BANK CORPORATE GOVERNANCE AND REAL ESTATE LENDING DURING THE FINANCIAL CRISIS

BANK CORPORATE GOVERNANCE AND REAL ESTATE LENDING DURING THE FINANCIAL CRISIS Emilia Peni a,*, Stanley D. Smith b,**, Sami Vähämaa a,*** a University of Vaasa, Department of Accounting and Finance b University of Central Florida, Department of Finance May 2, 2012 Abstract This paper examines the effects of bank corporate governance on real estate lending and loan losses during the financial crisis. The results indicate that banks with stronger corporate governance mechanisms generally had higher profitability during the period 2006 2009. Our findings on the effects of corporate governance on real estate lending performance are mixed and depend on the definition of the crisis period. Although banks with stronger governance practices had a lower amount of real estate loan losses during 2006 2008, our results also show that these banks experienced significantly larger losses in 2009. Finally, we document that banks with weaker corporate governance decreased their higher exposure to real estate loans after the meltdown of the real estate market. JEL classification: G01, G21, G30, R30 Keywords: corporate governance, bank performance, real estate lending, real estate losses, financial crisis * Address: University of Vaasa, Department of Accounting and Finance, P.O. Box 700, FI-65101 Vaasa, Finland; E-mail address: epeni@uwasa.fi ** Corresponding author. Address: University of Central Florida, Department of Finance, College of Business Administration, P.O. Box 161400, Orlando, FL, 32816-1400, USA; E-mail address: ssmith@bus.ucf.edu *** Address: University of Vaasa, Department of Accounting and Finance, P.O. Box 700, FI-65101 Vaasa, Finland; Tel. +358 6 324 8197; Fax: +358 6 324 8344; E-mail address: sami@uwasa.fi

1 1. Introduction This paper examines the effects of corporate governance on real estate lending and loan losses around the financial crisis and the meltdown of the real estate market. In particular, we aim to assess whether banks with stronger corporate governance mechanisms were associated with less severe real estate loan problems during the crisis. Given that the recent financial turmoil is to a large extent attributable to excessive risk taking by banks, particularly in terms of real estate lending, we consider the crisis to provide an expedient setting to examine whether effective governance practices enhance the performance and lending behavior of banks during periods of severe market stress. In essence, corporate governance is the mechanism for addressing agency problems and controlling risk within the firm. In the aftermath of the financial crisis, banking supervisors, central banks, and other authorities have argued that flaws in corporate governance practices may have had a significant role in the excessive risk taking of financial institutions and, consequently, on the development of the financial crisis (see e.g., Basel Committee on Banking Supervision, 2010; Board of Governors of the Federal Reserve System, 2010a-b; OECD 2010). Therefore, it is of paramount importance to empirically address the potential implications of corporate governance on bank performance in the midst of the recent crisis. It is widely acknowledged that the most important factor in the variability of bank performance and the risk of failure is loan losses (e.g. English, 2002). Given that bank performance was severely affected by real estate loan problems and declining property prices during the market turmoil, we examine the influence of corporate governance on real estate loan losses, and furthermore, on different types of real estate loans (residential, commercial, and construction and land

2 development loans). If strong corporate governance constrains inordinate risk taking in real estate lending, we should observe a negative relationship between real estate loan losses and the strength of governance mechanisms amidst the financial crisis. We focus on real estate lending behavior of banks for three main reasons. First, it is important to empirically analyze the association between corporate governance on real estate lending around the financial crisis because excessive risk taking in real estate lending is often considered among the main factors contributing to the development of the crisis (Guynn, 2010). Second, real estate loans are by far the largest loan category in the loan portfolios of most banks. For large U.S. bank holding companies, real estate loans accounted for approximately 60 % of total loans and for about 45 % of total assets during the crisis. Third, although the financial crisis had a strong adverse impact on all loan categories, the increase in real estate loan losses was substantially higher than the increase in other types of loans. While losses on consumer loans and commercial and industrial loans increased by 160 % and 659 %, respectively, from 2007 to 2009, the real estate loan losses of large banks surged by 1,122 % in the midst of the turmoil. 1 Therefore, it can be argued, at least ex post, that banks with larger real estate loan portfolios were associated with higher risk. 2 This argument about higher risks involved with real estate lending is also consistent with the empirical findings of Blasko and Sinkey (2006), who document that banks holding a vast proportion of their total assets in real estate loans are riskier and have higher probabilities of insolvency. Given the vast relative size of real estate loans in banks loan portfolios and the higher risk associated with these loans, it is of interest to examine whether the strength of bank corporate governance affects the level of real estate lending and the amount of real estate loan losses around the financial crisis.

3 The interactions between financial institutions and real estate markets have received considerable attention in the literature over the past few decades. Tripp and Smith (1993), Ambrose, Benjamin and Chinloy (2003), and Igan and Pinheiro (2010) discuss the relationship between real estate lenders, interest rates, and the availability of real estate loans. Allen, Madura and Wiant (1995), He, Myer and Webb (1997), and Elyasian Mansur and Wetmore (2010) examine the effects of real estate market conditions on bank stocks, and find that bank stock prices are very sensitive to changes in real estate prices. Finally, Davis and Zhu (2009) document a positive relationship between real estate price developments and bank profitability. There is considerable empirical evidence suggesting that firms with strong corporate governance practices are associated with better financial performance and higher firm valuation (see e.g., Gompers, Ishii and Metrick, 2003; Brown and Caylor, 2006; Bhagat and Bolton, 2008; Bebchuk, Cohen and Ferrell, 2009; Ammann, Oesch and Schmid, 2011). Furthermore, Friday and Sirmans (1998), Feng, Ghosh and Sirmans (2005), Bauer, Eichhotlz and Kok (2010), and Kohl and Schaefers (2011) examine the role of corporate governance in the real estate industry, and document that good governance may enhance profitability and valuation of real estate companies. 3 The effects of corporate governance on bank performance have been examined e.g. in Mishra and Nielsen (2000), Macey and O Hara (2003), Sierra, Talmor and Wallace (2006), Caprio, Laeven and Levine (2007), de Andres and Vallelado (2008), Cornett, McNutt and Tehranian (2009), Laeven and Levine (2009), and Webb Cooper (2009). These studies demonstrate that effective corporate governance mechanisms are positively associated with bank performance and stock market valuation. Overall, the prior literature on bank corporate governance indicates that the same corporate governance attributes that affect non-financial firms are also relevant in bank governance.

4 The relationship between corporate governance practices and risk taking of financial firms has also been examined in several studies. Akhigbe and Martin (2008) document that short and long term risk measures of financial firms vary inversely with the strength of governance characteristics. Laeven and Levine (2009) examine the underlying reasons of risk taking by banks, and find that the same regulation seems to have different effects on risk taking depending on the bank s corporate governance structure. Their findings further indicate that banks with large owners are riskier than banks with a more dispersed ownership structure. Pathan (2009) finds that strong bank boards (i.e., small and less restrictive boards) are positively related to bank risk taking. However, he also documents a negative relationship between independent directors and bank risk measures, suggesting that directors may view their role as balancing between the interests of shareholders and the other relevant bank stakeholders, including depositors and regulators. Finally, Fortin, Goldberg and Roth (2010) examine bank risk taking at the onset of the financial crisis, and document that banks with stronger corporate governance mechanisms were associated with higher risk taking before the outbreak of the market turmoil. Our analysis is most closely related to the recent studies by Aeb Sabato and Schmid (2012), Beltratti and Stulz (2012), Erkens, Hung and Matos (2012), Fahlenbrach and Stulz (2011), and Peni and Vahamaa (2012). Beltratti and Stulz (2012) examine stock returns of large banks in 31 countries in the midst of the financial crisis. Their results indicate that the worstperforming banks were larger and had lower amounts of deposits and equity capital. Regarding corporate governance attributes, Beltratti and Stulz (2012) find that banks with strong, shareholder-friendly boards performed worse during the crisis. Erkens et al. (2012) focus on the effects of corporate governance on bank stock returns in 30 countries from January 2007 to September 2008, and document that banks with more independent boards and larger institutional

5 ownership had lower stock returns amidst the market turmoil. Moreover, their results indicate that bank risk taking at the onset of the crisis was positively related to the level of institutional ownership. Fahlenbrach and Stulz (2011) investigate the influence of managerial ownership and compensation on the stock returns and profitability of U.S. banks during the crisis. They find that the worst-performing banks had larger CEO ownership, while the CEO s option compensation and cash bonuses were unrelated to bank performance during the crisis. Aebi et al. (2012) examine whether corporate governance mechanisms related to risk management are associated with bank performance during the financial crisis. Their findings indicate that the presence of a chief risk officer that reports to the board had positive effects on bank performance during the crisis, while traditional attributes of strong corporate governance had no effect or even a negative effect on stock returns and profitability. Finally, Peni and Vahamaa (2012) document that banks with strong corporate governance practices had lower stock market valuations amidst the crisis and were associated with higher risk-taking at the onset of the crisis. Overall, the prior literature indicates that banks with strong corporate governance attributes did not perform any better or performed even worse during the financial crisis. In this paper, we aim to extend the above literature by examining the effects of corporate governance on real estate lending performance of large U.S. bank holding companies. The empirical findings reported in this paper demonstrate that the relationship between bank corporate governance and real estate lending performance during the financial crisis is somewhat ambiguous. Our analysis is based on the largest publicly traded U.S. bank holding companies with some form of real estate lending during 2006 2009, and we apply the Corporate Governance Quotient of Institutional Shareholder Services (ISS) to measure the strength of governance mechanisms within these banks. Consistent with the prior literature on the effects of

6 corporate governance on bank performance during crisis (Aebi et al., 2012; Beltratti and Stulz, 2012; Erkens et al., 2012; Fahlenbrach and Stulz, 2011; Peni and Vahamaa, 2012), we find that banks with stronger corporate governance structures did not perform any better or performed even slightly worse than banks with weaker governance structures in terms of return on assets. Furthermore, we document that the relative amount of real estate lending is lower in larger banks that have stronger corporate governance mechanisms, and that banks with weaker governance decreased their higher exposure to real estate loans after the meltdown of the real estate market. Our findings on the effects of corporate governance on real estate lending performance are mixed and depend on the definition of the crisis period. The results demonstrate the strength of corporate governance is generally negatively associated with the amount of real estate loan losses during 2006 2008. Nevertheless, when real estate loan losses were at their highest level in 2009, banks with better corporate governance structures experienced significantly larger loan losses on different types of real estate loans. This finding suggests that effective governance practices did not improve banks underwriting and loan pricing decisions and constrain excessive risk taking in real estate lending. Overall, the finding that the strength of corporate governance mechanisms is positively associated with the amount of real estate loan losses in 2009 is broadly consistent with the results of Aebi et al. (2012), Beltratti and Stulz (2012), Erkens et al. (2012), Fahlenbrach and Stulz (2011), and Peni and Vahamaa (2012), and thereby provides further evidence to suggest that strong corporate governance practices did not necessarily moderate the adverse effects of the financial crisis on bank performance. The rest of the paper is structured in the following manner. Section 2 describes our data set on the largest U.S. bank holding companies and presents the methodology used in the

7 analysis. Section 3 reports our empirical findings on the association between corporate governance and real estate lending performance. Finally, Section 4 summarizes the results and provides conclusions. 2. Empirical setup 2.1. Data We use data on publicly traded bank holding companies that are included in the S&P 1500 index and had some form of real estate lending during 2006 2009. We exclude the bank holding companies with total assets below $3 billion, and banks that have less than 10 % of their total assets invested in real estate loans. 4 After further excluding the banks with insufficient financial and/or corporate governance information, we obtain a sample comprising of 49 individual bank holding companies and 191 firm-year observations for the fiscal years 2006 2009. The average sum of total assets of the banks included in our sample is about $6.8 trillion, and thereby the sample covers a substantial proportion of the total U.S. banking assets despite the small number of individual firms. All the financial statement data for the banks are obtained from the Federal Reserve Board s Bank Holding Company Performance Reports (BHCPR) that are publicly available on a quarterly basis from the National Information Center. 5 Prior literature has proposed several alternative measures of corporate governance. In this paper, we apply the industry-specific Corporate Governance Quotient (CGQ) index issued by Institutional Shareholder Services (ISS) to measure the strength of governance mechanism in bank holding companies. 6 These data are obtained from the RiskMetrics Group. 7 The CGQ index has been recently used e.g. in Chhaochharia and Laeven (2009), Ertugrul and Hegde

8 (2009), and Bauer et al. (2010). The ISS Corporate Governance Quotient is based on 67 different firm-specific factors, which present both the internal and external governance of the firm. The different corporate governance sectors included in the CGQ index are audit, board of directors, charter/bylaws, director education, executive and director compensation, ownership, progressive practices, and state of incorporation. The governance data underlying the CGQ index are collected from public filings, company websites, and surveys conducted by the ISS. The CGQ index values are based on the ranked strength of corporate governance mechanisms of each bank relative to other publicly traded banks, and are expressed in percentile basis. 8 Thus, the CGQ values may range from 0 to 100, with higher values of the index corresponding to stronger corporate governance so that the bank with the best corporate governance practices is assigned a CGQ value of 100. Because the prior literature suggests that the implications of governance practices are often seen with a lag, we use one-year lagged values of CGQ in our empirical analysis. Essentially, the question posed in this paper is whether better corporate governance improves bank performance during the financial crisis by constraining the risk taking associated with real estate lending. We measure bank profitability by the return on assets (ROA), calculated as the bank s net income to its total assets. We analyze the effects of corporate governance and three different real estate lending measures on ROA. Furthermore, we analyze the influence of corporate governance on those three measures of real estate lending: (i) real estate loans to total assets (LOANTA), (ii) real estate loan losses to total assets (LOSSTA), and (iii) real estate loan losses to real estate loans (LOSSLN). In addition, we also examine the effects of governance on the lending measures for three different types real estate loans, i.e., residential loans, commercial loans, and construction and land development loans.

9 In order to analyze the effects of corporate governance on bank performance and real estate lending, we need to control for certain factors. Usually, analysts try to control for at least two major factors when comparing banks. Size is typically considered as the most important basis of comparison because different sized financial institutions may have very different characteristics, business strategies, and product compositions. Moreover, bank size may also surrogate for numerous omitted variables. Following the prior literature (see e.g., Mishra and Nielsen, 2000; Sierra et al., 2006; Caprio et al., 2007; Cornett et al., 2009; Laeven and Levine, 2009; Webb Cooper, 2009), we measure size (SIZE) by the logarithm of the bank s total assets. Leverage or different capital ratio measures are typically the second basis of comparison. Previous studies have shown that leverage is important in explaining bank performance. Leverage is also a useful indicator of risk and a proxy for the financial health of the bank. We measure leverage (LEV) as the bank s tangible common equity to tangible total assets. 9 Although leverage is commonly measured as debt to assets or debt to equity for non-financial firms, the most common measures of leverage used by the banking industry and bank regulators are various measures of equity to total assets or risk-weighted assets. Therefore, higher LEV in our analysis actually corresponds to lower levels of debt. Other factors that analysts often control for are bank age and location. Age may be important for younger banks because it usually takes several years for a bank to become profitable. Given that our sample comprises of large established bank holding companies, age should not be a significant factor in our analysis. Location may also be important because a bank s performance is usually a reflection of the economy where it operates. Most of the banks in our sample have a national or large regional location.

10 2.2. Descriptive statistics Panel A of Table 1 reports the descriptive statistics for the different variables used in the analysis. As can be noted from the table, the bank holding companies included in our sample have, on average, strong corporate governance mechanisms relative to other publicly traded banks. The mean GOV value is 79.6, with index values ranging from a minimum of 13.3 to a maximum of 100. Therefore, the average bank in our sample is in the upper quartile of banks with respect to the strength of corporate governance. Bank profitability, as measured by return on assets (ROA), varies between 6.8 % and 2.7 % during the sample period. The mean real estate loans to total assets (LOANTA) is approximately 45 % and varies from a low of 14.4 % to a high of 71.9 %; therefore, our sample represents a wide range of lending activity. The mean and median of real estate loan losses to real estate loans (LOSSLN) during 2006 2009 are 0.98 % and 0.29 %, respectively, with a minimum of 0.02 % (or a recovery of previous charge-offs larger than the current charge-offs) and a maximum of 8.02 %. 10 The sample represents a wide range of loan losses relative to the size of the loan portfolio. Real estate loans losses to total assets (LOSSTA) has a mean of 0.46 % and a wide range from 0.01 % to 5.76 %. Panel A also shows that the largest U.S. bank holding companies are highly leveraged, with a mean tangible equity to tangible assets ratio of 5.8 %. 11 Finally, the logarithm of total assets in thousands (SIZE) varies from 15.01 ($3.31 billion) to 21.62 ($2.44 trillion), with a mean of 17.10 ($141.43 billion). (insert Table 1 about here)

11 Panel B of Table 1 provides the descriptive statistics for two subsamples sorted by the strength of corporate governance, GOV. The banks with Corporate Governance Quotient values above sample medians in each year comprise the stronger governance subsample, while the weaker governance subsample consists of the banks with below median CGQ scores. Several interesting features can be noted from these statistics. First, Panel B indicates that the banks with stronger governance have slightly higher return on assets (0.45 % vs. 0.35 %), thereby suggesting that strong governance may improve bank profitability. The difference in ROA between the two subsamples, however, is statistically insignificant. Furthermore, Panel B shows that the banks with stronger governance practices have a lower amount of real estate loans to total assets (40.41 % vs. 50.13 %). A t-test indicates that the difference in LOANTA is statistically highly significant. There is virtually no difference in the amount of real estate loan losses to real estate loans between the two subsamples, while the real estate loan losses to total assets for the stronger governance banks appears slightly higher than for weaker governance banks (1.09 % vs. 0.87 %). The differences in the both measures of real estate loan losses are statistically insignificant. As can be further noted from Panel B, there is no significant difference in the leverage figures between the two subsamples, while the banks with stronger governance mechanisms are, on average, substantially larger than the banks with weaker corporate governance. (insert Table 2 about here) Table 2 reports the sample means by year for the different variables used in the analysis. The table shows that the mean Corporate Governance Quotient (GOV) slightly increases during

12 the sample period. Bank profitability, as measured by ROA, decreases each year over our sample period, and is negative during the crisis years 2008 and 2009. Interestingly, return on assets is higher for the banks with stronger corporate from 2006 to 2008, but much lower in 2009 when the profitability of U.S. banks was at its lowest level. However, the difference in ROA between the two subsamples is statistically significant only in year 2006 (1.50 % vs. 1.14 %). The amount of real estate loans to total assets stays relatively constant from 2006 to 2008, and decreases in 2009. This decrease may reflect writedowns in the value of real estate loans and may also indicate that banks were trying to reduce their exposure to real estate lending after the meltdown of the real estate market. The proportion of real estate loans to total assets is about 10 percentage points lower for the banks with stronger corporate governance practices, and the differences in LOANTA are statistically significant throughout the sample period. The two real estate loan loss measures demonstrate the adverse effect of the financial crisis on real estate lending performance, as both loan loss measures have increased substantially in 2008 and 2009. Although the differences in real estate loan losses between the stronger and weaker governance subsamples are statistically insignificant, it should be noted that both LOSSTA and LOSSLN appear lower for the banks with stronger governance in 2008, and higher in 2009. Finally, Table 2 shows that the banks with stronger corporate governance structures are significantly larger, and that there is no difference in the leverage ratios between the stronger and weaker governance subsamples. (insert Table 3 about here)

13 Table 3 presents the correlation matrix of the variables used in the analysis. Corporate Governance Quotient appears strongly positively correlated with size (0.42) and negatively correlated with LOANTA ( 0.37). Thus, consistent with Tables 1 and 2, the correlations imply that banks with stronger governance structures are larger and have a lower amount of real estate loans to total assets. As expected, the return on assets exhibits a strong negative correlation with the two measures of real estate loan losses. ROA is also significantly negatively correlated with LOANTA. Moreover, Table 3 shows that the different real estate loan variables are positively correlated with each other, and the both loan loss variables are negatively correlated with LEV. Finally, it should be noted that bank size and leverage are negatively correlated, indicating that larger banks have less equity capital. 2.3. The empirical models We begin by examining the association between bank profitability, corporate governance, and real estate lending measures around the financial crisis with the following fixed-effects panel regression model: ROA t GOV 0 t 1 LOANTA 6 1 t GOV 2 t 1 LOSSLN 8 CRISIS SIZE t n 1 k 1 BANK k 4 k i t LEV 2009 5 y 2007 y t YEAR y i t (1) where ROA t denotes the return on assets for bank i at time t, GOV t is the Corporate Governance Quotient, CRISIS is a dummy variable which equals one in 2008 and 2009, SIZE t is the logarithm of total assets, and LEV j,t is the ratio of tangible common equity to tangible total assets, LOANTA is real estate loans to total assets, LOSSTA is real estate loan losses to total assets, LOSSLN is real estate loan losses to real estate loans, BANK denotes a dummy variable k i

14 for bank and y YEAR i is a dummy variable that indicates fiscal years. We also estimate an alternative version of Equation (1) in which we replace the GOV t-1 CRISIS variable with GOV t-1 YEAR 2008 and GOV t-1 YEAR 2009 interaction variables. After establishing the relation between profitability, governance, and real estate lending, we examine the influence of corporate governance on real estate lending with the following panel regression specification: REL t GOV 0 n 1 k 1 1 k t 1 BANK k i GOV 2 2009 y 2007 y t 1 YEAR CRISIS y i t SIZE 4 t LEV 5 t (2) where REL t denotes one of the alternative real estate lending measures (LOANTA, LOSSTA, or LOSSLN) for bank i at time t, and all the other variables are as defined above in Equation (1). Again, we also estimate an alternative version in which GOV t-1 CRISIS is replaced with GOV t-1 YEAR 2008 and GOV t-1 YEAR 2009 interaction variables. Given that bank profitability and real estate lending measures are likely to differ over time and across banks, we control for the potential time effects and unobserved heterogeneity across banks with y YEAR i and k BANKi dummy variables in Equations (1) and (2). Hence, throughout our panel regressions, we use a two-way fixed-effects specification, which allows for a different intercept for each bank in the sample, and also controls for the possible change in bank performance over time. Moreover, in order to avoid problems with residual correlations, we use standard errors corrected for clustering at the bank level in the regressions.

15 3. Results 3.1. Bank profitability, corporate governance, and real estate lending Table 4 reports the estimation results of alternative versions of Equation (1). The first regression specification (Model 1) is a parsimonious model that includes SIZE and LEV as the control variables and bank and year fixed-effects as additional controls. As can be seen from the table, this model has a good explanatory power for bank profitability, as the adjusted R 2 of the regression is 57.3 % and the F-statistic is significant at the 1 % level. Four of the control variables are statistically significant at the 1 % level with LEV having a positive coefficient, and the fiscal year dummy variables YEAR 2007, YEAR 2008, and YEAR 2009 having negative coefficients. In Model (2), we use GOV YEAR 2008 and GOV YEAR 2009 interaction variables instead of GOV CRISIS. The adjusted R 2 of this regression specification is 57.9 % and the coefficient estimates for the control variables remain almost unchanged. In other words, the estimates of Models (1) and (2) indicate that bank profitability was positively related to the amount of equity capital during 2006 2009, and as expected, our estimates indicate that the return on assets on large U.S. bank holding companies was substantially lower in 2008 and 2009. The test variables of interest in our regression specifications are GOV and the interaction variables GOV CRISIS, GOV YEAR 2008, and GOV YEAR 2009. The estimated coefficient for GOV is positive and statistically significant in both models, indicating that strong corporate governance generally improves bank profitability during the period 2006-2009. This finding is consistent with the prior literature on the positive effects of effective corporate governance practices in the banking industry (see e.g., Sierra et al., 2006; Caprio et al., 2007; de Andres and Vallelado, 2008). The magnitude of the coefficients suggests that a ten point increase in the

16 Corporate Governance Quotient increases ROA by approximately nine basis points. The coefficient estimate for GOV CRISIS appears statistically insignificant, and thereby suggests that the overall positive association between the strength of governance and bank profitability was not affected by the market turmoil. Interestingly, the estimates of Model (2) provide mixed evidence about the effects of corporate governance on bank performance. The positive and significant coefficient for GOV YEAR 2008 indicates that strong governance moderated the adverse effects of the financial crisis on bank performance during the crisis year of 2008. However, the estimated coefficient for GOV YEAR 2009 is negative and statistically significant, and also marginally larger in absolute terms than the coefficient for GOV, thereby demonstrating that banks with better corporate governance structures had significantly lower profitability in 2009. The coefficient of 0.012 in 2008 is slightly larger than the coefficient of 0.010 in 2009 which suggests that the positive effect of governance in 2008 was offset by the negative effect in 2009. Consistent with the empirical findings of Fortin et al. (2010), Aebi et al. (2012), Erkens et al. (2012), and Peni and Vahamaa (2012), the reversal in financial performance in 2009, when the profitability of U.S. banks was at its lowest level, may indicate that banks with better governance practices were taking more risk before the crisis. (insert Table 4 about here) The real estate lending measures are included as additional control variables in Models (3) and (4). In particular, we include LOANTA and LOSSLN to control for the level of real estate lending and the losses in the real estate loan portfolio. 12 Table 4 shows that the inclusion of

17 these two lending variables as additional controls increases the explanatory power of the bank performance regressions to about 79 %, and the estimates demonstrate the substantial impact of real estate loan losses on ROA. Given the importance of real estate loan losses, the results based on Models (1) and (2) should be interpreted with some caution due to the lack of an important explanatory variable. The statistically significant coefficients for the control variables indicate that smaller banks and banks with less leverage (or more equity capital) had higher profitability during 2006 2009. Regarding the estimated coefficients for the real estate lending measures, it can be noted from Table 4 that real estate loan losses (LOSSLN) are, not surprisingly, strongly negatively associated with return on assets, while the negative coefficients for LOANTA appear statistically insignificant at the conventional significance levels. Hence, our estimates indicate that the level of real estate lending (LOANTA) does not have any systematic effect on bank profitability during our sample period. 13 In Model (3), the coefficient estimate for GOV CRISIS is positive and highly significant, while the coefficient for GOV appears statistically insignificant. Similarly, in Model (4), the estimated coefficients for GOV YEAR 2008 and GOV YEAR 2009 are positive and statistically significant at the 1 % level and the coefficient for GOV is insignificant. Consistent with the extensive literature on the positive effects of corporate governance on firm performance, our estimates suggest that strong corporate governance mechanisms may have improved bank performance amidst the financial crisis after controlling for the negative effect of loan losses on profitability. It is important to recognize that alternative model specifications may lead to somewhat different results. Given that the coefficient for GOV YEAR 2009 is negative in Model (2) and turns positive after controlling for the amount of real estate loan losses in Model (4), our estimates indirectly suggest that loan losses are higher in 2009 for banks with higher GOV. The

18 positive coefficient in Model (4) may also indicate that good corporate governance has a positive effect on the remaining contributors to bank profitability after controlling for the negative effect of loan losses. Nevertheless, we acknowledge that it is very difficult to conceive that banks with good corporate governance practices would have been able to manage their other operations well and were simply worse in monitoring their real estate loan portfolios. 14 3.2. The effect of corporate governance on real estate lending performance As the next step of the analysis, we focus on the effects of corporate governance on real estate lending. There are two major real estate lending decisions for a bank. First, what percentage of the bank s assets or loans should be invested in real estate loans (or real estate loan subsets)? The descriptive statistics in Table 1 show a broad range of decisions in this area. The maximum value of 71.85 % for LOANTA shows that at least one bank made a very conscious decision to bet the bank on real estate loans. Further, once the level of real estate loans (or real estate loan subsets) has been set, which underwriting or credit standards should be used to make the loans? These standards will affect the level of losses for the real estate loans. The descriptive statistics in Table 1 show a broad range of experiences for LOSSLN. The interaction or the multiplication of the two ratios, LOSSLN and LOANTA, provides the overall effect of real estate lending on bank profitability, LOSSTA. (insert Table 5 about here)

19 The estimates of the alternative versions of Equation (2) are reported in Table 5. First, we regress the ratio of real estate loans to total assets (LOANTA) on the governance variables and bank-specific controls (Models 1 and 2). The adjusted R 2 s of these regressions are about 92 % and the F-statistics are significant at the 1 % level, thereby indicating a good fit of the models. Regarding the control variables, it can be noted form Table 5 that the negative coefficient for SIZE and the positive coefficient for YEAR 2007 are statistically significant in both models. In addition, the coefficient estimate for YEAR 2009 appears negative and statistically significant in Model (2). These estimates indicate that the relative amount of real estate loans is smaller in larger banks, and is decreasing substantially in year 2009. The latter finding may reflect writedowns in the value of real estate loans and may also indicate that banks were trying to reduce their exposure to real estate lending after the meltdown of the real estate market. Although the univariate tests reported in Table 1 show that stronger corporate governance practices may constrain the level of real estate lending, the insignificant coefficients for GOV in Models (1) and (2) demonstrate that there is no difference in the level of lending after controlling for the effects of other bank-specific factors. Moreover, the coefficient estimates for GOV CRISIS and GOV YEAR 2008 are also statistically insignificant, indicating that the strength of governance mechanisms is unrelated to the amount of real estate lending. However, the estimated coefficient for GOV YEAR 2009 is positive and significant at the 1 % level, and thereby suggests that banks with stronger corporate governance had a higher level of real estate lending to total assets in 2009 after controlling for the large decrease in LOANTA in 2009 for all banks (YEAR 2009 ), and for the differences in bank size and leverage. 15 Overall, our regression results indicate that effective governance practices did not constrain the level of real estate lending around the crisis, suggesting that the statistically significant differences in Tables 1 and 2 are

20 mostly caused by size-effects with larger banks generally having better governance mechanisms and lower amount of real estate lending. Our next question is to assess whether the banks with more effective corporate governance are associated with lower real estate loan losses amidst the financial turmoil. Thus, we use real estate loan losses to total assets (LOSSTA) as the dependent variable in Models (3) and (4). The ratio of real estate loan losses (or net charge-offs) to total assets should reflect the impact of real estate losses on ROA. We calculate LOSSTA by multiplying the ratio of real estate loans to total assets by real estate loan losses to real estate loans. As mentioned above, LOSSTA can be broken down into two variables or management decisions: (i) LOANTA is a management decision on how much of a product, i.e., real estate loans, to produce, while (ii) LOSSLN is a reflection of management decisions with respect to underwriting and pricing the loans, ultimately representing the quality of the real estate loans made by the bank. In Models (3) and (4), the coefficient estimates for SIZE and LEV are negative and statistically highly significant, while the coefficients the fiscal year dummy variables appear positive. 16 Thus, our estimates indicate that larger banks with more equity capital experienced lower real estate loan losses, and further confirm that the amount of real estate losses was significantly higher in the midst of the crisis. The coefficients for LEV indicate that a one percentage point increase in tangible equity to tangible assets is associated with a decrease in real estate loan losses of approximately 20 basis points. In other words, banks that had higher debt to assets ratios were associated with relatively higher real estate loan losses, consistent with the view that the excessive use of leverage by large bank holding companies was one of the factors that contributed to the development of the crisis.

21 As can be noted from Table 5, the estimated coefficients for GOV have the expected negative signs and are statistically significant at the 5 % level in both models, and also the coefficient for GOV YEAR 2008 is negative and highly significant in Model (4). Hence, our estimates suggest that banks with stronger corporate governance are generally associated with lower real estate loan losses and had even lower loan losses in 2008, thereby implying that effective governance practices may have improved lending performance during 2006 2008. Interestingly, however, the coefficient for GOV YEAR 2009 is positive and statistically significant in Model (4), indicating that banks with stronger governance mechanisms had significantly higher real estate loan losses in 2009. The combined estimated coefficients for GOV and GOV YEAR 2009 suggest that a ten point increase in the Corporate Governance Quotient is associated with approximately five basis points increase in LOSSTA in 2009. The coefficient for GOV YEAR 2009 of 0.009 slightly offsets the coefficient for GOV YEAR 2008 of 0.007, which explains the insignificant coefficient for GOV CRISIS in Model (3). Taken together with the profitability results reported in Table 4, the estimates with LOSSTA as the dependent variable indicate that better governance, as reflected by a higher GOV, had an insignificant impact on loan losses for the two-year crisis period 2008-2009, but had a negative impact on ROA in year 2009 through a higher amount of real estate loan losses. The last two regression specifications reported in Table 5 employ the ratio of real estate loan losses to real estate loans (LOSSLN) as the dependent variable (Models 5 and 6). While LOSSTA includes the effect of both LOSSLN and LOANTA, LOSSLN variable evaluates losses only relative to the size of the real estate loan portfolio. This measure is a more direct analysis of the bank s underwriting and pricing decisions, given a certain level of real estate lending. As mentioned earlier, location may also play a role in the level of LOSSLN; however, since most

22 banks in the sample have a national or large regional presence, location should not be an important factor in our analysis. The F-statistics of the LOSSLN models are significant at the 1 % level and the adjusted R 2 s are about 65 %. Similar to Models (3) and (4), the coefficients for SIZE and LEV are negative and statistically highly significant and the fiscal year dummy variables have positive coefficients. Hence, the estimates indicate that the proportion of real estate loan losses to real estate loans was substantially higher during the financial crisis, and moreover, that larger banks and banks with more equity capital are associated with lower real estate loan losses. Once SIZE is controlled for, LEV may tell us something about how conservative the management of the bank is. A higher amount of equity capital would suggest a more conservative management, which may have taken less risk. The negative and significant coefficients for LEV in the four models explaining LOSSTA and LOSSLN are consistent with this hypothesized effect. Regarding our test variables, it can be noted from Table 5 that the coefficients for GOV are negative and significant in Models (5) and (6). In Model (5), the coefficient for GOV CRISIS is statistically insignificant. In Model (6), GOV YEAR 2008 is negative and significant, while the coefficient for GOV YEAR 2009 is positive and significant. Therefore, consistent with the LOSSTA regressions, our estimates indicate that banks with stronger corporate governance structures generally have a lower amount of real estate loan losses per real estate loans and had even lower loan losses in 2008. When the financial crisis is defined as a two-year period as in Model (5), the effect of corporate governance on loan losses during the crisis remains the same as during the whole four-year sample period. Nevertheless, the estimates of Model (6) demonstrate that when the real estate loan losses were at the highest level in 2009, loan losses were significantly higher for banks with stronger governance practices. The

23 magnitudes of the combined coefficients for GOV and GOV YEAR 2009 indicate that a ten point increase in the Corporate Governance Quotient was associated with an approximately eight basis points increase in LOSSLN in year 2009. This finding suggests that effective corporate governance practices did not improve loan quality and constrain excessive risk taking in real estate lending with respect to a bank s underwriting and pricing decisions in the midst of the financial crisis. 3.3. Corporate governance and losses on different types of real estate loans To further analyze the effects of corporate governance on real estate lending, we next estimate the panel regression model given by Equation (2) with LOSSLN as the dependent variable separately for three different types of real estate loans: (i) residential loans, (ii) commercial loans, i.e., nonfarm nonresidential real estate loans, and (iii) construction and land development loans. The largest category of loans made by the large bank holding companies is real estate loans secured by 1-4 family homes, which includes the typical residential mortgage and home equity loans. 17 It should be noted that the different types of real estate loans are generally associated with different risk characteristics, with the residential loans typically considered as the least risky and the construction and land development loans as the most risky form of real estate lending. The risk characteristics of different real estate loan subsets are also distinctively reflected in the realized loan losses during the crisis; the average loss rates on residential loans were 1.41 % in 2009, whereas the corresponding loss rates on commercial loans and construction and land development loans were 2.81 % and 6.65 %, respectively. Moreover, while residential loan losses increased by 683 % from 2007 to 2009, losses on

24 commercial and construction and land development loans surged by 1,907 % and 2,560 %, respectively. (insert Table 6 about here) Table 6 reports the estimation results of the alternative LOSSLN regressions for the different types of real estate loans. In general, the estimates of these regressions are very consistent with those reported in Table 5, and thereby further indicate that stronger corporate governance is negatively associated with the amount of real estate loan losses until 2009, and strongly positively related to loan losses in 2009. As can be noted from the table, the adjusted R 2 s of the regressions vary between 51.4 and 62.2 %, and the F-statistics are statistically highly significant in each model. Similar to the loan loss regressions in Table 5, the coefficients for SIZE and LEV are negative and statistically significant in most specifications, while the fiscal year dummy variables have positive coefficients. Regarding the effects of corporate governance on real estate loan losses, it can be noted from Table 6 that the estimated coefficients for GOV are negative and statistically significant in each of the six alternative models. Thus, our estimates suggest that effective corporate governance practices generally reduced real estate loan losses for all three different types of real estate loans during 2006 2009. The coefficient estimate for GOV CRISIS is statistically insignificant in all three models, and thereby indicates that the negative association between corporate governance and real estate loan losses was not significantly affected by the financial crisis. However, in Models (2), (4), and (6), coefficients for GOV YEAR 2009 interaction variable are positive and statistically significant at the 1 % level. Therefore, consistent with Table 5, the

25 results indicate that losses on residential, commercial, and construction and land development loans were significantly higher in 2009 for banks with stronger corporate governance practices. In sum, the additional regressions reported in Table 6 suggest that effective corporate governance practices reduced real estate loan losses for all three different types of real estate loans until 2009, and in 2009, banks with stronger corporate governance had significantly higher real estate loan losses for all types of real estate loans. These findings provide further evidence to suggest that good corporate governance mechanisms did not necessarily improve bank s underwriting and loan pricing decisions and constrain excessive risk taking in real estate lending. 3.4. Discussion and further tests Overall, the results reported in Tables 4, 5 and 6 provide somewhat mixed evidence on the effects of corporate governance on bank performance and real estate loan losses during the financial crisis. Consistent with the extensive literature on the positive effects of strong governance on bank performance (see e.g., Sierra et al., 2006; Caprio et al., 2007; de Andres and Vallelado, 2008), our results indicate that banks with better corporate governance had higher profitability and better lending performance ratios during 2006 2008. Nevertheless, our empirical findings depend on the definition of the crisis period and we also document that banks with strong governance mechanisms experienced significantly larger real estate loan losses in 2009 when the loan losses of U.S. banks were at their highest level. Thus, our results indicate that effective corporate governance practices did not necessarily mitigate the adverse effects of the financial crisis on real estate loan losses and, ultimately, on bank performance.

26 Interestingly, our empirical findings demonstrate that the negative effect of strong corporate governance on real estate loan losses was reversed in 2009. We interpret this finding as evidence that banks with better corporate governance mechanisms took more risk in real estate lending before the crisis in order to maximize shareholder wealth, and that these risks were then materialized and exposed during the crisis. This interpretation is broadly consistent with the recent literature that suggests that banks with better governance practices were taking more risk before the crisis (Fortin et al., 2010; Erkens et al., 2012; Peni and Vahamaa, 2012), and had lower stock returns in the midst of the turmoil (Aebi et al., 2012; Beltratti and Stulz, 2012; Erkens et al., 2012). Although weaknesses in bank corporate governance have been cited among the main factors contributing to the development of the financial crisis, the empirical evidence generally demonstrates that banks with strong governance practices did not perform any better or may have performed even worse during the crisis. This indicates that good corporate governance is not a substitute for good managerial decisions. To further investigate why the banks with stronger corporate governance performed better in terms of real estate lending in 2008 and worse than the banks with weaker governance in 2009, we analyze the composition of the real estate loan portfolios and changes in equity capital around the crisis. First, we regress the components of the real estate loan portfolios on GOV and our control variables. These regressions (not tabulated) indicate that banks with stronger corporate governance mechanisms had larger amounts of riskier commercial real estate loans (i.e., commercial loans and construction and land development loans) at the onset of the crisis, and that the relative amount of commercial real estate lending in these banks was reduced after the financial turmoil began. Hence, the regression estimates provide further support for the