of risky-type loans, for example or greater o -balance sheet risk), their overall risk levels may be higher than that of small banks. Demsetz and Strahan (1997) report that large bank holding companies o set the potential bene ts of diversi cation through adopting more risky loan portfolios and operating with more leverage. The natural log of the book value of total assets is used as our measure of bank size. We use KeeleyÕs (1990) adaptation of TobinÕs Q as a proxy for the health of the individual banking rm. Q is equal to the sum of market value of common equity (price per share times number of shares) plus the book value of liabilities divided by the book value of assets. Keeley argues that this Q measure is a proxy for franchise value. He nds that banks with higher Q values have lower default risk as measured by lower risk premiums on their large, uninsured certi cates of deposit. Moreover, Demsetz et al. (1997) nd that risk at individual banks is inversely related to this Q measure. Monitoring of managerial risk taking may come from the nature of ownership of the banks as well as from the regulatory structure, depositors, and from other sources. We proxy the structure of ownership by including the percentage of shares held by una liated or outside blockholders. These blockholders consist of all una liated shareholders who own 5% or more of the voting shares. Blockholdings data is gathered from the 1987, 1989, 1992, and 1994 corporate proxy statements. 3. Empirical method We use a model similar to Saunders et al. (1990) to test our hypothesis Risk ˆ a 0 a 1 Holdings a 2 Keeleys Q a 3 Outside blockholdings a 4 ln Total assets a 5 Frequency e I : 4 Since there is some evidence (see McConnell and Servaes, 1991; Gorton and Rosen, 1995 for example) that the relationship between managerial holdings and bank risk is nonlinear, we estimate an additional piecewise version of this model, (as in Gorton and Rosen) and report both coe cients in the relevant

7 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383± tables. 7 We split our sample into two distinct periods: 1987±1989 and 1992± 1994, inclusive. The 1987±1989 period is characterized as an unhealthy period for the banking industry and one of regulatory laxity. The period from 1992 to 1994 is typi ed as a healthy banking industry and as one with increased regulatory constraints. We use panel data techniques since the regression disturbance term is likely to have a serially related component, a cross-sectionally related component, or both. To account for serial correlation and heteroskedasticity, we employ timeseries cross-sectional (TSCS) regression analysis. The TSCS procedure adds the individual and time-speci c random e ects to the variance±covariance matrix of the disturbance term. As a result, the variance±covariance matrix of the disturbance term is no longer a diagonal matrix with the same value at each diagonal element as with OLS, but a full matrix in which each element can assume a unique value. We use the Fuller and Battese (1974) TSCS method which ef- ciently estimates our parameters using generalized least squares. 8; 9 Hausman Speci cation tests, comparing xed e ect to random e ect panel data techniques, suggest that the two techniques provide signi cantly indi erent regression results. Therefore, we employ a random e ects model for our analysis. 4. Empirical results 4.1. Descriptive statistics Table 1 contrasts the market risk measures, managerial holdings, KeeleyÕs Q, outside blockholdings, total assets, and frequency of trading for the 1987± 1989 and 1992±1994 periods. While neither total risk nor rm speci c risk measures di er signi cantly between the two periods, each of the other variables does di er between the two periods. Systematic risk declined substantially between the two periods. Managers hold slightly fewer (9.63% vs %) shares in the 1992±1994 period. Moreover, the banks in the 1992±1994 period 7 We also estimated a model that included holdings and (holdings) 2. The (holdings) 2 variable was not signi cant in any of the regressions. 8 We also run all of our regressions using ordinary least squares. In many instances, the signi cance of the parameter estimates increased substantially compared to those found using panel data techniques, suggesting substantial loss of e ciency with OLS estimation. 9 We tested causality in our panel data set by using a cross-lagged regression model employed by Davidson et al. (1997) and Batemean and Strasser (1983). In the rst equation of the model, we regress our risk measures on lagged values of itself and the corresponding insider holding measure. For the second equation, the holding measure is regressed against lagged value of itself and the corresponding lagged value of risk. Signi cant coe cients of the cross-lagged parameters then indicate the direction of causality. Out tests provide an inconclusive result because neither of the cross-lagged parameters are signi cant.

8 1390 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383±1398 Table 1 Descriptive statistics a Variable 1987±1989 (N ˆ 450) 1992±1994 (N ˆ 450) Test statistics Pooled (N ˆ 900) Total risk (%) Mean Median Systematic risk (%) Mean ) Median ) Firm speci c risk (%) Mean Median Managerial holdings (%) Mean ) Median ) Total assets ($,000,000) Mean 10,714 16, ,463 Median KeeleyÕs Q Mean Median Outside blockholdings Mean Median ) Frequency of trading (%) Mean Median a This table shows the sample characteristics for our 900 observations or bank-years segregated from 1987 to 1989, from 1992 to 1994, and combined or pooled. Total risk is the standard deviation of daily returns expressed in percent for each bank with data taken from the CRSP tapes. Systematic risk is the di erence between total risk and rm speci c risk. Firm speci c risk is the standard deviation of the residuals expressed in percent from the single-factor market model for each rm. Managerial holdings are the aggregate holdings, in percent, of all o cers and directors of the bank. Total assets are the year-end book value of total assets. KeeleyÕs Q is the market value of common equity plus the book value of liabilities divided by the book value of assets. Outside Blockholdings is the percent of shares held by all una liated, ve-percent holders of the rmõs shares. Frequency of trading is the average daily volume of shares traded divided by shares outstanding. Test statistics are t-tests for di erence of means and Z-tests for di erence of medians. Signi cant p-values are listed to the right of each test statistics. * Signi cant at the 5% level. ** Signi cant at the 1% level. are larger in total assets (signi cant at the 1% level) and have stock that is traded more frequently (signi cant at the 1% level). Outside blockholdings are also higher in the 1992±1994 period (signi cant at the 1% level). Signi cant di erences in the mean and median values of KeeleyÕs Q observed in the 1987±1989 and 1992±1994 periods corroborate other (anecdotal) evidence suggesting that bank health improved in the early 1990s. While the di erences may not seem great, the inclusion of the book value of liabilities and the book value of assets in KeeleyÕs Q measure mitigates large uctuations over time. Another perspective on changes in market valuation of banking rms is given by the ratio of the market value of equity to the book value of equity. The mean values of the sample banksõ market-to-book ratio is in the 1992±1994 period as compared to in the 1987±1989 period. This di erence is statistically signi cant at the 1% level. Similar di erences are observed

9 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383± for the median values of the market-to-book ratios (1.360 vs. 1.10, statistically signi cant at the 1% level) Regression results Table 2 presents the results from the TSCS regression of each of the market risk factors on managerial holdings and other hypothesized determinants of risk. 10 Separate risk estimates are employed for the 1987±1989 and 1992± 1994 periods. The equations are all statistically signi cant at the 1% level, as judged by their F-values, with adjusted R 2 s that vary between 17.99% and 50%. The explanatory power of the equations is generally greater for the latter period. We present in Table 2 both the linear and piecewise linear versions of our model, since each provides separate insights into the in uence of equity holdings on managerial risk-taking. Insider holdings are statistically signi cant in the linear model for each of the equations. Increased managerial shareholdings are associated with greater total and rm speci c risk in the 1987± 1989 period. However, increased managerial shareholdings are associated with a decline in total and rm speci c risk in the 1992±1994 period following regulatory constraints imposed by the Financial Institutions Reform, Recovery, and Enforcement Act (FIRREA) in 1989 and the Federal Deposit Insurance Corporation Improvement Act (FDICIA) in As shown in the piecewise regressions, the increased risk associated with larger managerial holdings in the 1987±1989 period does not become signi cant until holdings exceed the 25% level, though the positive relationship for 5± 25% ownership is signi cant at the 10% level. Apparently, managers incentives to exploit underpriced deposit insurance in the earlier period did not overcome their risk aversion associated with a desire to retain their employment until their shareholdings became substantial. 11 In contrast, the negative association between insider holdings and risk in the latter period takes place at very low (<5%) level of managerial shareholdings. 12; 13 In contrast, there appears to be 10 All of the analyses reported in Table 2 are also done with the entire sample without screening for infrequent trading. While the results are less robust, the basic conclusions remain unchanged. 11 These results are consistent with those of Cebenoyen et al. (1999) who nd, using accounting data for savings and loans, that ``manager-owned thrifts have greater risk, but only at levels of ownership above a threshold of about 23±28% ownership'' (p. 51). 12 We explored the possibility that the negative association between managerial ownership and risk taking is driven by low franchise banks by dividing the sample into high and low franchise banks and replicating the regression analysis. We are unable to detect any meaningful di erences in the relationships based upon bank franchise values. 13 Decreases in the risk taking preference of managers in the 1992±1994 period may also re ect increased stock option value of managers associated with higher franchise value.

10 1392 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383±1398 Table 2 Time series, cross section regression results of capital market measures on managerial holdings a Total risk Linear Piecewise linear 1987± ± ± ±1994 Panel A: Regression with total risk as the dependent variable. Total risk is the fractional standard deviation of daily returns over a one-year holding period Intercept (5.04) (12.87) (5.20) (13.26) Insider holding ) ± ± (3.50) ()1.78) Hold (0±5%) ± ± ) ) ()0.29) ()2.36) Hold (5±25%) ± ± ) (1.89) ()0.23) Hold (>25%) ± ± ) (2.02) ()0.79) KeeleyÕs Q ) ) ) ) ()2.70) ()8.21) ()2.67) ()8.07) Outside blockholdings ) ) ) ) ()0.55) ()0.64) ()0.53) ()0.65) ln (Total assets) ) ) ) ) ()3.62) ()9.80) ()3.92) ()9.49) Frequency (3.25) (2.35) (2.92) (2.38) F-statistic Adj. R N Chow tests or F-tests: Holdings 1987±1989 ˆ Holdings 1992±1994 : Hold (0±5%) 1987±1989 ˆ Hold (0±5%) 1992±1994 : 1.13 Hold (5±25%) 1987±1989 ˆ Hold (5±25%) 1992±1994 : 9.74 Hold (>25%) 1987±1989 ˆ Hold (>25%) 1992±1994 : 12.93

11 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383± Systematic risk Panel B: Regression with systematic risk as the dependent variable. Systematic risk is the di erence between total risk and rm speci c risk Intercept ) ) ) ) ()2.76) ()1.81) ()2.49) ()1.39) Insider holding ) ) ± ± ()0.96) ()0.66) Hold (0±5%) ± ± ) ) ()0.24) ()0.14) Hold (5±25%) ± ± ) ) ()0.31) ()1.73) Hold (>25%) ± ± ) ()0.54) (1.67) KeeleyÕs Q ) ) ()0.40) ()1.58) (0.37) (1.73) Outside blockholdings ) ) ) ) ()0.31) ()1.46) ()0.32) ()1.62) ln (Total assets) (9.14) (9.91) (7.09) (7.83) Frequency (6.71) (13.98) (6.58) (13.70) F-statistic Adj. R N Chow tests or F-tests: Holdings 1987±1989 ˆ Holdings 1992±1994 : 0.29 Hold (0±5%) 1987±1989 ˆ Hold (0±5%) 1992±1994 : 1.36 Hold (5±25%) 1987±1989 ˆ Hold (5±25%) 1992±1994 : 0.05 Hold (>25%) 1987±1989 ˆ Hold (>25%) 1992±1994 : 0.55 Firm speci c risk Panel C: Regression with rm speci c risk as the dependent variable. Firm speci c risk is the fractional standard deviation of the residuals from the two-index market model Intercept (5.48) (13.41) (5.52) (13.27) Insider Holding ) ± ± (3.59) ()1.74) Hold (0±5%) ± ± ) )0.1139

12 1394 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383±1398 Table 2 (Continued) Total risk Linear Piecewise linear 1987± ± ± ±1994 ()0.85) ()2.31) Hold (5±25%) ± ± ) (1.97) ()0.16) Hold (>25%) ± ± ) (1.99) ()0.85) KeeleyÕs Q ) ) ) ) ()2.83) ()8.14) ()2.80) ()7.99) Outside blockholdings ) ) ) ) ()0.54) ()0.62) ()0.52) ()0.62) ln (Total assets) ) ) ) ) ()4.49) ()10.07) ()4.53) ()9.70) Frequency (2.40) (1.98) (2.12) (1.97) F-statistic Adj. R N Chow tests or F-tests: Holdings 1987±1989 ˆ Holdings 1992±1994 : Hold (0±5%) 1987±1989 ˆ Hold (0±5%) 1992±1994 : 1.55 Hold (5±25%) 1987±1989 ˆ Hold (5±25%) 1992±1994 : Hold (>25%) 1987±1989 ˆ Hold (>25%) 1992±1994 : a This table presents regression estimates of bank risk on managerial holdings, bank health, and several control variables. Holdings are the aggregate fraction of shares held by o cers and directors of the bank. KeeleyÕs Q is the market value of equity plus the book value of liabilities divided by the book value of assets. Outside blockholdings is the percent of shares held by all una liated, ve-percent holders of the rmõs shares. Total assets are the year-end book value of total assets for each rm. Frequency is the average daily volume of shares traded divided by total shares outstanding. t-statistics are given in parentheses and p-values to the right of each signi cant estimate. Chow test is the F-statistic result from a test examining whether the risk and holdings relation is signi cantly di erent between the 1987±1989 and 1992±1994 periods. * Signi cant at the 1% level. *** Signi cant at the 5% level. ** Signi cant at the 10% level.

13 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383± no statistically signi cant relationship between ownership and systematic risk in either of the two periods. We perform a Chow or F-test to determine whether the change in the risk± holdings relationship is signi cantly di erent between the 1987±1989 and 1992±1994 periods. Speci cally, the unrestricted model allows the coe cient value of each independent variable to vary between the two subperiods. The restricted model, however, allows each parameter value to vary between the two periods with the exception of holdings; the holdings term has the same coe cient value across the two periods. The resulting null hypothesis is that the coe cient of holdings in the 1987±1989 period is equal to the coe cient of holdings in the 1992±1994 period. See Johnston (1984) for additional details on tests of structural change. The results of the tests strongly reject the null hypothesis and indicate that the relationship between managerial holdings and total and rm speci c risk signi cantly changed from 1987±1989 to 1992±1994. The coe cient of KeeleyÕs Q is negative and statistically signi cant, indicating that high franchise value banks have less total and rm speci c risk. Indeed, Keeley's Q is statistically signi cant at the 1% level in its association with total and rm speci c risk in each of the regression equations. Two other variables that appear to in uence bank risk are size and trading frequency. Larger banks appear to have less total and rm speci c risk in both time periods, though more systematic risk. These results suggest that portfolio diversi cation and managerial expertise characteristics associated with bank size more than o set the greater potential total and rm speci c risk of individual activities at larger banks. Banks with more frequently traded stock also appear to have higher levels of risk, ceteris paribus. Outside blockholdings do not appear to in uence managerial risk taking Further analysis Our results suggest that total and rm speci c risk and managerial holdings are positively related in the 1987±1989 period and negatively related in the 1992±1994 period. This change in the managerial holdings to risk relationship could re ect relatively constant risk and changes in managerial holdings. However, our analysis indicates that managerial holdings change little on an individual bank basis. Managerial holdings drop by 1.95% per bank from 1987 to Moreover, only 2.5% of the sample experience a drop in holdings from greater than 20% in 1987 to less than 10% in We have no banks in the sample that increase holdings from less than 3% in 1987 to over 6% in Thus, managers that are characterized as entrenched in the 1987±1989 period are still characterized as entrenched in the 1992±1994 period and unentrenched managers also maintain this characterization throughout the period.

14 1396 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383±1398 We nd that the change in the relation between managerial holdings and total and rm speci c risk apparently began in Regression analysis on a year-by-year basis suggests that the positive relation between risk and managerial ownership holds for 1987, 1988, and In addition, the inverse relation between risk and holdings occurs only in 1992, 1993, and However, we are unable to detect any statistically signi cant relation for 1990 or 1991, suggesting that managerial behavior started to change during this period. We further investigated the change in risk taking propensity by applying the 1987±1989 coe cients of our basic model to the franchise value and other exogenous variables actually observed in the 1992±1994 period. Given the prior discussion about changing risk taking behavior, we would expect that the predicted risk levels in the 1992±1994 period using the 1987±1989 coe cients would be higher than the actual levels observed. This was the case for both total and rm speci c risk. Indeed, the predicted level of total risk was 46% above the actual level while the predicted level of rm speci c risk was 27% higher. In contrast, the predicted level of systematic risk was slightly lower. Our results thus far suggest that the relation between managerial shareholdings and total risk changes between the pre- and post periods. However, board characteristics and structure di er substantially among rms, suggesting that inside director preference for risk may be substantially di erent from that of an independent director. As a result, we randomly sampled 50 rms from our sample and gathered detailed board of director data from annual proxy statements. We classi ed board members following the taxonomy of Brickley et al. (1994). Inside directors are employees, retired employees, and family members of an executive within the rm. Examples of a liated directors are attorneys, investment bankers, and consultants. Lastly, independent directors are directors whose only relationship to the rm is their directorship. We then estimated again our regression model from Table 2 with this sample of 50 banking organizations with the insider holdings variable now de ned as the shares held by inside and a liated directors. The results (available from the authors) are not materially di erent, with the relationship between managerial holdings and risk remaining positive in the 1987±1989 period and again becoming negative in the 1992±1994 period. For total risk, for example, the insider holdings coe cient for 1987±1989 is in Table 2, signi cant at the 1% level, and , signi cant at the 5% level for the rede ned insider holdings variable. For the 1992±1994 period, Table 2 shows an insider coe cient of )0.120, signi cant at the 1% level, while the rede ned insider coe cient for the 50-bank sample is ) For the piecewise regressions, the patterns are again similar, though the positive and statistically signi cant insider holdings coe cient in 1987±1989 occurs in the 5±25% ownership range (vs. > 25% in Table 2) and the insider holdings coe cients in the 1992±1994 period are not statistically signi cant.

15 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383± Conclusions and implications Our results suggest that managerial shareholdings do in uence bank total and rm speci c risk. However, the association is di erent in the 1987±1989 period when banks were less regulated and under nancial stress than in the 1992±1994 period following legislation (FIRREA and FIDICIA) designed to restrict risk-taking and after the industry returned to pro tability. Managers with substantial equity holdings took more risk in the 1987±1989 period. In contrast, managers with substantial equity position took less risk in the 1992± 1994 period in response to regulatory changes designed to reduce incentives for risk-taking and improvements in the nancial health of the banking industry. Evidence from the other variables also provides potentially important insights into bank risk-taking. Franchise value appears to be an important determinant of bank risk-taking: banks with high franchise values are less likely to take risk than banks with low franchise value. In contrast, outside blockholders have, at best, limited in uences on bank risk taking. While it does appear that managerial shareholdings do in uence bank risk taking and that this relationship is di erent in the 1992±1994 period then in the 1987±1989 period, identifying the causes of the change must be done in a more tentative way. Large increases in franchise values in the 1990s suggest reduced incentives for banks to take risk, as do signi cant changes in the bank regulatory regime designed to reduce bank risk taking. Further evidence on whether the enhanced regulatory restrictions have been su cient to control management incentives to take risk will be provided the next time that bank franchise values decline substantially. References Anderson, R., Lee, D.S., Ownership studies: The data source does matter. Journal of Financial and Quantitative Analysis 32, 311±329. Bateman, T., Strasser, S., A cross-lagged regression test of the relationships between job tension and employee satisfaction. Journal of Applied Psychology 68, 439±445. Brickley, J., Coles, J., Terry, R., Outside directors and the adoption of poison pills. Journal of Financial Economics 35, 371±390. Cebenoyen, A., Cooperman, E., Register, C., Ownership structure, charter value, and risk taking behavior for thrifts. Financial Management 28, 43±60. Chen, C., Steiner, T., Whyte, A., Risk-taking behavior and management ownership in depositors institutions. The Journal of Financial Research 20, 1±16. Davidson, W., Rangan, N., Rosenstein, S., Regulation and systematic risk in the electric utility industry: A test of the bu ering hypothesis. The Financial Review 32, 163±184. Demsetz, R., Strahan, P., Diversi cation, size, and risk at bank holding companies. Journal of Money, Credit and Banking 29, 300±313. Demsetz, R., Saidenberg, M. and Strahan, P., Agency problems and risk taking at banks. Working paper, Federal Reserve Bank of New York.

16 1398 R.C. Anderson, D.R. Fraser / Journal of Banking & Finance 24 (2000) 1383±1398 Esty, B., Ownership concentration and risk-taking in the S&L Industry. Working paper, Harvard Business School. Flannery, M., James, C., The e ect of interest rate changes on the common stock returns of nancial institutions. Journal of Finance 39, 1141±1153. Fuller, W., Battese, G., Estimation of linear models with crossed-error structure. Journal of Econometrics 2, 67±68. Giliberto, M., Interest-rate sensitivity in the common stocks of nancial intermediaries: A methodological note. Journal of Financial and Quantitative Analysis 20, 123±126. Gorton, G., Rosen, R., Corporate control, portfolio choice, and the decline of banking. Journal of Finance 50, 1377±1420. Johnston, J., Econometric Methods. New York, McGraw-Hill. Keeley, M., Deposit insurance, risk, and market power in banking. American Economic Review 80, 1184±1200. Kmenta, J., Elements of Econometrics. Prentice-Hall, Englewood Cli s, NJ. McConnell, J., Servaes, H., Additional evidence on equity ownership and corporate value. Journal of Financial Economics 27, 595±612. Merton, R., An analytical derivative of the cost of deposit insurance and loan guarantees: An application of modern option pricing theory. Journal of Banking and Finance 1, 3±11. Saunders, A., Strock, E., Travlos, N., Ownership structure, deregulation, and bank risk taking. The Journal of Finance 45, 643±654. Unal, H., Kane, E., Two approaches to assessing the interest-rate sensitivity of depositoryinstitutions equity returns. Research in Finance 7, 113±138.