Prevention is better than cure – Corporate governance relations within banks

(1)

Prevention is better than cure – Corporate governance relations

within banks

Abstract

This study examines the relations between board independence, the strength of the risk

management function and bank risk-taking. A sample of 85 US bank holding companies in the period 2008-2016 is examined. Using the Generalized Method of Moments, this study finds evidence for a negative association between bank risk taking and strength of the risk

management function. In addition, the results suggest that the level of bank risk-taking is positively associated with board independence levels. Therefore, the results suggest that former bank risk-taking levels influence the decisions made regarding corporate governance.

Key Words: Banking, Risk Management, Board Independence, Risk-Taking, GMM Regression

JEL classifications: G21, G32, C22, C38

University: University of Groningen Faculty: Economics and Business Date: January 2019

Name: Fenneke B. Korteweg Student number: S2569590

Supervisor: Prof. Dr. K.F. Roszbach

(2)

Table of Contents

1. Introduction 3

2. Literature Review and Hypothesis Development 4

3. Variables and Data 9

4. Methodology 14

5. Results 16

6. Discussion and Conclusion 23

7. References 25

(3)

1. Introduction

Klaas Knot, the Dutch Central Bank’s president, believes that nowadays banks still strive for excessive returns. Banks declare that they have no other option; their shareholders demand it.

1 Bank risk-taking has been a hot topic since the financial crisis of 2008. During the crisis it

became apparent that banks had engaged in high risk-taking behaviour to achieve high returns. When it became evident that these high returns, due to the high risk-taking, were no longer sustainable, many banks had to be saved from bankruptcy. Whether banks have learned from the financial crisis remains debatable.

Since the financial crisis of 2008, regulators have tried to prevent excessive and unsustainable risk-taking by banks. In many countries the regulator has increased its supervision and extra rules have been introduced with regard to the governance of banks (Baker and Wurgler, 2015). In addition to the attention of the regulators, academic literature has partly shifted their focus from returns towards bank risk-taking as well (De Haan and Vlahu, 2015).

Several studies have investigated corporate governance and its effect on bank risk-taking. Prior literature has focused on a board attribute or the risk management and its effect on bank risk-taking. However, according to Srivastav and Hagendorff (2016) some indications are present that board independence, risk management and bank risk-taking interact with each other. However, the interaction and the holistic understanding of these concepts is missing. To create a deeper understanding and a more holistic view, this study examines all the relations between board independence, risk management and bank-risk taking.

This study makes a contribution to the existing literature by examining the relations between board independence, the risk management index and bank risk-taking. It is

specifically relevant for banks and their stakeholders since corporate governance mechanisms are important to prevent excessive risk-taking and its related costs. In addition, regulators benefit from research on the effectiveness of corporate governance mechanisms, since this enables them to adjust regulation accordingly. Lastly, this study might be relevant for investors in banks, as research regarding bank’s risk profile and related governance mechanisms enables them to optimize their portfolio.

This study uses two-stage GMM regression to examine the relations between board independence, the strength of the risk management function and bank risk-taking. The results indicate a negative association between board independence and strength of the risk

(4)

management function. In addition, the results suggest that the level of bank risk-taking is positively associated with the level of board independence. Therefore, the most important finding of this study is that former bank-risk taking levels significantly influence corporate governance mechanisms.

The remainder of this study is structured as follows: in section two, the relevant literature is elaborated upon and hypotheses are developed, in section three the sample and methodology that are used are described, in section four the results are explained, and in section five, the conclusions and the limitations of this study are reviewed.

2. Literature Review and Hypothesis Development

The agency theory of Ross (1973) has been one of the most important theories in economic and management literature. The theory states that the interests of the manager (agent) and the owner (principal) are not aligned. Due to information asymmetry between the two parties, the agent is able to fulfil his own desires instead of the principal’s, and agency problems are created. To mitigate the agency problem, the owner could monitor the actions of the manager with several corporate governance mechanisms. Given these different interests, according to Schultz et al. (2010), corporate governance mechanisms are necessary to ensure that

management takes actions that benefit the shareholders.

However, these corporate governance mechanisms do not function in the same way for all companies. The subject of interest of this study is financial bank holding companies, which are fundamentally different from non-financial companies. According to De Haan and Vlahu (2015), bank holding companies are inherently different because they are heavily regulated, have a different capital structure and have a more complex and opaque structure and business model. Because of these differences, corporate governance structures that work for non-financial companies do not have the same effects in the banking environment.

(5)

more shareholder-friendly provided shareholders with a higher return. However, this profitability did not favour all the stakeholders of the bank. Shareholders preferred and

received a higher return due to higher risk-taking of the bank. On the other hand, creditors and the ‘taxpayer' have a limited upside on their claim on the bank, and consequently preferred lower risk-taking. Therefore, the higher profitability levels of banks before the crisis was at the expense of the ‘taxpayers' and creditors (Pathan, 2009). After the crisis, banks have focused on the stakeholder approach instead of the shareholder approach (Bolton et al. 2015). Also, academic literature has shifted towards a focus on risk-taking compared to return.

In the domain of corporate governance and its effects on bank risk-taking, recent research has mainly focused on board attributes, risk management, and executive

compensation. The risk management aspect has received less academic attention since ‘risk management’ is more difficult to quantify and data is harder to obtain (Srivastav and Hagendorff, 2016). However, previous studies have found that stronger risk management (Keys et al., 2009; Ellul and Yeramilli, 2013), risk management committees (IMF; 2014) and less persistent risk-culture (Fahlenbrach et al., 2012) are associated with lower bank risk taking. However, these aspects have only been studied in isolation, their effect on bank risk-taking is not combined with the other corporate governance mechanisms. Also, the

interrelations, between these mechanisms have not been studied. As suggested by Srivastav and Hagendorff (2016), this study combines two aspects of corporate governance and bank risk-taking to examine their interrelations. Therefore, this research makes a contribution to existing literature on corporate governance and bank risk-taking.

Hypothesis Development

The research objective of this study is to examine the relations between board independence, strength and quality of risk management and bank risk-taking. The question is whether a causal relation exists and if so, what the direction of this relationship is.

Risk management function

(6)

might be too difficult or too risky for independent board members to monitor. Banks that have less effective risk management might attract board members who do not attach much value to effective risk management either. Therefore, the quality and strength of the risk management might positively influence the number of independent members of the board. Therefore, the expectation is,

Hypothesis 1a: The strength and quality of risk management have a positive effect on the levels of board independence.

The effect of risk management on bank risk-taking has been studied by Ellul and Yerramilli (2013). They find that higher strength and independence of the risk function is associated with lower tail risk at banks. They argue that the risk function is necessary since both customers and the external market might not be able to contain the manager's desire for risks that enhance short-term performance. The structure of the risk function determines how effectively risk information is shared between the board and the business segments. However, due to the compensation schemes of top management, which encourage risk-taking, risk management needs independence and strength to have significant power to control these incentives. In conclusion, a strong and independent risk management function is expected to prevent managers from taking excessive risk, which decreases bank risk taking.

Other literature finds similar results. Keys et al. (2009) have found that stronger risk management is associated with lower bank risk taking. Furthermore, risk management committees (IMF; 2014) and less persistent risk-culture (Fahlenbrach et al., 2012) are associated with lower bank risk taking.

This leads to the following expectation;

Hypothesis 1b: The strength and quality of risk management has a negative effect on the levels of bank risk-taking.

Board Independence

According to Srivastav and Hagendorff (2016), the board of a bank is responsible for monitoring risks and engaging in risk management. However, previous literature has

established that the effectiveness of the board to maintain adequate risk management differs due to different board attributes. This study focuses on board independence.

(7)

monetary ties at risk when advising or monitoring management. Therefore, independent board members are believed to be more objective compared to inside executive board members.

According to Hagendorff and Vallascas (2011), managers of the bank often receive compensation packages that are dependent on the profits of the bank. If compensation

packages depend on returns only, managers have incentives to increase the return of the bank regardless of the increase in risk levels. As explained, this risk-taking is not in the interest of all the stakeholders of the bank, such as the creditors. Boards with more independent board members are likely less tempted by management to increase risk levels and evaluate these proposals more objectively since these board members do not have monetary or personal ties at risk. Therefore,

Hypothesis 2a: Higher levels of board independence have a negative effect on bank risk-taking levels.

In line with the argument above, this study expects that boards with a relatively high number of independent board members are more likely to curtail management’s desire to increase risk levels. (Erkens et al., 2012; IMF, 2014) Board members are more likely to ensure a strong risk management function, since this is crucial for their possibilities to curtail risk levels. The measure of the strength and quality of the risk management will be extended upon in the methodology section.

Therefore, this study expects;

Hypothesis 2b: Higher levels of board independence have a positive effect on the strength and independence of risk management.

Bank Risk Taking

Many studies that examine the effects of corporate governance mechanisms, do not consider the possibility of endogeneity (Schultz et al, 2010). Endogeneity means that the error term in the model is correlated with the independent variables. Multiple corporate governance studies acknowledge that endogeneity might cause a problem, yet do little to mitigate the problem (i.e. Erkens et al., 2012; V. Aebi et al., 2012; Minton et al. 2014).

(8)

and bank risk-taking might all be affected by firm-specific characteristics, such as the risk culture present in the bank. (Schultz et al, 2010; Haubrich, 1998). Therefore, this study explicitly examines the possibility of endogeneity.

Ellul and Yerramilli (2013) acknowledge in their research that the level of risk-taking could have an effect on risk management. They argue that the risk culture in banks might determine the choice for the strength of the risk management function. Based on that line of argumentation, conservative BHCs chose to take little risks and ensure a stronger risk management function is in place. Where on the other hand, high risk-taking BHCs might be less conservative and also have weaker risk management function. From that perspective, bank risk-taking also determines the importance of risk management.

Hypothesis 3a: Bank risk-taking has a negative effect on the strength and quality of risk management.

Based on the line of argumentation above, the risk culture of the bank could also influence the bank risk-taking and the choice for independent board members. Banks that have a culture of high risk-taking might be less conservative in their choice of supervision and appoint inside, less objective board members. On the other hand, banks with a culture of low-risk taking might be more conservative in their choice of supervision and appoint more independent board members. Therefore,

Hypothesis 3b: Bank risk-taking has a negative effect on the level of board independence. Figure 1 - Conceptual model

(9)

3. Variables and Data

Strength and quality of risk management

Academic literature has not been able to establish a valid and reliable measure for risk management for financial institutions. Some studies have focused on using survey data, where executives expressed their ideas and attitudes towards the risk management function.

However, these surveys rely on self-reported data and may suffer from a social desirability bias. An alternative for survey data is to employ measures that rely on publically observable variables. An advantage of using these measures is that companies are limited in their ability to influence their outcomes and therefore these measures are more objective. On the other hand, measures that rely on publicly available data may be less valid measures of the ‘real importance' the bank has given risk management. Since obtaining surveys is beyond the scope of this research and may create biases, this study uses the Risk Management Index (RMI) created by Ellul and Yerramilli (2013). The RMI uses observable variables such as the presence of a Chief Risk Officer (CRO) at the bank as proxies for the strength and quality of the risk management function within the bank. After the collection of these observable variables this study uses Principal Component Analysis (PCA) to construct the index. The main advantage of this method is that it establishes an index of these variables without

choosing arbitrary weights, The data used to construct the risk management index is collected from BoardEx and ExecuComp. Whenever data is not available, the data is manually

collected from annual proxy statements from the SEC government website.

Information of the CRO of the bank comprises one part of the Risk Management index. The variables included are:

CRO present: A dummy variable that indicates whether a CRO (or an equivalent function)

responsible for enterprise-wide risk management is present within the bank.

CRO executive: a dummy variable that identifies whether the CRO is an executive officer of

the bank.

CRO Top5: a dummy variable that identifies whether the CRO is amongst the five

highest-paid executives at the bank.

CRO centrality: The ratio of the CRO’s total compensation to the CEO’s total compensation,

(10)

According to Ellul and Yerramilli (2013), the other component of the RMI index measures the quality of the risk oversight by the risk committee of the bank. The variables used are:

Risk committee experience: a dummy variable that identifies whether at least one of the

independent directors serving on the board’s committee has banking and finance experience.

Active Risk committee: a dummy variable that identifies whether the board’s risk

committee met more frequently during the year compared to the average board risk committee across all banks.

Board Independence

In accordance with Erkens et al. (2012), this study classifies board members as independent if they are non-executive directors and have no direct personal or monetary ties to management. This leads to,

𝐵𝑜𝑎𝑟𝑑 𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑒 = # 𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑑𝑖𝑟𝑒𝑐𝑡𝑜𝑟𝑠

# 𝑇𝑜𝑡𝑎𝑙 𝑑𝑖𝑟𝑒𝑐𝑡𝑜𝑟𝑠 (1)

The information regarding the board independence is collected from the GMI ratings from the MSCI database.

Bank Risk Taking

The bank risk-taking variable is based on the credit risk taken by the bank. Credit risk is defined as the risk that the creditor of the bank is not able to pay back his loan in full. The fraction of non-performing loans can be used to measure the credit risk. According to Louzis et al. (2011), non-performing loans are loans that have not received a scheduled payment for at least 90 days. Loans that are not paid back fully reduce the earnings of banks. Therefore, low non-performing loan ratios are essential for banks’ survival. Banks that have relatively high non-performing loans ratios are riskier than banks with lower non-performing loan ratios. Therefore, the non-performing loans ratio can be used as a valid measure of bank risk-taking, as the chances that non-performing loans are paid back fully are substantially lower. Following Ellul and Yeramilli (2013), this study measures credit risk using the fraction of non-performing loans. Consequently,

𝐵𝑎𝑛𝑘 𝑅𝑖𝑠𝑘 − 𝑇𝑎𝑘𝑖𝑛𝑔 =𝑛𝑜𝑛−𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑖𝑛𝑔 𝑙𝑜𝑎𝑛𝑠

(11)

The information on bank risk-taking is collected from the Compustat database.

Credit risk is not the only risk measure used in studies regarding bank risk-taking. For instance, Erkens et al. (2012) use equity risk and Keys et al. (2009) use portfolio risk. It is difficult to conclude which type of risk measure is most valid measure of bank risk-taking. In this study, we focus on credit risk, however, future research could examine the effects on different risk measures.

Control Variables

This study uses several control variables to reduce the effect of confounding variables.

Control variables are based on prior studies that have found an effect of these variables on our main variables of interest. Size is used since several studies have found that larger companies have more independent board members (Armstrong et al., 2010) and stronger risk controls (Ellul and Yeramilli, 2012). The deposits to assets ratio (D/A) is employed as a measure of banking activity. The assumption is that banks with higher deposits compared to other assets use a larger percentage of their assets for banking activities. These banks could have different risk management, bank risk-taking or board independence compared to banks with relatively low deposits.

In addition, several measures that influence the risk levels of the bank are taken into account. Return on assets (ROA) is used as a measure of profitability; more profitable companies are more likely to take more risk (John et al., 2000). Liabilities to assets (L/A) ratio is chosen to control for the debt-equity ratio; highly leveraged banks entail more risk that could influence the risk management or the board independence (Keys et al., 2009). The reserves to impaired loans ratio (R/ I L) is chosen since banks with high reserves compared to their loans entail less risk than banks without these reserves. The capital-assets ratio (C/A) is included as a measure of risk, since banks with higher risk might employ stronger risk management or employ more independent board members (Ellul and Yeramilli, 2013; Armstrong et al., 2010).

Sample

(12)

with more than 10 billion US dollars revenue. Through the BHC Peer Reports from the Federal Reserve System Information Center a sample is created which consists of the largest BHCs. The reason this study focuses on the US market is that data on BHCs is widely available and easy to obtain. A disadvantage is that using the US only, the generalisability of the results is limited. This study focuses on the years 2008-2016 to examine the most recent developments in corporate governance and their effects on bank risk-taking. The original sample consists of 108 BHCs.

Table 1 – Sample Selection

Initial Sample 108

Company years missing 4

Missing observations 19

Final Sample 85

Using the BHC Peer Reports of the period 2008-2016 with a revenue of more than 10 billion US dollars, a sample of 108 BHCs is created. First the data from the Compustat database is collected. Since the analyses require at least three lags, all BHCs with fewer than four year observations are deleted. These BHCs either failed, or existed for less than four years. Which led to the removal of four BHCs.

After the data collection from the Compustat database, the other data from BoardEx and Execucomp is collected. Every BHC that has missing observations for two or more databases, is removed from the sample. This leads to the final sample of 85 BHCs. A list of the examined BHCs is included in Appendix A.

Descriptive Statistics

(13)

The independence of the board is 79% on average. The level of board independence is in line with other studies on board independence within banks (Erkens et al., 2012; IMF, 2014). Whereas the risk-taking percentage is 1% on average, which is in line with the study of Ellul and Yeramilli (2013).

Table 2 – Descriptive Statistics

Table 2 present the descriptive statistics for the main variables used in the analyses. All variable definitions can be found in Appendix B.

Mean Std. Dev. Min Max N

CRO Present 0.812 0.391 0 1 750

CRO Executive 0.783 0.413 0 1 750

CRO Top5 0.196 0.397 0 1 744

CRO Centrality 0.104 0.227 0 0.967 744

Exp. Risk Committee 0.612 0.488 0 1 750

Active Risk committee 0.281 0.45 0 1 750

RMI 0.000 1.619 -2.873 3.601 744

Board Independence 0.79 0.174 0 1 750

Bank Risk Taking 0.01 0.015 0 0.108 750

Size 119527.3 360904.5 556.24 2573126 750 ROA 0.626 1.16896 -16.20 3.678 750 (C/A) 15.217 3.426 9.1 44 750 (R/ I L) 1.934 6.466 0.053 148.572 750 (L/A) 0.885 0.305 0.619 0.967 750 (D/A) 0.723 0.116 0.033 0.889 750

Pearson Correlation Matrix

(14)

4. Methodology

The objective of this study is to examine the relationships between bank risk-taking, board independence and the strength of the risk management. To examine these relationships, the following models are tested;

𝐵𝑜𝑎𝑟𝑑 𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑒_𝑗,𝑡 = 𝛼 + 𝛽 ∗ 𝑅𝑀𝐼_{𝑗,𝑡−1}+ 𝛽 ∗ 𝐵𝑎𝑛𝑘 𝑅𝑖𝑠𝑘 𝑇𝑎𝑘𝑖𝑛𝑔_{𝑗,𝑡−1} +𝛽 ∗ 𝐵𝑜𝑎𝑟𝑑 𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑒_{𝑗,𝑡−1}+ 𝛽 ∗ 𝐵𝑜𝑎𝑟𝑑 𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑒_{𝑗,𝑡−2} + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠_{𝑗,𝑡−1}+ 𝜀 (3) 𝑅𝑀𝐼_𝑗,𝑡 = 𝛼 + 𝛽 ∗ 𝐵𝑜𝑎𝑟𝑑 𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑒_{𝑗,𝑡−1}+ 𝛽 ∗ 𝐵𝑎𝑛𝑘 𝑅𝑖𝑠𝑘 𝑇𝑎𝑘𝑖𝑛𝑔_{𝑗,𝑡−1}+ 𝛽 ∗ 𝑅𝑀𝐼_{𝑗,𝑡−1+}+ 𝛽 ∗ 𝑅𝑀𝐼_{𝑗,𝑡−2}+ 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠_{𝑗,𝑡−1}+ 𝜀 (4) 𝐵𝑎𝑛𝑘 𝑟𝑖𝑠𝑘 𝑡𝑎𝑘𝑖𝑛𝑔𝑗,𝑡 = 𝛼 + 𝛽 ∗ 𝐵𝑜𝑎𝑟𝑑 𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑒𝑗 ,𝑡−1+ 𝛽 ∗ 𝑅𝑀𝐼𝑗,𝑡−1 +𝛽 ∗ 𝐵𝑎𝑛𝑘 𝑅𝑖𝑠𝑘 𝑇𝑎𝑘𝑖𝑛𝑔_{𝑗,𝑡−1}+ 𝐵𝑎𝑛𝑘 𝑅𝑖𝑠𝑘 𝑇𝑎𝑘𝑖𝑛𝑔_{𝑗,𝑡−2} +𝛽 ∗ 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠_{𝑗,𝑡−1}+ 𝜀 (5)

In these models, the dependent variables in one model are the independent variables in the other models. Therefore, it is likely that the independent variable will be correlated with the past and possibly current error term. This phenomenon is called endogeneity. When endogeneity is present, an assumption for effective OLS regression is violated. According to Wintonki (2012), an estimation procedure more appropriate in the presence of endogeneity is the Generalized Method of Moments (GMM) procedure. In such a regression the endogenous variable is replaced with an instrumental variable. A valid instrument variable should be strongly correlated with the regressor and uncorrelated with the error term. Wintonki (2012) argues that GMM is more appropriate when endogeneity is present since firm-fixed effects are included to account for unobservable heterogeneity. In addition, unlike OLS, analysed

dependent variables can be influenced by prior values of independent variables. Lastly, the GMM procedure allows using past values as internal instruments to account for simultaneity. However, according to Roodman (2009), the GMM procedure can be complicated and therefore more likely to create invalid estimates.

(15)

which simplifies the estimation procedure. IV estimation is more efficient in estimating correct coefficients especially in smaller samples (Baum, et al. 2003). However, the errors must be homoscedastic to produce valid estimates.

The IV and GMM procedures both have their advantages and disadvantages. Therefore, this study performs a test proposed by Pagan and Hall (1983) that is able to test the

homoscedasticity of the errors. If the errors appear homoscedastic, IV is used. If the errors are not homoscedastic, the GMM procedure is used.

Both estimation procedures can suffer from bias due to autocorrelation (Roodman, 2009). Therefore, this study uses a heteroskedastic and autocorrelation-consistent estimation procedure for both IV or GMM.

Following recommendations of Wintonki (2012), this study uses lags of variables as instruments. However, the number of lags used as instruments differ per estimation. According to Baum et al. (2003), instruments chosen should adhere to the order condition. This condition states that the number of instruments should at least be equal to the number of endogenous regressors. When too few or too many instruments are included, over- or under-identification problems arise and regression estimates could be biased.

Multiple tests are employed to assess the strength and validity of the instruments. To test possible under-identification, the Anderson test statistic can be employed. This test has as null hypothesis the under-identification of the coefficients. In addition, Baum et al. (2003)

recommend a weak identification test, the Cragg-Donald Wald test. The weak identification test measures the extra bias introduced by the instrumental variables compared to OLS regression. High bias in the estimation means that the used instruments can be classified as weak. Stock et al. (2002) have created critical values, depending on the regression to assess the bias compared to an OLS regression.

(16)

5. Results

Instrument strength results

In the first stage of the analysis, OLS regression is used to determine the relation between the endogenous variable and the instruments chosen. As explained in the methodology section, a valid instrument variable should be strongly correlated with the regressor. Table 4 presents the results of this analysis and shows the strength of the instruments.

The instruments with the highest significance regarding RMI_𝑡−1 are the second and third order lag. The higher order lags of RMI are not significantly associated with RMI_𝑡−1. The lags chosen as instruments for Board Independence 𝑡−1, the second and fifth order lag are

significantly associated with Board Independence _𝑡−1. The lags chosen as instruments for Bank Risk − Taking _𝑡−1 are all significantly associated with the regressor.

The instruments from table 4 that are not strongly correlated with the regressor are still included, because the variable’s correlation with the regressor is not the only condition which determines instruments’ validity. Identification, redundancy and the Wald F-statistic are important aspects that influence the choice of instruments. Table 6, panel C presents the other tests employed to assess the strength of the instruments.

First, in the three estimations the Hansen J statistic is insignificant at the 5% level. This means that the results suggest that the over-identifying restrictions implied by the instruments are valid. Second, the Kleibergen-Paap LM statistic shows a significant p-value at the 1% level. Therefore, the results provide no evidence for under-identification, i.e. the instruments are sufficiently correlated with the endogenous variable. Third, the Wald F-statistic is larger than the critical value of Stock Yogo (2005) in the regressions regarding RMI and Board Independence. This means that in these regressions, the relative bias of GMM compared to OLS is below the 5% level. The Wald F-statistic is below the critical value of Stock Yogo (2005) in the estimation of bank risk-taking. Therefore, the bias of GMM compared to OLS is larger. This suggests that these results are less valid compared to the other regressions.

Choosing different lags with this estimation unfortunately does not solve the problem. Fourth, this study uses the Hansen (1983) test to examine the null hypothesis stating that endogenous variables should be treated as exogenous. The results suggest that this hypothesis should be rejected. Therefore, these results provide evidence that the instruments should be treated as endogenous. Lastly, the results from Breusch instrument relevance test suggest that the instruments included in the regressions are not redundant.

(17)

Table 4 –Instruments

Table 4 presents the OLS regression results of the first stage of the analysis. The *, **, and *** are used to denote the 1%, 5%, and 10% significance levels, respectively. Errors robust to heteroscedasticity are denoted between the brackets.

RMI𝑡−1 Board Independence 𝑡−1 Bank Risk − Taking 𝑡−1

Panel A: RMI instruments

RMI 𝑡−2 0.807*** (0.051) RMI𝑡−3 0.118* (0.061) RMI𝑡−4 -0.042 (0.059) RMI𝑡−5 -0.039 (0.047) Panel B: Board independence instruments

Board Independence 𝑡−2 0.0657 (0.050) Board Independence 𝑡−3 -0.191*** (0.052) Board Independence 𝑡−4 0.0334 (0.052) Board Independence 𝑡−5 0.609*** (0.076) Panel C: Bank Risk-Taking instruments

Bank Risk − Taking 𝑡−2 0.723***

(0.033)

Bank Risk − Taking 𝑡−3 -0.073**

(0.035)

Bank Risk − Taking 𝑡−4 0.056**

(0.024)

Bank Risk − Taking 𝑡−5 -0.038***

(0.014)

Constant 0.123*** 0.312*** 0.001***

(0.047) (0.064) (0.000)

Observations 320 320 320

(18)

Pagan and Hall (1983) results

Table 5 presents the results of the test described by Pagan and Hall (1983) used to determine the estimation procedure. All test statistics are insignificant, therefore it is likely that

heteroscedasticity is present in the errors. When heteroscedasticity is present, GMM procedure is more appropriate than IV estimation. Therefore, I use GMM estimation to examine the relations in this study.

Table 5 - Pagan Hall test statistics

This table presents the results of the test described by Pagan and Hall (1983).

RMI Bank Risk Taking Board Independence

Pagan Hall Test statistic 29.82 87.709 52.60

Pagan Hall p-value 0.019 0.000 0.000

Hypothesis results

Table 6 presents the two-stage GMM results for equations 3, 4 and 5. Based on these results the hypotheses are discussed.

First, hypothesis 1a declares that the strength and quality of risk management has a positive effect on the levels of board independence. The results suggest that there is no significant association between the RMI and board independence. Therefore, results provide no

supporting evidence for this hypothesis and this hypothesis is rejected. An explanation could be that the independent board members do not take the risk management systems into account in their decision to join the board of the BHC. Another explanation could be that the BHC is not transparent regarding its risk management and is therefore not taken into account by independent board members.

Second, hypothesis 1b assumes that the strength and quality of risk management has a negative effect on the levels of bank risk-taking. The beta of the association between the RMI and bank risk-taking is not significantly different from zero. Consequently, the results provide no supporting evidence for a negative effect on bank risk-taking. Therefore, hypothesis 1b is rejected. An explanation could be that the RMI is not able to capture the real efforts in the risk management area of the BHC or that new regulation limits the possibilities of banks to take excessive risks.

(19)

hypothesis 2a and the hypothesis is rejected. An explanation for this relation could be, in line with Haan and Valhu (2015), that banks are complex organizations. Independent board

members less involved in the bank and its functioning might be unable to prevent risk outside the risk appetite. On the other hand, independent board members might be able to improve the transparency from the BHC and might therefore be more transparent and conservative

regarding the risk levels of the BHC. Another possible argument, further developed in the conclusion is that independent board members more likely to satisfy the shareholders and therefore encourage risk-taking.

Fourth, hypothesis 2b assumes that higher levels of board independence have a positive effect on the strength and quality of risk management. The results provide supporting

evidence for this hypothesis, since board independence is significantly positively associated with RMI. Therefore, the results of the analysis suggest that higher levels of board

independence lead to a stronger risk management system.

Fifth, hypothesis 3a declares that bank risk-taking has a negative effect on the strength and quality of risk management and board independence. The results suggest that higher levels of bank risk-taking lead to lower levels of strength and quality of the risk management function. Therefore, the results provide supporting evidence for this hypothesis. Consequently, I fail to reject hypothesis 3a.

(20)

Table 6 - Results table

Table 6 presents the regression results of the two-stage GMM analysis of equation 3, 4 and 5. The *, **, and *** are used to denote the 1%, 5%, and 10% significance levels, respectively. Errors robust to heteroscedasticity are denoted between the brackets. The estimates presented are autocorrelation consisted. In this regression, the main variables in panel A are

instrumented using the instruments from table 4. Panel B shows the results of the control variables. In these regressions the control variables:(L⁄A), (D/A), (R ⁄ I L), Size, ROA and (C ⁄ A) are included as well. To save space, the results of these control variables are shown in Appendix C. The statistics presented in panel C are used to determine the robustness of the method and instruments used.

RMI (3)

Board Independence (4)

Bank Risk Taking (5) Panel A: Main variables

RMI 𝑡−1 0.005 0.000

(0.005) (0.000)

Board Independence 𝑡−1 0.859*** 0.655***

(0.055) (0.0867)

Bank Risk − Taking 𝑡−1 -0.808** 1.886**

(0.358) (0.812)

Panel B: Control Variables

RMI 𝑡−1 -6.926

(4.270)

RMI 𝑡−2 0.048

(0.054)

Bank Risk − Taking 𝑡−1 0.001

(0.001)

Bank Risk − Taking 𝑡−2 -0.043

(0.054) Board Independence 𝑡−1 0.318*** (0.034) Board Independence𝑡−2 0.059* (0.031) Constant 0.712 0.017*** -0.309 (1.819) (0.006) (0.279) Observations 320 320 320 R-squared 0.777 0.811 0.296

Panel C: Instrument strength

Hansen J statistic 7,497 6.984 12.021

Hansen p-value 0.2773 0.3223 0.062

Kleibergen-Paap LM statistic 64.643 106.706 63.406

Kleibergen-Paap Wald p-value 0.000 0.000 0.000

Kleibergen-Paap Wald F-statistic 18.980 76.554 13.674

Stock-Yogo critical value 17.70 17.70 17.70

Hausmann Test p-value 0.019 0.000 0.000

(21)

Robustness Tests

The results of the robustness tests are reported in Appendix D. Only the main variables of interest are presented, full results can be provided upon request by the author.

Implementing board expertise

This study examines board independence, however, other board attributes might have a significant impact on bank risk-taking as well. One particular attribute of interest is the expertise of the board members of the bank. Walker (2009) argues that board expertise is important, since banks are opaque and complex organizations. Therefore, the board members should have sufficient knowledge and experience to effectively manage risk. Walker (2009) states that board members who have more financial experience are better able to assess the effect of bank policies on bank's risk-taking. Therefore, boards with less financial expertise could be less effective in installing a strong and high-quality risk management function in a bank. As an additional analysis, this study uses board financial expertise to examine the effects of including a different board attribute.

Board experience is defined following Minton (2014) and is a percentage variable that indicates the percentage of independent board members has financial expertise compared to the total number of directors. A board director is classified as a financial expert if he or she: (1) has worked at a bank or;

(2) currently works at a financial institution or;

(3) has finance-related role within a non-financial company or; (4) works at an academic institution related to finance or; (5) is a professional investor.

Information about this experience of board directors is collected via BoardEx and directors are classified based on the criteria above.

Table 7 presents the results of the test where board independence is replaced by board expertise. The main finding is that bank risk-taking is negatively associated with board expertise, as well as board independence. Therefore, the results suggest that former bank risk-taking levels and thereby possibly the risk culture also influences the level of board expertise. This further strengthens the main finding of this paper, namely bank risk-taking levels

(22)

levels. Therefore, the results suggest the higher levels of board expertise does not have the same consequences for bank-risk taking as board independence does.

Crisis Years

The financial crisis of 2008 had disrupting effects for BHCs. To ensure that the results of this study are not driven by variables impacted by extreme results, the sample is divided into two subsamples. One sample from 2008-2011 is used to examine the associations between the main variables within the crisis. The period 2008-2011 is chosen since our analysis requires at least three lagged variables. In future research, other years could be chosen to reflect the financial crisis of 2008. Another subsample, 2012-2018, is used to examine the relations between the main variables within non-crisis years.

Tables 8 present the results of the period 2008-2011 and 2012-2016. The results suggest that in the crisis period, as well as in the non-crisis period the levels of bank risk-taking significantly influence the levels of board independence and RMI. This results is in line with the main analysis, since this relation does not appear to be influenced by the time period chosen. On the other hand, the board independence is not significantly related with levels of bank-risk taking in the crisis period. This result is therefore less robust, as the results suggest this relation is different in financial crises.

Risky vs. Non-risky BHCs

Another possible limitation within the main analyses is that the effects are mostly driven by high risk-taking BHCs. Therefore, this study divides the sample into two subsamples. The first subsample contains the 40% highest risk-taking BHCs. The other subsample contains the lowest 60% risk-taking BHCs. The results of these subsamples are compared to examine whether the results found in the main analysis are different for risky or non-risky BHC’s.

(23)

6. Discussion and Conclusion

This study examined the relations between board independence, the RMI and bank-risk taking. The results suggest a positive relation between bank risk-taking and board

independence. In addition, the results of this study suggest that higher levels of bank risk-taking lead to lower levels of strength and quality of the risk management function. This result is in line with the paper of Wintonki (2012). This results suggest that BHCs take former bank risk-taking levels into account regarding the level of independent board members and the strength of the risk management. The robustness tests suggest that this result does not differ based on a different time-period or riskiness level. Also, an another board attribute appears to lead to the same conclusion regarding the influence of bank risk-taking levels. Therefore, the main finding of this study is that former levels of bank risk-taking influence the decisions regarding corporate governance.

The interpretations of the findings above are in line with the ones from Ellul and Yeramilli (2013) and Fahlenbrach (2012). Namely, it could be that the bank risk-taking levels are an indicator of the risk culture within the BHC. Possibly, BHCs adjust their corporate governance mechanisms on the basis of their needs and risk culture. As stated by hypothesis 3a, high bank risk-taking levels, and thereby an aggressive risk culture, leads to weaker risk management. The results of this study find supporting evidence for this hypothesis.

In line with this argument, BHCs with an aggressive risk culture might want to appoint board members that will not discourage the risk-taking. According to Haan and Vlahu (2015), independent board members are more likely to encourage risk-taking by banks. This is

because independent board members are more likely to satisfy the shareholders, since they do not have personal or monetary ties with management. These board members are aware that increasing risk-levels lead to higher returns for the shareholders at the expense of the other stakeholders. Therefore, high bank risk-taking levels, as indication of high risk-taking culture, might lead to an increase in the levels of independent board members.

Furthermore, the results of this study suggest that the level of board independence positively influences the bank risk-taking levels. This results is not in line with prior research of Erkens et al. (2012) or the IMF (2014). However, this result could be interpreted within the argument above. As explained, independent board members know that increasing risk-levels leads to higher returns for the shareholders at the expense of the other stakeholders.

(24)

In addition, the results of this study suggest that higher levels of board independence lead to a stronger risk management function. This finding is in line with prior literature of Erkens et al. (2012) and the IMF (2014). However, the robustness tests suggest that this relation is not present in subsamples. Therefore, this result should be interpreted with caution.

Furthermore, the results of this study do not support the findings of Keys (2009) since this study does not find evidence for a negative association between the risk management index and the bank risk-taking levels.

The results of the robustness tests suggest that some findings appear to be different for the subsamples examined. This poses limitations on the generalizability of this study. Another limitation is choosing a sample of the largest BHCs listed on the NYSE Exchange. Therefore, the results are not generalizable regarding other countries, or smaller BHCs. In addition, the measures impose their limitations to our results. As mentioned in the methodology section, it is difficult to obtain the true strength of the risk management function, since this requires substantive observational research. Lastly, this study uses credit risk as a measure of bank risk-taking. However, there are multiple measures of bank-risk taking which might lead to different results.

Future research could further investigate the relations between bank risk-taking and corporate governance decisions. Especially studies including both bank risk-taking levels and risk culture could enrich the understanding of corporate governance within banks.

Furthermore, the results of this study suggest that endogeneity are important for bank risk-taking and corporate governance studies. Therefore, the recommendation is to use estimation procedures that account for endogeneity. Lastly, future research could focus on constructing valid and relevant measures of risk management within banks.

On the one hand, this study made a contribution to the literature on corporate

(25)

7. References

Aebi, V., Sabato, G., & Schmid, M. (2012). Risk management, corporate governance, and bank performance in the financial crisis. Journal of Banking & Finance, 36(12), 3213-3226. Armstrong, C. S., Guay, W. R., & Weber, J. P. (2010). The role of information and financial reporting in corporate governance and debt contracting. Journal of Accounting and

Economics, 50(2-3), 179-234.

Avraham, D., Selvaggi, P., & Vickery, J. I. (2012). A structural view of US bank holding companies.

Baker, M., & Wurgler, J. (2015). Do strict capital requirements raise the cost of capital? Bank regulation, capital structure, and the low-risk anomaly. American Economic Review, 105(5), 315-20.

Baum, C. F., Schaffer, M. E., & Stillman, S. (2003). Instrumental variables and GMM: Estimation and testing. Stata Journal, 3(1), 1-31.

Beltratti, A., & Stulz, R. M. (2012). The credit crisis around the globe: Why did some banks perform better? Journal of Financial Economics, 105(1), 1-17.

Bolton, P., Mehran, H., & Shapiro, J. (2015). Executive compensation and risk-taking. Review

of Finance, 19(6), 2139-2181.

Breusch, T., Qian, H., Schmidt, P., & Wyhowski, D. (1999). Redundancy of moment conditions. Journal of Econometrics, 91(1), 89-111.

De Haan, J., & Vlahu, R. (2016). Corporate governance of banks: A survey. Journal of

Economic Surveys, 30(2), 228-277.

Ellul, A., & Yerramilli, V. (2013). Stronger risk controls, lower risk: Evidence from US bank holding companies. The Journal of Finance, 68(5), 1757-1803.

Erkens, D. H., Hung, M., & Matos, P. (2012). Corporate governance in the 2007–2008 financial crisis: Evidence from financial institutions worldwide. Journal of Corporate

Finance, 18(2), 389-411.

Fahlenbrach, R., Prilmeier, R., & Stulz, R. M. (2012). This time is the same: Using bank performance in 1998 to explain bank performance during the recent financial crisis. The

Journal of Finance, 67(6), 2139-2185.

Hagendorff, J., & Vallascas, F. (2011). CEO pay incentives and risk-taking: Evidence from bank acquisitions. Journal of Corporate Finance, 17(4), 1078-1095.

Haubrich, J. G. (1998). Bank diversification: laws and fallacies of large numbers. Economic

Review-Federal Reserve Bank of Cleveland, 34, 2-9.

(26)

Hermalin, B. E., & Weisbach, M. S. (2001). Boards of directors as an endogenously

determined institution: A survey of the economic literature (No. w8161). National Bureau of

Economic Research.

International Monetary Fund (2014). Risk-taking by banks: The role of governance and executive pay. In Global Financial Stability Report: Risk-taking, liquidity, and shadow

banking: Curbing excess while promoting growth. Washington, DC: IMF.

John, K., Saunders, A., & Senbet, L. W. (2000). A theory of bank regulation and management compensation. The Review of Financial Studies, 13(1), 95-125.

Keys, B. J., Mukherjee, T., Seru, A., & Vig, V. (2009). Financial Regulation and

Securitization: Evidence from subprime loans. Journal of Monetary Economics, 56(5), 700-720.

Louzis, D. P., Vouldis, A. T., & Metaxas, V. L. (2012). Macroeconomic and bank-specific determinants of non-performing loans in Greece: A comparative study of mortgage, business and consumer loan portfolios. Journal of Banking & Finance, 36(4), 1012-1027.

Minton, B. A., Taillard, J. P., & Williamson, R. (2014). Financial expertise of the board, risk-taking, and performance: Evidence from bank holding companies. Journal of Financial and

Quantitative Analysis, 49(2), 351-380.

Pagan, A. R., & Hall, A. D. (1983). Diagnostic tests as residual analysis. Econometric

Reviews, 2(2), 159-218.

Pathan, S. (2009). Strong boards, CEO power and bank risk-taking. Journal of Banking &

Finance, 33(7), 1340-1350.

Roodman, D. (2009). How to do xtabond2: An introduction to difference and system GMM in Stata. The Stata Journal, 9(1), 86-136.

Ross, S. A. (1973). The economic theory of agency: The principal's problem. The American

Economic Review, 63(2), 134-139.

Schultz, E. L., Tan, D. T., & Walsh, K. D. (2010). Endogeneity and the corporate governance-performance relation. Australian Journal of Management, 35(2), 145-163.

Stock, J. H., Wright, J. H., & Yogo, M. (2002). A survey of weak instruments and weak identification in generalized method of moments. Journal of Business & Economic

Statistics, 20(4), 518-529.

Srivastav, A., & Hagendorff, J. (2016). Corporate governance and bank risk‐taking. Corporate

Governance: An International Review, 24(3), 334-345.

Walker, D. (2009). A review of corporate governance in UK banks and other financial industry entities.

(27)

8. Appendix

Appendix A – Analyzed Bank Holding Companies

BHC Period included Assets ($, 2016) JPMorgan Chase & Co. 2008-2016 2490972

Wells Fargo & CO 2008-2016 1930115

Citigroup INC 2008-2016 1792077

Bank of Montreal 2008-2016 687935

US Bancorp 2008-2016 445964

PNC Financial Services Group, Inc. 2008-2016 366380 Capital One Financial Corporation 2008-2016 357033 Bank of New York Mellon Corp 2008-2016 333469

State Street Corporation 2008-2016 242698

BB&T Corp 2008-2016 219276

SunTrust Banks, Inc. 2008-2016 204875

Ally Financial Inc. 2012-2016 163728

Fifth Third Bancorp 2008-2016 142177

Santander Holdings USA, Inc 2008-2016 137370,5

KeyCorp 2008-2016 136453

Regions Financial Corp 2008-2016 125968

Northern Trust Corp 2008-2016 123926,9

M & T Bank Corp 2008-2016 123449,2

Huntington Bancshares 2008-2016 99714,1

Discover Financial Services 2008-2016 92308

Synchrony Financial 2013-2016 90207

Comercia INC 2008-2016 72978

CIT Group Inc. 2008-2016 64170,2

Zions Bancorporation 2008-2016 63239,17

E*TRADE Financial Corporation 2008-2016 48999 New York Community Bancorp, Inc. 2008-2016 48926,56

SVB Financial Group 2008-2016 44683,66

People's United Financial, Inc 2008-2016 40609,8

Popular Inc. 2008-2016 38661,61

East West Bancorp Inc. 2008-2016 34788,84

First Citizens Bancshares 2008-2016 32990,84 BOK Financial Corporation 2008-2016 32772,28

City National Corp 2008-2014 32610,36

Cullen/Frost Bankers Inc. 2008-2016 30196,32

Synovus Financial Corp. 2008-2016 30104

Associated Banc-Corp 2008-2016 29139,32

First Horizon National Corp 2008-2016 28555,23

(28)

Webster Financial Corporation. 2008-2016 26072,53 Wintrust Financial Corp. 2008-2016 25668,55 Commerce Bancshares Inc. 2008-2016 25641,42

First Merit Corp 2008-2015 25524,6

Umpqua Holdings Corporation 2008-2016 24813,12 Investors Bancorp, Inc. 2008-2016 23174,68 Valley National Bancorp. 2008-2016 22864,44 Prosperity Bancshares, Inc. 2008-2016 22331,07

PacWest Bancorp 2008-2016 21869,77

F.N.B. Corporation 2008-2016 21844,82

Texas Capital Bancshares, Inc. 2008-2016 21697,13

IberiaBank Corporation 2008-2016 21659,19

UMB Financial Corporation 2008-2016 20682,53

MB Financial, Inc. 2008-2016 19302,32

Stifel Financial Corp. 2008-2016 19129,36

Fulton Financial Corp. 2008-2016 18944,25

Chemical Financial Corporation 2008-2016 17355,18 Western Alliance Bancorporation 2008-2016 17200,84

Bank of Hawaii Corp 2008-2016 16492,37

Washington Federal, Inc. 2008-2016 14888,06

Old National Bancorp 2008-2016 14860,24

BancorpSouth 2008-2016 14724,39

Astoria Financial Corporation 2008-2016 14558,65

Cathay General Bancorp 2008-2016 14520,77

Flagstar Bancorp, Inc. 2008-2016 14053

Hope Bancorp, Inc. 2008-2016 13441,42

Trustmark Corp 2008-2016 13352,33

TFS Financial Corporation 2008-2016 12914,08

Hilltop Holdings 2008-2016 12738,06

First Midwest Bancorp, Inc. 2008-2016 11422,56 Pinnacle Financial Partners 2008-2016 11194,62 Banc of California, Inc. 2008-2016 11029,85 United Community Banks Inc. 2008-2016 10708,66

WesBanco Bank, Inc. 2008-2016 9790,877

National Penn Bancshares, Inc. 2008-2015 9598,902 Provident Financial Services Inc. 2008-2016 9500,465

NBT Bancorp Inc. 2008-2016 8867,268

Renasant Corp. 2008-2016 8699,851

Simmons First National Bank 2008-2016 8400,056 Park National Corporation 2008-2016 7467,586 Taylor Capital Group Inc. 2008-2013 5685,818

1st Source Corporation 2008-2016 5486,268

(29)

PrivateBancorp, Inc. 2008-2015 4495,502

First Bancorp 2008-2016 3614,862

Sun Bancorp, Inc. 2008-2016 2262,262

United Bankshares, Inc. 2008-2016 633,119

Appendix B – Variable definitions

Variable Definition

RMI Risk Management Index, measures the

strength and independence of the risk management function (see variables section).

Board Independence The number of independent directors of

the board divided by the total number of board members.

Bank Risk-Taking The fraction of non-performing loans

divided by total assets.

(L A⁄ ) Total liabilities divided by total assets.

(D/A) Total (customer) deposits divided by

total assets.

(R I L⁄ ) The total amount of reserves for loan

losses divided by the total amount of impaired loans.

Size The total amount of assets.

ROA Return on assets, net income divided by

total assets.

(C A) ⁄ The total amount of Tier-1 capital

divided by total assets.

Board Expertise The number of board members classified

(30)

Appendix C – Additional Tables

Table 3 – Pearson Correlation Matrix

This table presents the Pearson correlations between the main variables of this study. The *, **, and *** are used to denote the 1%, 5%, and 10% significance levels, respectively.

RMI

Bank

Risk-Taking Board Indep (L/A) (D/A) (R/I L) ROA (C/A) Size

RMI 1

Bank Risk Taking -0,223*** 1

(31)

Table 6 – Control Variables

Table 6 presents the regression results of the GMM analysis of equation 3,4 and 5. The *, **, and *** are used to denote the 1%, 5%, and 10% significance levels, respectively. Errors robust to heteroscedasticity are denoted between the brackets. The estimates presented are autocorrelation consisted. In this regression, the main variables in panel A are instrumented using the instruments from table 4. Panel B shows the results of the control variables.

RMI (3)

Board Independence (4)

Bank Risk Taking (5) Panel A: Main instrumented

variables

RMI𝑡−1 0.005 0.000

(0.005) (0.000)

Board Independence 𝑡−1 0.859*** 0.655***

(0.055) (0.0867)

Bank Risk − Taking 𝑡−1 -0.808** 1.886**

(0.358) (0.812)

Panel B: Control Variables

RMI𝑡−1 -6.926 (4.270) RMI𝑡−2 0.048 (0.054) Board Independence 𝑡−1 0.318*** (0.034) Board Independence𝑡−2 0.059* (0.031) 0.001

Bank Risk − Taking 𝑡−1 (0.001)

-0.043

Bank Risk − Taking 𝑡−2 (0.054)

(32)

Appendix D – Results from Robustness Tests

Table 7 – Board Expertise

Table 7 presents the regression results of the two-stage GMM analysis, including board expertise. The *, **, and *** are used to denote the 1%, 5%, and 10% significance levels, respectively. Errors robust to heteroscedasticity are denoted between the brackets. The estimates presented are autocorrelation consisted. In this regression, the main variables in panel A are instrumented using the instruments from table 4. Panel B shows the results of the control variables. In these regressions the control variables: (L⁄A), (D/A), (R ⁄ I L), Size, ROA and (C ⁄ A) are included as well. To save space, the results of these control variables are available upon request. The statistics presented in panel C are used to determine the robustness of the method and instruments used.

RMI Board Expertise Bank Risk Taking

Panel A: Main instrumented variables

RMI𝑡−1 0.001 0.000

(0.002) (0.000)

Board Expertise 𝑡−1 0.213 -0.001

(0.231) (0.001)

Bank Risk − Taking 𝑡−1 -0.858*** 0.849***

(0.0515) (0.051)

Panel B: Control variables

RMI𝑡−2 0.0420

(0.0495)

Board Expertise 𝑡−2 0.098**

(0.050)

(0.057)

Constant 0.378 0.007 0.015***

(1.730) (0.105) (0.006)

R-squared 0.777 0.934 0.814

Hansen J statistic 1.496 5.567 9,624

Hansen p-value 0.9598 0.473 0.141

Kleibergen-Paap LM p-value 0.000 0.000 0.000

(33)

Table 8 – Crisis Years versus non-crisis years

Table 8 presents the regression results of the two-stage GMM analysis for the period 2008-2011 and 2012-2016. The *, **, and *** are used to denote the 1%, 5%, and 10% significance levels, respectively. Errors robust to heteroscedasticity are denoted between the brackets. The estimates presented are autocorrelation consisted. In this regression, the main variables in panel A are instrumented using the instruments from table 4. Panel B shows the results of the control variables. In these regressions the control variables: (L⁄A), (D/A), (R ⁄ I L), Size, ROA and (C ⁄ A) are included as well. To save space, the results of these control variables are available upon request. The statistics presented in panel C are used to determine the robustness of the method and instruments used.

Crisis period (2008-2011)

RMI Board Independence Bank Risk Taking

0.0102 -0.001

(0.006) (0.001)

Board Independence 𝑡−1 -1.420 0.010

(1.305) (0.010)

(0.164) (0.085)

RMI𝑡−2 0.355**

(0.140)

Board Independence 𝑡−2 -0.027

(0.079)

(0.105)

Constant 1.663 0.350 -0.066*

(4.913) (0.239) (0.036)

R-squared 0.662 0.843 0.792

Hansen J statistic 0.000 1.599 0.238

Hansen p-value 0.983 0.450 0.888

(34)

Non-crisis period (2012-2016)

RMI𝑡−1 -0.002 0.000

(0.004) (0.000)

Board Independence 𝑡−1 0.763 0.006**

(0.734) (0.003)

(0.064) (0.0555)

RMI𝑡−2 0.0168

(0.0538)

Board Independence 𝑡−2 0.394***

(0.065)

Bank Risk − Taking 𝑡−2 -0.095*

(0.0494)

Constant 4.030 -0.442* 0.014*

(2.844) (0.245) (0.008)

R-squared 0.803 0.689 0.809

Hansen p-value 0.377 0.677 0.2751

Stock-Yogo critical value 5% 5.44 11.04 13.14

Table 9 – Risky BHCs versus less risky BHCs

Table 9 presents the regression results of the two-stage GMM analysis. The *, **, and *** are used to denote the 1%, 5%, and 10% significance levels, respectively. Errors robust to

heteroscedasticity are denoted between the brackets. The estimates presented are

autocorrelation consisted. In this regression, the main variables in panel A are instrumented using the instruments from table 4. Panel B shows the results of the control variables. In these regressions the control variables:(L⁄A), (D/A), (R ⁄ I L), Size, ROA and (C ⁄ A) are included as well. To save space, the results of these control variables are available upon request. The statistics presented in panel C are used to determine the robustness of the method and instruments used.

Risky BHCs

(35)

RMI𝑡−1 0.025 0.000

(0.014) (0.000)

Board Independence 𝑡−1 -0.607 0.003*

(0.379) (0.002)

(0.060) (0.062)

RMI𝑡−2 0.004

(0.041)

Board Independence 𝑡−2 0.041

(0.058)

(0.063)

Constant -3.312 -0.442* -0.012

(2.316) (0.245) (0.014)

R-squared 0.807 0.098 0.844

Hansen p-value 0.479 0.210 0.664

Stock-Yogo critical value 5% 4.06 11.04 5.44

Non-risky BHCs

RMI𝑡−1 0.026 -0.000

(0.015) (0.000)

Board Independence 𝑡−1 -1.152** 0.005*

(0.513) (0.002)

(0.058) (0.075)

RMI𝑡−2 0.052

(0.140)

Board Independence 𝑡−2 0.048

(0.055)

(0.091)

Constant 2.101 0.149 0.027***

(2.389) (0.424) (0.009)

(36)

R-squared 0.522 0.116 0.753 Panel C: Instrument strength

Hansen p-value 0.115 0.204 0.344