Corporate Governance of Banks and Financial Stability: International Evidence 1

Corporate Governance of Banks and Financial Stability: International Evidence 1 Deniz Anginer Virginia Tech, Pamplin College of Business Asli Demirguc-Kunt Word Bank Harry Huizinga Tilburg University and CEPR Kebin Ma Warwick Business School This draft: December 2015 Abstract: This paper finds that shareholder-friendly corporate governance is positively associated with bank insolvency risk and a bank s contribution to financial-sector systemic risk for an international sample of banks over the 2004-2008 period. Banks are special in that good corporate governance increases bank risk relatively more for banks that are large and located in countries with generous financial safety nets, as banks aim to exploit the financial safety net compared to non-financial firms. Good corporate governance is specifically associated with higher asset volatility, more non-performing loans, and a lower tangible capital ratio. Furthermore, good corporate governance is associated with more bank risk taking at times of rapid economic expansion. These results underline the importance of the financial safety net and too-big-to-fail policies in encouraging excessive risk-taking by banks. Key words: Corporate governance; Bank insolvency; Systemic risk JEL Classification: G21, M21 1 Anginer: Danginer@gmail.com; Demirguc-Kunt: Ademirguckunt@worldbank.org; Huizinga: Huizinga@uvt.nl; Ma: Kebin.ma@gmail.com. We thank an anonymous referee and participants at the 18th annual International Banking Conference at the Federal Reserve Bank of Chicago for useful comments and suggestions. This paper s findings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. 1

1. Introduction Corporate managers may be more risk averse than shareholders, as corporate bankruptcy generally causes managers to lose their jobs, part of their personal wealth and their reputation. Good corporate governance - promoting the interests of shareholders may serve to counteract managers bias against risk taking. Consistent with this notion, John, Litov, and Yeung (2008) find that better corporate governance in the form of stronger investor protection brings about increased risk-taking and higher growth for an international sample of non-financial firms. For financial institutions, the calculus regarding the optimal degree of risk taking is different compared to non-financial firms, as banks tend to be supported by the financial safety net if they become distressed. Banks, in particular, benefit from deposit insurance and may receive generous public support to prevent their failures. The financial safety net provides banks with an incentive to take on excessive risks in order to increase the value of these benefits. Hence, the association between risk taking and corporate governance may be different in the case of banks (see for example, Laeven, 2013). Large banks may benefit relatively more from the financial safety net, as they may be deemed too big to fail by regulators (see Acharya, Anginer and Warburton, 2014, and Bertay, Demirguc-Kunt, and Huizinga, 2013). For this reason, shareholder-friendly corporate governance may increase risk taking more in the case of large banks. Similarly, banks may gain more from increasing risk if the financial safety net is more generous. Therefore, shareholder-friendly corporate governance is likely to increase bank risk taking especially in countries with generous financial safety nets. 2

This paper empirically examines the relationships between bank risk and corporate governance for an international sample of banks from 22 countries over the 2004-2008 period. We distinguish between a bank s stand-alone insolvency risk, and a bank s contribution to systemic risk to gauge the threat a bank poses to overall financial stability. Specifically, an individual bank s insolvency risk is proxied by the Z-score and the distance to default, while a bank s contribution to financial-sector systemic risk is captured by the marginal expected shortfall (MES) variable proposed by Acharya, Engle and Richardson (2012), and alternatively by the CoVaR proposed by Adrian and Brunnermeier (2015). Following Aggarwal et al. (2009), we use an overall index of shareholder-friendly corporate governance dealing with board composition, compensation, auditing and takeover-related issues based on data from Institutional Shareholder Services. We find that a bank s insolvency risk and its contribution to systemic risk are both positively associated with the overall index of shareholder-friendly corporate governance. This positive relationship between bank risk and shareholder-friendly corporate governance is robust to instrumental variables estimation, where a bank s corporate governance is instrumented by the annual country-mean value of the corporate governance variable for all non-financial firms. Furthermore, we find that bank insolvency and systemic risks are positively associated with shareholder-friendly corporate governance especially for large banks and for banks located in countries with generous financial safety nets. These results are consistent with the view that shareholder-friendly corporate governance increases bank 3

risk more if the additional bank risk increases a bank s contingent claim on the financial safety net. Going beyond measures of insolvency and systemic risk, we also examine the channels through which a better governed bank takes on more risk. On the asset side, we find that asset volatility derived from Merton s option pricing model, and the share of non-performing loans in total loans, are positively related to good corporate governance. On the liability side, we find a negative relationship between a bank s tangible equity ratio and the corporate governance index. Our results further suggest that corporate governance affects the relationship between bank risk taking and the business cycle. Specifically, we find that banks with good corporate governance report less non-performing loans at times of high economic growth. This suggests that banks with good corporate governance bias their reporting of non-performing loans downward during periods of rapid economic expansion so as to preserve capital to be able to continue to increase credit. In addition, we see that banks with good corporate governance have relatively low tangible capital ratios when the economy grows rapidly. Non-financial firms are less likely to benefit from the financial safety net compared to banks. This suggests that the relationship between shareholder-friendly corporate governance and firm insolvency risk should be weaker for non-banks than for banks. We expand our sample of international banks to include non-bank corporations, and find that the positive relationship between insolvency risk and shareholder-friendly corporate governance is significantly more pronounced for banks than for non-banks, especially when we compare large banks to large non-banks. 4

Finally, to further alleviate endogeneity concerns, we examine the impact of new governance rules mandating greater board independence, that were adopted in 2003 by stock exchanges in the US, on bank insolvency risk. The new rules increased board independence for some banks but not others, as only banks that were initially noncompliant with the new rules had to increase their board independence. We find that insolvency risk increased relatively more at banks that were directly affected by the reforms. This is strong evidence that shareholder-friendly corporate governance causes higher bank insolvency risk. Our study fits in an emerging literature that has addressed the impact of corporate governance on bank risk taking. 2 Pathan (2009) finds that small boards and boards that are not controlled by the CEO lead to additional bank risk as reflected in market measures of risk and the Z-score for a sample of US bank holding companies over the 1997-2004 period. Chen, Steiner, and Whyte (2006), find a positive impact of optionbased executive compensation and wealth on market measures of risk for a sample of US commercial banks during the period 1992-2000. DeYoung, Peng, and Yan (2013) find that CEO risk-taking incentives lead to riskier business policy decisions (with respect to loans to businesses, non-interest based banking activities, and investment in mortgagebacked securities) at US commercial banks over the 1994-2006 period, especially for larger banks and in the second half of the sample period after deregulation in the banking sector. Calomiris and Carlson (2014) examine bank ownership and risk-taking at US 2 Recent surveys are offered by Becht, Bolton, and Roell (2011) and Mehran, Morrison and Shapiro (2012). Stulz (2015) discusses how corporate governance and risk management should be designed to ensure that banks only take good risks that add value. 5

banks in the 1890s, and find that higher managerial ownership is associated with lower bank default risk. Several papers have also examined how banks with different corporate governance regimes fared during the crisis, with mixed results. Berger, Imbierowicz, and Rauch (2015) find that high shareholdings of lower-level management imply a substantially higher probability of bank failure for US commercial banks over the 2007-2010 period. Fahlenbrach and Stulz (2011) find some evidence that US banks with CEOs whose incentives were better aligned with the interests of shareholders in 2006 had worse share price performance during the subsequent crisis. Ellul and Yerramilli (2013) report that US bank holding companies that had a strong and independent risk management function in place before the onset of the financial crisis fared better in terms of operating and stock performance during the crisis. However, multi-country studies of bank corporate governance and risk taking are relatively scarce. Laeven and Levine (2009) examine the relationship between bank ownership and bank risk taking for an international sample of banks. They find that stronger cash flow rights of large owners are associated with greater bank risk, consistent with the hypothesis that bank shareholders favor risk-taking as compared to managers and creditors. These authors also consider the interaction between bank regulation and ownership, finding that deposit insurance is associated with an increase in risk only when the bank has a large equity holder. More recently, using international data, Erkens, Hung, and Matos (2012) find that financial institutions with more independent boards and higher institutional ownership experienced worse stock returns during the global financial crisis. 6

Our contribution to this literature is three-fold. First, we use international banklevel data from 22 countries to study the association between bank individual and systemic risk measures and corporate governance, adding to a literature that has mostly relied on US data. Using multi-country data enables us to exploit differences across country safety nets to study how the relationship between bank risk-taking and corporate governance varies with the generosity and credibility of the safety net and banks ability to engage in risk-shifting. Second, this paper is the first to study the relationship between a bank s corporate governance and its contribution to systemic risk. Third, we are able to study how banks increase their risk-taking with more shareholder friendly corporate governance, identifying increased asset risk, more non-performing loans, and reduced capitalization as potential channels. Overall, our findings on the interaction of bank-level corporate governance variables and the financial safety net have important implications for corporate governance reforms in the banking sector, as policy makers question the extent to which governance failures have contributed to excessive risk taking and financial instability. In particular, our results suggest that one has to be cautious to call for better corporate governance at banks as long as generous financial safety nets and too-big-to-fail policies are in place, as this may actually induce banks to take on more risk with potentially negative repercussions for the stability of the financial system. In the remainder, section 2 discusses the data. We present the empirical results in section 3. We start with an analysis of the relationships between corporate governance and bank insolvency and systemic risks. Then we consider the channels on the assets and liabilities sides of the bank s balance sheet through which corporate governance 7

affects overall bank risk, and the relationship between corporate governance and bank risk taking over the business cycle. Next, we consider the relationship between corporate governance and insolvency risk for a sample of firms including both banks and nonbanks. Finally, we examine the impact of reform towards greater board independence in the US on bank insolvency risk. Section 4 concludes with policy implications. 2. The data In this study, we relate measures of bank insolvency and systemic risk to summary indicators of corporate governance for an international sample of banks over the 2004-2008 period. In addition, we consider the relationship between corporate governance and insolvency risk for an international sample of firms comprising banks as well as non-bank corporations. We describe the samples of banks only, and of both banks and non-banks in turn. 2.1 The sample of banks Accounting and market data necessary to construct our bank risk measures are taken from Bankscope of Bureau Van Dijk, CRSP and Compustat Global. Data on bank corporate governance are from the Corporate Governance Quotient database assembled by Institutional Shareholder Services (ISS). We examine two measures of individual bank insolvency risk. First, the Z-score represents the number of standard deviations that a bank s rate of return on assets has to fall for the bank to become insolvent. The Z-score is constructed as the sum of the rate of return on assets and the equity to assets ratio divided by the standard deviation of the return on assets (see Table A1 in the Appendix for variable definitions and data sources). 8

A higher Z-score signals that a bank has lower insolvency risk. We calculate a Z-score for a bank, if at least three annual observations of its rate of return on assets are available. Our second insolvency risk variable, the distance to default, measures the difference between the asset value of the bank and the face value of its debt, scaled by the standard deviation of the bank s asset value (see Campbell, Hilscher and Szilagyi, 2008, p. 2899). The distance to default variable is computed as an annual average of monthly values (see the Appendix for details on how the distance to default is estimated). Following the recent literature, we use two variables to measure a bank s contribution to systemic risk. First, following Acharya, Engle and Richardson (2012), marginal expected shortfall (MES) is the average bank stock return when the stock market return as a whole is in the lower tail of its distribution. Specifically, MES is constructed as the average bank stock return when the market return is in its lowest 5% bracket in a given year. Second, following Adrian and Brunnermeier (2015), CoVaR is defined as the value at risk of the overall financial system conditional on a bank being in distress and the value of risk of the overall financial system conditional on a bank being in a normal state, where a bank distress and normal states are defined as a bank s stock return being at the 5 th and 50 th percentile in a given year respectively. Lower values for MES and CoVaR signal higher bank contributions to systemic risk. 3 Next, we construct several variables to capture different aspects of a bank s overall risk strategy. These variables reflect a bank s asset allocation and income mix strategies, its capitalization and funding strategies, and its overall asset growth strategy. 3 See Appendix A1 for details on how MES and CoVaR are constructed. 9

To start, asset volatility is the annualized standard deviation of the asset return computed from the Merton s option pricing model. The asset volatility variable has a mean of 0.05. Asset risk weight is an indicator of the average riskiness of a bank s assets, and is computed as the ratio of risk-weighted assets to total assets, using the risk weights as defined in the Basle capital adequacy framework. Lower values of asset risk weight indicate less risky assets. The asset risk weight variable has a mean of 0.70. Bank loans are generally riskier than other investments, such as holdings of government securities. We use the loans variable, computed as the ratio of loans to total assets, as a proxy for asset risk. The loans variable has a mean of 0.70. The non-performing loans variable, computed as the ratio of non-performing loans to total loans, is an index of loan quality. On average, 1.4% of loans are nonperforming. A bank s asset allocation affects the composition of its income which is generally in the form of interest income, fees, commissions and trading income. The non-interest income variable, constructed as the ratio of net interest income to total operating income, is an index of the riskiness of a bank s income. It has a mean of 0.29. On the liability side of a bank s balance sheet, we consider three alternative capitalization ratios. First, Tier 1 capital is a regulatory capital ratio constructed as Tier 1 capital divided by risk-weighted assets. The Tier 1 capital ratio has a mean of 11.1%. Second, regulatory capital is a broader regulatory capital ratio computed as the sum of Tier 1 capital and Tier 2 capital divided by risk-weighted assets with a mean of 13.0%. A bank s common equity can be divided into tangible common equity and non-tangible common equity. The latter category includes tax deferred assets and mortgage servicing 10

rights, which are capital categories with only limited loss absorption capacity. As a third capital ratio, we construct the tangible capital ratio as the ratio of tangible equity divided by tangible assets (i.e., total assets minus non-tangible assets). The tangible equity ratio has a mean of 7.3%. A bank s short-term funding comprises customer and other deposits and nondeposits such as short-term borrowing in the interbank market. The non-deposit funding variable, computed as the share of non-deposit, short-term funding in total short-term funding, is an index of the riskiness of a bank s short-term funding. It has a mean of 16.8%. High bank asset growth may signal higher bank risk, as a bank may only be able to grow fast by investing in riskier assets, for instance by lending to customers that represent higher risk. Our asset growth variable is the growth rate of total assets, with a mean value of 6.0%. In addition, banks with highly procyclical asset growth rates may be more risky, since such banks may be overly optimistic about asset quality at the peak of the business cycle. We construct the asset procyclicality variable as the correlation between a bank s asset growth rate and the GDP growth rate. The mean asset procyclicality variable is 0.13. Similarly, the lending procyclicality variable is the correlation between a bank s loan growth rate and the GDP growth rate with a mean of 0.20. In addition, we consider the correlations of a bank s non-performing loan rate and its tangible capital ratio with the GDP growth rate with means of -0.38 and 0.18, respectively. Our corporate governance variables are indices that summarize extensive information on detailed governance attributes that are indicative of increased power of 11

minority shareholders. We use the indices as formulated by Aggarwal et al. (2009) based on individual governance attributes assembled by Institutional Shareholder Services. The individual attributes are dummy variables that take on a value of 1 if the characteristic is relatively shareholder-friendly, and a value of zero otherwise. An overall index, called corporate governance, summarizes information on 44 attributes, and it is scaled to range between 0 and 1. In addition, there are four sub-indices, labeled board, compensation and ownership, auditing and takeover that summarize information on 25, 10, 3 and 6 attributes related to these various aspects of corporate governance, respectively. The takeover sub-index, for instance, has a higher score, if there are fewer corporate governance-related barriers to takeovers. A listing of the individual attributes that are represented by the overall index, and the four sub-indices, is provided in Table A2 in the Appendix. This detailed information on corporate governance is available for banks located in 22 countries. For the country coverage, see Table A3 in the Appendix. Overall corporate governance has become more shareholder-friendly over the 2003-2007 period. As seen in Figure 1, the average overall index increased from 0.58 in 2003 to 0.63 in 2007 for US banks, while it increased from 0.52 to 0.59 for non-us banks. These differential trends in corporate governance for US and non-us banks (and, therefore, also for individual banks) allow us to estimate relationships between bank risk variables and corporate governance in specifications that include bank fixed effects. The financial safety net variable is a summary measure of the strength of the financial safety net obtained through a principal components analysis of deposit insurance design features following Demirguc-Kunt and Detragiache (2002). Specifically, we collect data on deposit insurance characteristics in the year 2003 from 12

Demirguc-Kunt, Karacaovali and Laeven (2005), and construct financial safety net as the sum of four principal components derived from eight deposit insurance characteristics: (1) existing coverage of foreign currency deposits, (2) existing coverage of interbank deposits, (3) an absence of coinsurance, (4) coverage per depositor per bank per account, (5) existence of funding ex ante, (6) existence of funding by the government, (7) existence of a risk-insensitive insurance premium, and (8) the ratio of insurance coverage and deposits per capita. In each instance, a higher value for the deposit insurance feature suggests a more generous financial safety net and a greater potential to induce bank risk taking. In the analysis, we use several bank-level control variables. The assets variable, constructed as the log of a bank s total assets, proxies for the bank s absolute size. Larger banks may pursue riskier banking strategies, if they are considered to be too big to fail. In addition, the assets to GDP variable, or total bank assets divided by GDP, represents the bank s size relative to the national economy with a mean of 0.06. The overhead variable is constructed as overhead expenses divided by total assets. The average overhead variable is 0.03. Inefficient banks with large overhead expenses may choose relatively risky bank strategies to maintain a certain expected return on assets. Finally, the collateral variable is the ratio of assets that can be easily used as collateral divided by total assets. The collateral variable on average is 0.27. Banks with assets that can be used as collateral may find it easier to pursue risky banking strategies, as their financial costs may be less sensitive to overall bank risk. Finally, we include several macroeconomic and country-level institutional control variables. Inflation is the consumer price inflation rate. GDP growth is the rate of real 13

GDP growth. GDP per capita is GDP per capita in thousands of constant U.S. dollars. The variable restrict is a composite index of regulatory restrictions on bank activities from Barth et al. (2004). Specifically, it is an indicator of the degree to which banks face regulatory restrictions on their activities in securities markets, insurance, real estate, and their ownership of shares in non-financial firms. Capital stringency is an index of regulatory oversight of bank capital, summarizing information about the nature and the magnitude of bank capital requirements, with higher values indicating greater stringency. Official is an index of the power of the commercial bank supervisory agency to undertake specific actions to prevent and correct problems at a bank, with higher values indicating greater power. Diversification is an index of loan diversification guidelines imposed on banks. Finally, financial freedom is an index of financial market freedoms available from the Heritage Foundation. 2.2 The sample of banks and non-banks Summary statistics for the sample of banks and non-banks are provided in Panel A of Table 2. We use the same two risk measures, the Z-score and the distance to default, for both banks and non-bank corporates. The average Z-score and the average distance to default for the combined bank and corporate sample are 3.06 and 5.68, respectively. However, we use a different set of controls when we compare financial firms to nonfinancial firms, since some variables, such as the capital stringency variable, are only applicable to banks. Following the literature, we use the book-to-market ratio, the return on assets, and firm size proxied by the assets variable as control variables. Firms with lower market valuations may be closer to insolvency, and hence display higher insolvency risk. To reflect this, we use the Book-to-market variable, which is the book 14

value of total equity divided by the market value of total equity with a mean of 3.09. More profitable firms with a higher return on assets may have lower insolvency risk. ROA is the return on assets with a mean of 0.66%. Finally, larger firms, especially banks, may pursue riskier strategies, if they are deemed to big to fail. The average assets variable is 6.71. 3. Empirical results 3.1 Bank insolvency risk and corporate governance We begin our analysis by examining the baseline relationship between bank risk and corporate governance. In particular, we estimate the following panel model for an international sample of banks: Risk ijt = α + β 0 Governance ijt 1 + β 1 X ijt 1 + β 2 Z jt 1 + γ i + δ t + ϵ it (1) Riskijt is a measure of risk for bank i in country j at time t. We use four different measures of risk. Z-score and distance to default measure bank level risk, while MES and CoVar measure a bank s contribution to the risk of the system. Governanceijt is a corporate governance variable. As mentioned earlier, we use an overall governance index and its major subcomponents following Aggarwal et al. (2009). Xijt is a set of firm-level controls, and Zjt is a set of country-level controls as described in Section 2. Finally, we include firm fixed effects, γ i, and year fixed effects, δ t, to control for time-invariant firm level heterogeneity and macro shocks that affect all firms in a given year. For all governance measures we use, a higher value of the Governanceijt variable indicates that corporate governance better serves the interests of shareholders. Shareholders potentially stand to gain from higher bank risk to the extent that they can shift risk to debt holders 15

and the financial safety net. Lower values of all four bank risk measures, Z-score, distance to default, MES, and CoVaR, point towards higher risk. Accordingly, in the regression specified in equation (1), we expect to find that β 1 < 0. In the estimation, we cluster the errors at the bank level. All independent variables are lagged by one year to reduce endogeneity concerns. Table 3 reports the results. In Panel A, we report regression results that include the overall corporate governance variable, while in Panel B we report regression results that include the four corporate governance subindices that make up the overall index. In the Z-score regression in column 1 of Panel A, the overall corporate governance index has a negative coefficient of -0.832 that is significant at the 10% level, suggesting that more shareholder-friendly corporate governance increases bank insolvency risk. Among the controls, we find that the Z-score is negatively and significantly related to the assets and assets to GDP variables, indicating that larger absolute and relative bank size are associated with higher bank insolvency risk. This is consistent with the notion that larger banks take on more risk as they benefit from a too-big-to-fail status. The Z-score is positively and significantly related to the GDP growth variable, as bank insolvency risk may be reduced by economic growth. In addition, the Z-score is negatively and significantly related to GDP per capita. This result may reflect that banks in wealthier countries benefit from a more credible financial safety net, which induces them to take on more risk. Furthermore, the Z-score is positively and significantly related to the diversification variable, suggesting that guidelines promoting diversification contribute to bank safety. Similarly, the Z-score is positively and significantly related to the economic 16

freedom variable, as economic freedom may enable a bank to diversify into activities that reduce overall bank insolvency risk. In regression results reported in column 2, the dependent variable is distance to default. Otherwise, this regression is analogous to regression results reported in column 1. We obtain similar results. The overall corporate governance index has a negative coefficient that is significant at the 5% level, indicating that bank insolvency risk is positively related to shareholder-friendly corporate governance. In regressions 3 and 4, the dependent variables are MES and CoVaR which measure a bank s contribution to systemic risk. In both regressions, the estimated coefficients for the corporate governance variable are negative and significant at the 1% level, consistent with a positive relationship between shareholder-friendly corporate governance and a bank s contribution to systemic risk. Overall, the results of Panel A of Table 3 suggest that bank insolvency and systemic risks vary positively with shareholderfriendly corporate governance, consistent with our conjecture. In Panel B, we examine the sub-components of the overall governance index together in the same regression. We only report coefficients on the governance subindices to save space. The results are weaker when we consider individual components of governance. We find that the distance to default variable is negatively and significantly related to the board-related corporate governance subindex. The MES and CoVaR variables are negatively and significantly related to the compensation and ownership, and takeover subindices. As robustness checks, we used two alternative measures of bank risk and riskshifting: the fair value of the liability insurance implicit in the financial safety net (this is 17

the IPP variable explained in Appendix A1), and average interest expense calculated as interest expense divided by average interest-bearing liabilities. In unreported regressions similar to those of Table 3 Panel A, these two indices of bank risk and risk-shifting are positively and significantly related to the corporate governance variable. Bigger banks may be riskier, because they expect to receive a more generous treatment by bank regulators in case of insolvency because of their too-big-to-fail status. Hence, the positive relationship between bank risk and good corporate governance may be driven by the larger banks in the sample. To examine whether the relationship between bank risk and corporate governance depends on bank size, we include an interaction term between the corporate governance variable and the assets variable in the regressions of Panel A of Table 3. We report the results in Table 4. In regressions reported in columns 2-4, the coefficient on the variable that interacts the overall corporate governance index with size is negative and statistically significant. This is consistent with the notion that good corporate governance increases insolvency and systemic risks of especially large banks, consistent with greater risk-shifting incentives for larger banks on account of their too-big-to-fail status. To further investigate risk-shifting by larger banks, we also examine how the impact of corporate governance on bank risk varies with the strength of the safety net. Countries with strong financial safety nets are more likely to bail out distressed banks that are deemed too big to fail. Therefore, banks with shareholder-friendly corporate governance may be riskier if they are located in country with a strong financial safety net. To examine this, we estimate regressions that include the financial safety net variable as an index of the strength and generosity of the financial safety net protecting banks. 18

Specifically, in the regressions of Table 4 we include triple interactions of the financial safety net, corporate governance, and assets variables (and also double interactions of financial safety net with corporate governance and assets). We report the results in Table 5. In regressions reported in columns 1, 3 and 4, we see that the triple interaction term involving the financial safety net variable has significant negative coefficients. This is evidence that good corporate governance increases insolvency and systemic risks especially at banks that are large and located in countries with generous financial safety nets. 3.2 Endogeneity We recognize that corporate governance may, to some extent, be endogenously determined. For instance, a strong preference for risk on the part of a bank s shareholders may jointly give rise to both considerable bank risk taking and shareholder-friendly corporate governance. To alleviate concerns about endogeneity, we include bank fixed effects in all regressions in the paper, thereby controlling for any time-invariant unobservable bank characteristics that affect both bank corporate governance and bank risk. Going beyond this, we analyze the relationship between corporate governance and bank risk using an instrumental variables approach. In particular, we instrument for a bank s corporate governance variable by using the country and year average of this variable for all non-financial firms in the country. Such country-year averages are good instruments to consider, as a shock to a bank s risk is unlikely to affect the corporate governance of non-financial firms. Similar IV approaches were previously used by John, Litov, and Yeung (2008), Aggarwal et al. (2009), and Laeven and Levine (2009). The IV results, reported in Table 6, show negative and significant coefficients for the corporate 19

governance variable in the distance to default, MES, and CoVaR regressions in columns 2-4. These IV regressions thus provide additional evidence that shareholder-friendly corporate governance increased bank insolvency and systemic risks over the sample period covering the years 2004-2008. 3.3 Bank risk strategies A bank s Z-score and its distance to default are summary measures of bank insolvency risk that reflect a range of bank risk-related strategies associated with its asset allocation, income mix, and capitalization and funding strategy. Next, we consider the impact of corporate governance on a range of indicators that reflect a bank s broader risk strategy. To start, Table 7 reports results on the associations between a bank s asset and income strategies and indices of corporate governance. In regression 1, the asset volatility variable is positively and significantly related to the overall corporate governance index which indicates that more shareholder-friendly corporate governance is associated with more asset risk. In regressions 2 and 3, the asset risk weight and loans variables are positively related to the corporate governance variable, but these relationships are statistically insignificant. Regression 4 shows that the non-performing loan rate varies positively and significantly with the overall corporate governance variable, suggesting that banks with shareholder-friendly corporate governance make riskier loans that more frequently become non-performing. Finally, the non-interest income share is not significantly related to the corporate governance variable in regression 5. Overall, the results of Table 7 suggest that banks with shareholder-friendly corporate governance maintain risker asset portfolios as reflected in a higher asset return volatility and a higher non-performing loan rate. 20

Next, we consider whether corporate governance is associated with risky capitalization, funding and growth strategies. In regressions 1-3 of Table 8, the Tier 1 capital ratio, the regulatory capital ratio, and the tangible capital ratio are negatively related to the overall corporate governance index. In the case of the tangible capital ratio regression 3, the negative estimated relationship between capitalization and the corporate governance variable is statistically significant. This provides some evidence that bank capitalization varies negatively with shareholder-friendly corporate governance. 4 We do not find that the non-deposit funding variable, as an index of relatively risky non-deposit short-term funding, is significantly related to the corporate governance variable in regression 4. Finally, regression 5 shows a positive and insignificant relationship between the asset growth and corporate governance variables. Taken together, Tables 7 and 8 are consistent with the view that banks with shareholder-friendly corporate governance employ several strategies to increase bank risk, including higher asset return risk, higher non-performing loan rates, and lower capitalization. 3.4 Banking procyclicality Next, we consider how corporate governance affects the cyclicality of bank asset growth and other proxies of bank-risk taking. Banks with shareholder-friendly corporate governance may take on additional risk by expanding their balance sheet during economic booms. To test this, we relate the asset procyclicality variable, which is the correlation between bank asset growth and GDP growth, to the overall corporate governance index in regression 1 of Table 9. This yields an estimated coefficient that is 4 Anginer, Demirguc-Kunt, Huizinga, and Ma (2013) consider in detail how board-related and takeoverrelated corporate governance features and executive compensation affect capitalization strategies for an international sample of banks over the 2003-2011 period. 21

positive and insignificant. In regression 2, we relate a similar loan procyclicality variable, constructed as the correlation between bank loan growth and GDP, to the corporate governance variable, yielding a positive and insignificant coefficient. Next, we consider the correlation coefficient between the non-performing loan rate and GDP growth. Regression 3 shows that this measure of non-performing loans procyclicality is related negatively and significantly to the overall corporate governance variable, indicating that banks with more shareholder-friendly corporate governance report less non-performing loans during economic booms. This could reflect that these banks manipulate the reporting of non-performing loans downward during periods of high economic growth in order to preserve capital and continue to be able to expand credit. Finally, the tangible capital procyclicality variable is computed as the correlation between the tangible capital ratio and GDP growth. In regression 4, the tangible capital procyclicality variable is related negatively and significantly to the corporate governance variable. This is evidence that banks with more shareholder-friendly corporate governance maintain lower tangible capital ratios during economic upswings. Overall, the results of Table 9 suggest that banks with shareholder-friendly corporate governance tend to take more risks at the peak of the business cycle by maintaining relatively low non-performing loans and tangible capital ratios. 3.5 Additional analyses to address endogeneity In this section, we carry out two additional analyses to further alleviate potential endogeneity concerns. First, we show that the effect of shareholder-friendly corporate governance in increasing insolvency risk is significantly higher for banks compared to 22

non-financial firms. Second, we use an exogenous regulatory change in the US that increased board independence for some banks but not others, and show that insolvency risk increased significantly more for the banks affected by this regulatory change. Diverging interests regarding firm risk between shareholders and management exist at financial as well as non-financial firms. The evidence of section 3.1, however, suggests that shareholders of banks may be especially interested in additional firm risk, as banks in distress can potentially benefit from the financial safety net. This may imply that insolvency risk varies more positively with shareholder-friendly corporate governance for banks than for non-banks. Furthermore, we would expect that the shareholders of large banks are more interested in additional risk relative to the shareholders of non-banks of comparable size, as large banks are more likely to benefit from the financial safety net. To test these relationships, we relate the Z-score and the distance to default to the corporate governance variable for an international sample of firms that includes banks as well as non-banks. To distinguish the two types of firms, we construct a dummy variable called bank that takes on a value of one if a given firm is classified as a bank according to ISS industry codes. 5 In the analysis, we use the book-to-market, ROA, and assets variables as controls. We allow the coefficients for these control variables to be different for banks and non-banks by including interactions of each control variable with the bank dummy variable. In addition, an interaction term of the corporate governance and bank dummy variables is included to test for a differential impact of shareholder-friendly corporate governance on insolvency risk for banks and non-bank corporations. The results are reported in Table 10. 5 In particular, the bank dummy is set to one if the firm is in the banks or diversified financials industry categories. 23

In the Z-score regression 1, the corporate governance variable has a positive and significant coefficient, while the interaction of corporate governance with the bank dummy variable has a negative and significant coefficient. These results suggest that shareholder-friendly corporate governance increases insolvency risk at banks relatively more, which is consistent with the view that the shareholders of banks are relatively more interested in firm-level risk. In the distance to default regression 2, the interaction of corporate governance with the bank dummy variable similarly has a negative and significant coefficient. This is further evidence that insolvency risk varies more positively with shareholder-friendly corporate governance for banks than for non-banks. 6 In the Z- score and distance to default regressions 3 and 4, we include an additional triple interaction term of the corporate governance, bank dummy and assets variables. In both regressions, the triple interaction term has negative and significant coefficients, consistent with the view that the shareholders of large banks stand to gain relatively more from increasing risk on account of their banks too-big-to-fail status. Overall, the results of regressions 1-4 indicate that insolvency risk varies more positively with shareholderfriendly corporate governance at banks than at non-banks, especially in the case of large firms. These results are consistent with the view that shareholder-friendly corporate governance leads a bank to adopt a higher level of risk to serve the interests of shareholders. 6 As a robustness check, we replicated the analysis reported in columns 1-2 of Table 10 using US data on corporate governance for banks and non-banks that are available for a longer period. In particular, as corporate governance variables we used the G-Index introduced by Gompers, Ishii and Metric (2003) giving rise to a sample for the period 1990-2012, and alternatively the share of independent board members giving rise to sample for the period 2000-2012. For these samples, we obtain significant results that are similar to those reported in Table 10 (unreported). 24

In the remainder of this section, we consider new regulations introduced in 2003 by the NYSE and NASDAQ exchanges, requiring firms to have more than 50% independent directors, as an exogenous event affecting corporate governance. 7 The new regulations significantly increased the proportion of independent directors in some firms but not others. The firms were required to comply with these new rules starting in 2004. 8 We use the fact that some firms already had a majority of independent directors on their boards (and thus complied with the new rules) at the time they were adopted, while other firms had to increase the number of independent directors after the rules came into effect. We classify these banks as being affected by the reform. We then examine the change in our two risk measures for the affected and non-affected banks following the adoption of the reforms. 9 To do this, we create a Post dummy variable that takes on a value of one for the time period after the implementation of reforms starting in 2004, and we interact it with an Affected dummy variable, indicating whether a bank was affected by the reforms. In our sample, about 12% of banks were affected by the new rules, as seen in Panel B of Table 2 with summary statistics. 10 The interaction of the Post and Affected variables is included in Z-score and distance to default regressions to ascertain whether the forced increase in board independence on account of the new regulations caused an increase in bank insolvency risk. The results are reported as regressions 5 and 6 in Table 10. In both 7 Several papers have used the introduction of NYSE and NASDAQ rules requiring majority board independence, and the Sarbanes-Oxley regulations requiring majority independence in the audit committee as exogenous shocks to governance. See, for instance, Duchin, Matsusaka and Ozbas (2010), Chhaochharia and Grinstein (2009), Linck, Netter and Yang (2008), and Armstrong, Core and Guay (2014). 8 NYSE-listed and NASDAQ-listed firms were required to implement the new requirement by their first annual meeting occurring after January 15, 2004, but no later than October 31, 2004 and October 15, 2004, respectively. 9 We obtain similar results using an instrumental variables approach to changes in the share of independent directors from 2000 to 2004 as in Duchin, Matsusaka and Ozbas (2010) (unreported). 10 Consistent with Armstrong, Core and Guay (2014), we find that initially non-compliant firms had a significant increase in independent directors compared to initially compliant firms. 25

regressions, the interactions of Post and Affected variables have a negative and significant coefficient, which suggests a causal effect of greater board independence on bank insolvency risk. Overall, the evidence in this section shows that bank insolvency risk varies more positively with shareholder-friendly corporate governance for banks than for non-banks, and that this effect is more pronounced for larger banks compared to larger non-bank corporations. These results suggest a causal relationship between shareholder-friendly corporate governance and insolvency risk. Further, we find that exogenous increases in board independence of US banks following new regulations are positively related with increases in bank insolvency risk as strong evidence of a causal link between shareholder-friendly corporate governance and bank insolvency risk. 4. Conclusion This paper provides evidence that more shareholder-friendly corporate governance is associated with greater bank insolvency and systemic risks for an international sample of banks. These empirical relationships are robust to including bank fixed effects and instrumental variable estimation, alleviating endogeneity concerns. We further find that good corporate governance is associated with increased asset volatility, more non-performing loans, and a lower tangible equity ratio. Our findings that good corporate governance is associated with increased risk taking at financial firms are consistent with earlier research showing that better investor protection reduces excessive risk-avoidance at non-financial firms explained by the fact that managers earn private benefits from reducing risk. Banks, however, are special in 26

that they benefit from the financial safety net. The financial safety net provides banks with an incentive to take on too much risk, as banks receive financial support from the financial safety if they become distressed. This suggests that risk taking at banks is determined by the interaction of corporate governance regimes and the financial safety net. We find empirical support for this hypothesis by showing that good corporate governance increases bank risk taking especially for banks that are large and located in countries with generous financial safety nets. For these banks, more risk can be expected to increase their contingent claim on the financial safety net. Good corporate governance thus reinforces the tendency for banks to exploit the financial safety net, if they are in a position to do so. For an international sample of banks and non-banks, we find that shareholderfriendly corporate governance is more positively related to insolvency risk at banks than at non-banks, especially in the case of large firms. This is further evidence of the special nature of banks, giving rise to a stronger relationship between shareholder-friendly corporate governance and risk taking than in the case of non-bank corporations. For the case of the US, we further find that regulatory reform towards greater board independence in the US approved in 2003 increased bank insolvency risk at banks that were affected by this regulation, which is evidence indicating a causal link between shareholder-friendly corporate governance and bank risk taking. The interaction of corporate governance and the financial safety net in determining bank insolvency and systemic risks has important implications for public policy towards corporate governance at banks. In particular, the case for more shareholder-friendly corporate governance at banks is much weaker than in the case of 27