Effect of bank competition on financial stability

Effect of bank competition on financial stability

E.A. Helling1 Studentnr: s1824260 Master’s Thesis Finance Supervisor: K. Roszbach

10th January 2014

Abstract

This thesis examines the relation between bank risk and the degree of competition for 5,222 banks located in the 28 member countries of the European Union, for the period 2001 to 2011. I define bank

risk as either overall bank risk or loan portfolio risk, and infer the degree of competition using the Lerner index, and a more novel competition measure, the Boone indicator. This study takes into account the effect of regulatory, supervisory and institutional factors on bank risk. I use both an OLS technique as IV technique with GMM estimator to control for potential endogeneity of the competition

measure. The results suggest that banks with more market power have less risky loan portfolios, in support of the competition fragility view. The data provide no evidence for a relation between

competition and overall bank risk.

Keywords: Bank competition; Financial stability; Regulation; Banks in Europe

JEL Codes: G21; G28

1

1

1. Introduction

To properly design competition policy for the banking sector, it is important to understand the effect of bank competition on financial stability. Two different views about this effect have emerged in the literature: the competition fragility view and the competition stability view. In the competition fragility view, an increase in competition is supposed to decrease the market power of banks, leading to lower interest rates. The lower interest rates subsequently lead to lower profit margins for the bank. If the bank wants to keep the same return, it will have to take on more risk leading to a less stable financial system (Keely, 1990; Allen & Gale, 2004; Carletti 2008). In the competition stability view, competition also erodes market power of the bank, and the bank will demand lower interest rates. The lower interest rates have a positive effect on potential agency problems; moral hazard declines as borrowers will not have to take on riskier project to be able to repay the interest rate and adverse selection reduces as the pool of loan applicants does not consist of only risky borrowers. Under the competition stability view, increased competition leads to a more stable financial system (Boyd and DeNicolo, 2005). These two views are both backed by empirical research, and therefore it still not conclusive which dominates.

2 In each of these studies different datasets, methodologies and definitions of competition and risk are used, leading to mixed results. More research is required to resolve the competition-stability/fragility debate. In this thesis, I use different measures of competition and risk in order to find the effects of bank competition on bank risk taking for banks located in the 28 member countries of the European Union for the period 2001 to 2011. Market, institutional and regulatory variables are taken into account to assess whether they have influence on the relation between competition and stability. The different aspects of the competition and bank risk taking relation are important when designing competition and macro prudential policy for the European Union.

This paper contributes to existing literature by analyzing whether the effect of bank competition on financial stability is stable through time for EU-countries. The study includes years in which the banking sector suffered from the financial crisis to test whether the relation of competition on risk-taking is the same for crisis and non-crisis years. To assess overall bank risk, I use the score. The Z-score is based on accountancy information and can therefore be calculated easily for the banks in the sample. To analyze the effect of the degree of competition on the loan portfolio separately, I include the ratio of non-performing loans to total loans as dependent variable following Berger et al. (2009). Finally, I infer the degree of competition using the Lerner index and the Boone indicator. The Lerner index, which determines the individual price setting power by the company’s ability to raise prices above its marginal cost, takes into account actual market conditions. It includes both interest income and non-interest income, and therefore reflects competition in broader banking activity making it an appropriate measure of bank competition (Carbo et al., 2009). However, several studies doubt the robustness of the Lerner index as competition measure as the Lerner index might increase due to more intense competition (e.g. Amir, 2003), leading to wrong inferences. Therefore, in addition to the Lerner index, I use the Boone indicator as competition measure. The Boone indicator is a more innovative measure of competition and is based on the idea that more efficient firms have higher profits or market shares, and that this effect is increasing in the degree of competition (Boone, 2001). Leuvensteijn (2008) finds that the Boone indicator is better able to identify different regimes of competition than the elasticity-adjusted Lerner index for a sugar-factory company. The Boone indicator has not been applied much in banking studies yet, and is therefore a relative new measure of competition for the banking sector.

I find evidence for a positive relation between loan portfolio risk and the degree of competition, in support of the competition fragility view. The data provide no evidence for a relation between bank risk and the degree of competition.

3

2. Literature Review

Two different strands of literature have emerged over time considering the impact of bank competition on financial stability: the competition fragility view and the competition stability view. Section 2.1 and Section 2.2 contain the relative literature on respectively, the competition fragility view and the competition stability view.

2.1 Competition fragility view

Keeley (1990) proposes a theoretical empirical framework to explain the bank defaults in the 1980s following the US deregulations in the 1970s. Deregulation of the banking sector leads to an increase in competition which makes it more difficult for banks to earn monopoly rents (the returns available for the managers and shareholders). The reduction in the monopoly rents subsequently lead to an increase in agency problems between the bank owner and the government deposit insurance fund. The bank owner has an incentive to take on more risk; the bank enjoys a guarantee of funds due to the deposit insurance and only has a limited downside as the bank has a non-convex pay-off function. The quality of the loan portfolio deteriorates as more marginal loan applicants receive financing leading to an increase in financial instability. The failures of banks in the 1980s following the US deregulations support this line of reasoning and indicate that there is a negative relationship between bank competition and financial stability.

A large amount of literature follows Keeley’s paper stressing the importance of proper bank regulation to mitigate the negative effect of bank competition on risk taking incentives of banks (e.g. Allen & Gale, 2004). The moral hazard effect might even be amplified by the size of the banks; big banks might be too-big-to fail, too-interconnected-to fail, or even too-complex-to fail. The implicit (or explicit) subsidies the banks receive via government safety nets may lead to more excessive risk-taking (see e.g. Beck et al, 2013).

2.2 Competition stability view

The second strand of literature is initiated by Boyd and De Nicolo (2005). By including some different features to existing models, they find that the result of the negative trade-off between competition and stability is not robust. The risk of the bank’s loan portfolio decreases when entrepreneurs (the borrowers of the bank) choose the risk of the investment project conditional on the loan rate set by the bank. A bank with a higher degree of market power demands a higher loan rate, and the entrepreneurs confronted with these higher funding costs will optimally choose higher risk projects. Thus, in reverse; banks with a lower degree of market power demand lower loan rates, and subsequently investors will invest in lower risk projects.

4 government bond. The model predicts a negative relationship between competition and financial stability. Using a cross-sectional sample of US banks for the year 2003 and a sample of banks located in 134 non-industrialized countries for the years 1993-2004, they find that the probability of default (the Z-score) is negatively related to concentration, supporting their model. They also show that the loan to asset ratio is negatively related with concentration; indicating that when concentration increases, the amount of total assets that is put to lending increases. The economic meaningfulness of their findings can however be doubted. Although the correlation coefficient between concentration and the Z-score is significantly negative, its value is only -0.06. The same holds for the relation between the parameter estimate of concentration in relation to financial stability, which value is respectively -0.0004. Moreover, it can be noted that Boyd, De Nicolo and Jalal use concentration as a measure of competition, which is not a reliable indicator of competition (Beck, Demirguc-Kunt and Levine, 2006).

Martinez-Miera and Repullo (2010) argue that the competition stability view does not necessarily hold when loan defaults are imperfectly correlated. Increased competition may have two effects: the risk-shifting effect, where increased competition reduces the borrower’s probability of default, and the margin effect, where increased competition may also reduce interest payment from performing loans, which serve as a buffer to cover loan losses. Their model predicts a U-shaped relationship between competition and risk: in highly competitive markets the margin effect dominates leading to more instability for the banks, and in highly concentrated markets the risk-shifting effect dominates and more competition reduces bank risk. Liu, Molyneux and Wilson, (2013) find evidence for this theory using a sample of banks located in 11 European countries for the period 2000 to 2008. This study uses regional variables as country level variables are supposed to be incorrect for banks that are more regionally orientated. The theory also holds for the Spanish banking market, where a non-linear relationship exists between market competition and the ratio of non-performing loans using standard measures of market concentration for both loan and deposit markets (Jiménez, Lopez, and Saurina, 2013). However, when concentration is substituted by the Lerner index calculated using bank specific interest rates, there is a positive relationship between risk and competition for the loan market, supporting the competition fragility hypothesis.

5 Schaeck and Cihak (2010) investigate a possible transmission channel for the effect of competition on financial stability. They reason that competition positively affects efficiency of banks, and efficient banks have incentives to screen and monitor borrowers better and will be characterized by a lower ratio of non-performing loans. First, they analyze the effect of competition measured using a Lerner index on various measures of profit efficiency. Second, they relate the Boone indicator, which is a measure of competition that focuses on the impact of competition on performance of efficient banks, to bank soundness. Their results indicate that, for a sample of European countries in the period 1995-2005, the degree of competition is positively related to bank soundness through the efficiency channel.

In contrast to the above mentioned studies, Xiaoqing Fu, Yongjia Lin and Molyneux, (2013) employ a market-based measure of bank risk instead of accountancy based measures of risk. They calculate the probability of bankruptcy using Merton’s distance to default model for banks located in 14 Asia Pacific economies from 2003 to 2010 to investigate the influence of bank competition, concentration, regulation and national institutions on individual bank fragility. They find that both lower pricing power as a more concentrated market is associated with financial fragility. Their results also indicate that banks are more stable in countries with tougher entry restrictions, and less stable in countries with strong deposit insurance schemes. Entry restrictions and deposit insurance in part explain the large cross-variation in the relation between competition and stability in the banking sector. Other variables that influence the tradeoff between competition and financial stability are lower systemic fragility, better developed stock exchanges and more effective credit sharing (Beck, De Jonghe and Schepens, 2013).

The above mentioned studies use different methodologies and datasets, leading to diverging results. The focus of this study is on the competition-stability trade-off for member countries of the European Union, and analyzes whether the relationship between competition and bank risk is stable over time and which factors affect this relationship.

3. Methodology

(1)

6 invariant bank heterogeneity, bank-fixed effects are added to the regression ( ) and time fixed effects ( ) to control for time varying global business cycle effects. I run a Hausman test for each regression to check whether fixed effects are preferred over random effects, and a redundancy test to check whether the fixed effects are jointly zero (Brooks, 2008). To address potential heteroskedasticity in the error term, I use robust standard errors clustered at the cross-section. The exogenous variables are lagged by one year to mitigate potential endogeneity problems between bank risk, economic conditions and other bank-specific features (Liu, Molyneux and Wilson, 2013). All financial variables and ratios are winsorized at the first and 99th percentile to mitigate the effect of outliers.

Endogeneity problems might be present due to a loop of causality between bank risk and the competition measure. This reflects in correlation between the error term and the competition measure. In the presence of an endogenous variable, using an ordinary least squares (OLS) regression would be inappropriate as it will produce biased and inconsistent estimates. An instrumental variable (IV) technique can give consistent parameter estimates by controlling for unobserved heterogeneity. Heterogeneity of an unknown form can best be addressed by a GMM estimator, as a GMM estimator does not require to make distributional assumptions on the error terms. A GMM estimator also accounts for heteroskedasticity, and is therefore preferred above the 2SLS technique (Hall, 2005). In order to get reliable results from the IV-technique, it is important to select instruments that have the following two properties: the instruments should be highly correlated with the competition measure, but not correlated with the errors (Brooks, 2008). Thus, the instrument should be related to the dependent variable, bank risk, only through the potentially endogenous variable, competition. If the correlation between the instruments and the endogenous variable is low, then a ‘weak instrument’ problem might occur. This may lead to two problems; first, the endogeneity bias will not be removed, and might even be worse, and secondly, the estimates of the coefficient standard errors will be biased, making hypothesis testing unreliable (Stock, Wright and Yogo, 2002). Hence, in this study I employ both an IV-technique with a GMM estimator and an OLS regression.

7 A high percent of government owned banks, or a low percent of foreign owned banks, points to a relatively closed economy, in which competition is limited. Table 1 contains a correlation matrix of the instruments, the competition and risk measures. The simple correlation between the Lerner index and its instruments is quite low, suggesting that a weak instrument problem might be present.

Table 1. Correlation matrix of the instruments for the competition measures and bank size. The variables are

defined as follows: Zscore is the natural logarithm of the Z-score, NPL/TL is the ratio of non-performing loans to total loans, Lerner is the Lerner index, Boone is the Boone indicator. The remaining variables are instruments, and defined as follows: Actrest is the activity restrictions index of a country, FFI is the Financial Freedom index, Foreign and Govern is the percent of foreign and government owned banks, MKT is the natural logarithm of the market share of a bank and FIXTA is the ratio of fixed assets to total assets of a bank.

Zscore NPL/TL Lerner Boone Actrest FFI Foreign Govern MKT FIXTA Lerner 0.176 0.005 1.000 Boone -0.105 -0.026 -0.034 1.000 Actrest -0.060 0.012 0.009 -0.169 1.000 FFI -0.114 -0.046 0.013 0.086 -0.371 1.000 Foreign -0.055 0.093 0.021 -0.421 -0.189 0.269 1.000 Govern 0.122 -0.043 -0.079 -0.152 -0.112 -0.432 -0.149 1.000 MKT -0.047 0.051 0.080 -0.237 -0.094 0.172 0.538 -0.138 1.000 FIXTA -0.047 0.007 -0.077 0.025 0.058 -0.081 -0.007 0.022 0.029 1.000

For the Boone indicator, the same endogeneity problem between competition and bank risk might be present. However, also endogeneity between bank size and the Boone indicator can be a problem. Banks might ‘gamble for resurrection’ when they are more fragile (Schaeck and Cihak, 2008). Banks will extend more risky loans, and this might be understood as an increase in competition, although the degree of competition is not altered. This study uses the same instruments for the Boone indicator as for the Lerner index: Activity Restrictions, the Financial Freedom Indicator, the percent of foreign owned banks and the percent of government owned banks. The following two variables also serve as instruments for the Boone indicator and bank size: the natural logarithm of market share, and the fixed assets to total assets of the banks (Schaeck and Cihak, 2008). The market share is likely to affect bank soundness indirectly through bank size, and the ratio of fixed assets to total assets indicates whether a bank has a wide ranging branche office network, which can serve as an indicator of market power and bank size. The Boone indicator is highly correlated with most of its instruments, and therefore the ‘weak’ instrument problem is assumed to be of less a problem when the Boone indicator is used in an IV technique with a GMM estimator.

8 potentially endogenous. Hence, the First stage F test cannot be applied in this case2. To test the exogeneity of the instruments, meaning that the instruments are uncorrelated with the error term, I use Hansen’s J test of over identification.

Model one provides us with an indication of the relation between the degree of competition and bank risk. However, there are variables related to the supervision of banks in a country, the regulation and the institutions, that might explain variance in bank risk. Cross-section fixed effects control for unobserved heterogeneity that does not vary over time, and which is correlated with independent variables (Brooks, 2008). Thus, unobserved heterogeneity that varies over time, and is correlated with included variables, may lead to an omitted variable bias. The consequence would be that estimated coefficients on other variables will be biased and inconsistent (Brooks, 2008). Inclusion of irrelevant variables may however lead to inefficient estimates. To reduce the omitted variable bias, I include supervisory, regulatory and institutional variables to the model, which I derive from economic theory to assure the relevance of these variables (see Section 4.3).

Hence, I estimate a second model (equation two), in which I add several institutional, regulatory and supervisory variables to the regression specification. This will also enable us to check which of these variables is related to bank risk.

(2)

Where respectively , and stand for a set of institutional, supervisory and regulatory variables. These are all at the country level, and consist of the following variables: the Credit information sharing index, the Financial structure, the Financial development, the Capital stringency index, the Deposit insurance index, the Supervisory power index, the Private monitor index, the Moral hazard index, the Diversification index, the Entry requirements index, and the Investor protection index.

4. Data

The initial sample consists of 5,222 savings, cooperative and commercial banks located in the 28 member countries of the European Union3 for the period 2001 to 2011. Not all variables are available

2 _{Eviews, the statistical program I use, does not report the results of the Cragg Donald test for panel data. The}

Cragg Donald test can assess the relevance of the instruments in case more than one endogenous regressor is present in the model.

3

9 for each bank year observation; hence the sample size varies across regression specifications. Each country has more than ten cross-sections, and over 90 bank year observations.

The country level variables are derived from the World Bank’s World Development Indicator database (WDI) and the bank specific variables from Bureau van Dijk’s Bankscope database. Table A1 of Appendix A contains an overview of the definitions and sources of the bank and country level variables, and table B1 of Appendix B the summary statistics for these variables.

The regulatory, institutional, and supervisory variables come from either the WDI database, the World Bank’s ‘Doing Business’ database, the World Bank’s Financial Development and Structure (FDFS) database, data gathered by the Heritage Foundation, or the Bank Regulation and Supervision (BRS) database (Barth, Caprio and Levine, 2013). The BRS database consists of surveys carried out in the years 2001, 2003, 2007 and 2012. For this study, the missing years are completed by using the previously available survey data until a new survey becomes available. For an overview of the definition and sources of the regulatory, institutional and supervisory variables, see table A3 of Appendix A. Summary statistics for these variables are in table B3, of Appendix B.

The rest of this section is organized as follows. The first subsection contains a description of the two risk measures I use in this research. The second subsection elaborates on the two competition measures. The final subsection discusses the relevant supervisory, institutional and regulatory variables.

4.1 Risk measures

As risk measures, this study uses the Z-score to measure overall bank risk, and the ratio of nonperforming loans to measure loan portfolio risk.

Z-score

The Z-score is an accounting based measure of financial fragility. It takes into account the profitability of the bank, the bank’s capital buffers and associated risks. It is calculated using the following equation:

10 using a standard deviation of return on assets of the full sample I include in the analysis as robustness test.

The Z-score increases when either the return on assets or the capitalization level increases, and decreases when the standard deviation of return on assets decreases. As the Z-index is highly skewed, I take the natural logarithm of the Z-score, which follows a normal distribution (Laeven and Levine, 2009). The average Z-score has a mean value of 1.47, with a maximum of 5.09. The correlation matrix in table 2, points out that the Z-score is positively correlated with the Lerner index (0.224), and negatively with the Boone indicator (-0.138). The Boone indicator is negative, and a more negative value of the Boone indicator points to more competition. Accordingly, a negative correlation between the Boone indicator and the Z-score denotes a positive relation between competition and the Z-score.

Ratio of nonperforming loans

In order to measure the risk of the loan portfolio of a bank, this study uses the ratio of nonperforming loans to total loans. The mean value of the ratio of nonperforming loans to gross loans is 3.49%, with a maximum value of 26.6%. Only 13.162 observations are available, which corresponds to approximately one quarter of the total observations in the sample. The correlation matrix in table 2 indicates that loan portfolio risk is not significantly correlated with either the Z-score, or one of the competition measures.

4.2 Competition measures

Choice of a competition measure is rather difficult as there seems to be limited consistency between the different competition measures (Carbo et al., 2009). In this study I use two measures of competition: the Lerner index and a more innovative measure of competition, the Boone indicator.

Lerner index

The Lerner index determines individual price setting power by the company’s ability to raise prices above its marginal cost. It has the following specification:

(4)

Where is the price of assets and is measured as the number of total revenues (interest and non-interest income) over total assets, and is the marginal cost of each bank which is derived from a Translog cost function that is estimated per country (see Appendix C). The Lerner index takes values between zero and one, whereby larger values indicate less competition and more market power. It is a non-metric scale and therefore interpretations can be drawn only from relative differences. The mean value of the Lerner index is 0.20 in the sample, with a maximum of 0.69.

Boone Indicator

11 2008). Schaeck and Cihak (2010) compute the Boone indicator for the banking market per country per year, and use the following empirical setup:

( ) (5)

Where are the profits of bank at time , denotes marginal cost, and is the Boone indicator. Schaeck and Cihak (2010) assume that the relation between marginal cost and profit is downward sloping. If marginal cost rise and prices stay at the same level, this would imply lower profit margins for the banks. However, if prices will rise, output will be reduced, and the market share of the bank will decline. The reduction of profits that arise due to an increase in marginal cost is expressed by the Boone indicator. The more negative the Boone indicator, the stronger competition, and thus the larger the fall in profits if marginal cost rise.

The Boone indicator can be obtained per year per country from the World Bank’s database, and is calculated following the same methodology as Schaeck and Cihak (2010). The average value of the Boone indicator is -0.034, and the maximum is 5.968. The standard deviation of the Boone indicator is 0.361, indicating that there is considerable dispersion around the mean.

The ECB (2005) provides the results of two measures of banking competition for each country in the European Union each year. These measures are the Herfindahl-Hirschman index (HHI) and the market share of the five largest banks (CR5), which this study uses as robustness check. The HHI measures market concentration by adding the squares of the market shares of all banks in a country. A higher value for the HHI indicates less dispersion in market shares and therefore an increase in market power. The HHI and CR5 are positively correlated with the Lerner index, respectively correlations of 0.062 and 0.094, indicating that an increase in market power is associated with an increase in the concentration measures. The Boone indicator is negatively correlated with the HHI (-0.235). An increase in the Boone indicator, and thus a decrease in competition, is associated with a decrease in the HHI. The Boone indicator is positively correlated with five bank asset concentration (0.057). As competition and concentration are not the same, the correlation between the competition and concentration variables may diverge. The low correlation between the Lerner index and Boone indicator (-0.057) is however striking, as both are supposed to measure the degree of competition.

Table 2. Correlation matrix of risk and competition measures. The variables are defined as follows: Z-score is

the natural logarithm of the Z-score, NPL/TL is the ratio of non-performing loans to total loans which measures loan portfolio risk, Boone is the Boone indicator, Lerner is the Lerner index, HHI is the Herfindahl-Hirschman index and CR5 is the market share of the five largest banks.

Zscore NPL/TL Boone Lerner HHI CR5

12

4.3 Regulatory, Supervisory and Institutional Variables

Regulatory, supervisory and institutional variables can influence the competition stability trade-off due to the effect they can have on the scope for adverse selection and moral hazard (Beck, De Jonghe and Schepens, 2013). These agency problems are caused by asymmetric information between the entrepreneurs and the banks, and can be reduced by increasing the degree of information sharing. Therefore, I include the variable ‘Credit information sharing’ in the model, which measures the degree of information that credit registry institutions have on borrowers. A better information sharing system will reduce the incentive of the entrepreneur to take on more risk and leads to a better screening of the borrowers before extending a loan.

A well-developed stock market also enhances information sharing due to increased information disclosure and transparency for listed firms. It further reduces moral hazard as a listed firm will suffer reputation losses in other markets too if it misbehaves. Hence, I add a ‘Financial development’ variable to the model, which is defined as the Stock market capitalization divided by GDP. Also, this study takes into account the availability of alternative sources of funding for firms by including a ‘Financial Structure’ variable which is measured as private credit divided by GDP. More alternative sources of funding might make it easier for banks to obtain funding, and therefore increase potential moral hazard.

13 The effectiveness of the supervision of banks depends on several aspects. First of all, a supervisor that has no power is not able to correct the behavior of misbehaving banks. For this reason, I add a ‘Supervisory Power’ index to the model which captures the authority and power of supervisors. Apart from the official supervisors, also private monitors might be active in the market. The ‘Private Monitor’ index is based on the incentives given for private monitoring in a country. The higher the index, the more incentives there are for private monitors to effectively monitor the bank and reduce potential agency problems. Incentives for investors to monitor the bank are reduced when a country has a high protection of its investors. The ‘Investor protection’ measures the strength of investor protection in a country. The final variable I add to the model is ‘Entry requirements’. Tighter entry requirements may increase bank profits due to a less competitive environment. It might however also induce market inefficiency. Hence, the impact of entry requirements on financial stability can go two directions.

Table 3. This table contains the expected relation between bank risk and one of the institutional, regulatory and

supervisory variables. The independent variable is expected to be either positively (+), or negatively (-) related to bank risk.

Variables Expected relation with bank

risk Credit information sharing -

Financial development - Financial structure + Capital stringency - Deposit insurance + or - Moral hazard - Diversification - Supervisory power - Private monitor - Investor protection + Entry requirements + or -

5. Empirical Results

In this section, I discuss and interpret the main results. I start with the discussion of the impact of competition on overall bank risk and loan portfolio risk. In the following two subsections I add regulatory, supervisory and institutional variables to the model to see whether these variables affect the relation between bank risk taking and competition.

5.1 Impact of competition on bank risk

14 these tests are not reported. In regression (3) and (4), I use an IV technique with a GMM estimator to control for potential endogeneity of the Lerner index in regression (3), and of the Boone indicator and bank size in regression (4).

The parameter of interest is the parameter of the competition measure. The positive or negative value, the size and significance of this parameter gives an indication of the potential relation between bank risk and competition.

In regression (1) of table 4, the parameter of the Lerner index enters the regression significantly positive (1.791). As a higher value for the Lerner index indicates more market power, there is a negative relation between competition and the Z-score. This is in line with the competition fragility view, in which more competition leads to a less stable financial system. However, the Lerner index might be endogenous, leading to biased and inconsistent estimates if OLS is used. Therefore, in regression (3), I use an IV technique with GMM estimator. I use the following instruments for the Lerner index: Activity restrictions, Financial Freedom, the percent of government owned banks and the percent of foreign owned banks. I drop the percent of government owned banks from the instrument list, to ensure the exogeneity of the instruments. The percent of government owned banks has the biggest impact on the Hansen’s J-statistic, suggesting that it might not be exogenous to the model. The p-value of the Hansen’s J-statistic should be above 5% to assure that the instruments are exogenous. The p-value of the J-statistic of regression (3) is 0.313, confirming the exogeneity of the instruments. Note that I use the same line of reasoning for the other regressions that make use of an IV technique. The First-stage F-test assesses the relevance of the instruments. If the instruments are relevant, they are correlated with the endogenous variable conditional on the other explanatory variables in the model. As a rule of thumb, the value of the First-stage F-test should be larger than 10 for the instruments to be relevant (Stock, Wright and Yogo, 2002). In regression (3), the value of the First-stage F-test is 74.769, suggesting that the instruments are relevant.

The relation between the Z-score and the Lerner index is positive in regression (3), but larger in magnitude than in regression (1) (6.337 compared to 1.791). The standard deviation of the parameter of the Lerner index is considerably larger in regression (3) than in regression (1) (1.542 versus 0.189). If the competition measure would be exogenous to the model, the parameter estimate should not be very different if either OLS or an IV technique is used. The large difference in magnitude of the parameter of the competition measure indicates that the competition measure is probably endogenous, and using OLS would lead to biased and inconsistent estimates. Although the instruments are valid and relevant, we need to be aware of the potential presence of a weak instrument problem. If the instruments are weak, the consequence can be increased bias in the estimated IV coefficients.

15 negative value of the Boone indicator. Thus, there is a positive relation between the degree of competition and the Z-score. In contrast to the results for the Lerner index, the findings support the competition-stability view, in which an increase in the degree of competition is related to a higher Z-score. However, the Boone indicator and bank size are potentially endogenous. Hence, in regression (4) of Table 4 I use an IV technique with GMM estimator. The instruments for the Boone indicator and bank size are: Activity restrictions, Financial freedom, the percent of government owned banks, the percent of foreign owned banks, the natural logarithm of the market share and the ratio of fixed assets to total assets of the bank. I however drop the percent of government owned banks from the instrument list, to improve the value of the J-statistic. In line with results for the Lerner index, I find a positive relation between the bank’s Z-score and market power (7.623). Note that the parameter of the competition measure is different in sign and magnitude if I use an IV technique instead of OLS (7.623 compared to -2.908). This large difference in the parameter estimates implies that bank size and the competition measure are probably endogenous, and using an IV technique is correct.

The results in table 4 also show that banks located in countries that are globally integrated, and banks that have a low ratio of loan loss provisions to total assets, experience higher Z-scores. All four regressions support these findings. The ratio of loan loss provisions to total assets measures asset quality, and therefore a deterioration of a bank’s asset quality is related to more overall bank risk. A possible explanation is that the bank’s assets generate the income for the bank, and lower quality thus implies increased risk for the income generating activities, which is reflected in a lower Z-score. Striking are the results for the crisis dummy and country size. In regression (1) and (2), the crisis dummy enters the regression significantly positive, suggesting that banks located in countries that suffer from a banking crisis experience higher Z-scores. Due to the financial turmoil caused by a banking crisis, we would however expect banks to have lower Z-scores. In case I use an IV technique, the sign of the crisis dummy is reversed making the results more plausible. Country size is significantly negative related to the Z-score if the estimation method is OLS, and significantly positive in case I use an IV technique.

The economic development of a country seems to play a positive role for financial stability in regression (1) and (2), in which the parameter of economic development is significantly positive. This result does not hold in regression (3) and (4). In regression (3) and (4) on the other hand, the deposits to total assets which measures the reliance on deposits for funding of banks, is positively related to the Z-score. The leverage ratio of the bank does not explain variance in the Z-score in regression (1) and (2), and has a contradicting sign in regression (3) and (4).

16 rates from its borrowers, and thereby increase its return. There is no relation between the capitalization level (Equity/TA) of the bank and the Lerner index. The standard deviation of the return on assets based on a three year rolling window, (St. dev. ROAA), is negatively related to the degree of market power, indicating that an increase in market power is associated with a decrease in the standard deviation of the return on assets. These results support the findings that increased competition leads to financial fragility. In regression (4), the Z-score is calculated using the standard deviation of the return on average assets based on the full sample. The overall inferences stay the same. In regression (5) and (6), I use the HHI and five bank asset concentration ratio as competition measure. There is no significant relation between the HHI and bank risk, and a negative relation between the five bank asset concentration and the Z-score. This is not in line with my other findings, and it should be noted that concentration and competition are not two sides of the same coin. Hence, it should not be surprising that they may lead to different outcomes.

In Table D2 of Appendix D, the competition measure is the Boone indicator and the dependent variable is one of the components of the Z-score or the Z-score computed using a standard deviation based on the full sample. There is a negative relation between the return on assets and market power which is significant at the 10% level. No significant relation between the standard deviation of the return on assets and the competition measure seems to exist, and a negative relation between the competition measure and the capitalization rate at the 1% level. Notable is that both the Lerner index and Boone indicator are related to the return on assets, but with different sign. The negative relation between the Boone indicator and the capitalization rate implies that banks with more market power have lower capitalization rates. Banks with more market power may be assumed to more easily stay profitable, and therefore can increase their leverage ratio without raising concerns among investors. To control for potential multicollinearity problems, I run another robustness test. Multicollinearity can cause two problems when ignored; first, the R-squared of the model can be very high, but the individual parameters may not be significant due to high standard errors. Second, the regression becomes sensitive to small changes in the model (Brooks, 2008). However, due to the large sample size, I assume the influence of multicollinearity on the parameter estimates and their standard deviations to be minimal. Country size and Global integration are highly correlated (-0.88), and therefore I run the same regressions excluding either country size or global integration. Overall inferences stay the same, and results are not reported.

18

* Table 3. This table contains results of regressions that have as dependent variable the Z-score and as competition measure either the Lerner index or Boone indicator. Regression (1) and (2) have as estimation method OLS with bank and time fixed effects and standard errors clustered at the cross-section. In regressions (3) and (4), I use an IV technique with a GMM estimator. The GMM-estimations have a White period covariance matrix to control for heteroskedasticity at the cross-section, and time fixed effects. In regression (3) I use the following instruments for the Lerner index: Financial freedom, Activity restrictions and the percent of foreign owned banks. The instruments for the Boone indicator and bank size in regression (4) are: Financial freedom, Activity restrictions, the percent of foreign owned banks, the fixed assets to total assets of the bank, and the natural logarithm of the market share of the bank. The Hansen's J-statistic tests whether the instruments are valid, and follows a chi-square distribution with the number of instruments less the parameters to be estimated as the degrees of freedom. The First stage F-statistic indicates whether the instruments are relevant. In parenthesis are robust standard errors clustered per cross-section. * indicates significance at the 10% level, ** at the 5% level and *** at the 1% level.

Regression 1 2 3 4

Dependent variable Z-score

Estimation method OLS OLS GMM GMM

Lerner 1.791 6.337 (0.189)*** (1.542)*** Boone -2.908 7.623 (1.208)** (2.252)*** LN (Total Assets) -0.065 -0.058 0.023 0.035 (0.048) (0.046) (0.010)** (0.035) Leverage 0.002 0.000 0.013 -0.007 (0.003) (0.003) (0.005)** (0.002)*** LLP / TA -27.382 -19.107 -57.363 -20.563 (3.113)*** (2.881)*** (9.394)*** (2.412)***

Net income /TA 0.100 0.075 -0.05 -0.027

(0.119) (0.112) -0.069 (0.072) Deposits /TA -0.025 -0.005 0.502 0.735 (0.319) (0.286) (0.145)*** (0.115)*** Country size -10.751 -10.075 0.511 0.427 (1.465)*** (1.428)*** (0.031)*** (0.022)*** Economic Development 22.462 2.353 0.009 -0.053 (0.623)*** (0.652)*** -0.035 (0.048) Economic stability -0.004 -0.001 0.005 -0.002 (0.003)** -0.003 (0.003)* (0.003) Crisis 0.24 0.244 -0.194 -0.201 (0.114)** (0.109)** (0.097)** (0.102)** Global Integration 0.018 0.016 0.009 0.009 (0.002)*** (0.002)*** (0.001)*** (0.001)*** Observations 20,660 21,637 19,119 17,603 R-squared 0.493 0.492 F-statistic 4.755 4.652 J-statistic 2.325 7.560 P-value J-statistic 0.313 0.056

First stage F-statistic 74.769 n.a.

19

5.2 Impact of competition on loan portfolio risk

Table 5 shows the results from the estimation of equation (1), with as dependent variable the risk of the loan portfolio, which is defined as the ratio of non-performing loans to total loans (NPL/TL). In regression (1) and (2) the estimation method is OLS with bank and time fixed effects, and I control for unobserved heteroskedasticity at the cross-section by using White’s period standard errors. The results for the Hausman test, which I do not report, indicate that fixed effects are preferred over random effects, and the redundancy test confirms appropriateness of using cross-section and time fixed effects in the OLS estimations. In regression (3) and (4) I use an IV technique with a GMM estimator to control for potential endogeneity of the Lerner index in regression (3), and of the Boone indicator and bank size in regression (4). Again I drop the percent of government owned banks from the instrument list to assure that the instruments are exogenous to the model based on the Hansen’s J-statistic4

. In case the estimation method is OLS, there is no significant relationship between loan portfolio risk and the degree of competition (see results for regression (1) and (2)). The results for regression (3) and (4), in which I use an IV technique, point to significant negative relation between the competition measure and loan portfolio risk. The coefficient of the Lerner index is significant and has a value of -19.053, indicating that banks situated in more competitive environments have more risky loan portfolios. I find the same relation between bank risk and the Boone indicator (-15.511). The F-statistic for regression (3) points out that the instruments are relevant. The GMM estimates deviate substantially from the OLS estimates, and the standard deviations are considerably larger. This large difference in the parameter estimates implies that bank size and the competition measure are probably endogenous, and using an IV technique is correct. However, if instruments are weak, an IV technique might worsen the bias in the estimated coefficients.

Among the control variables, banks situated in economically more developed and stable countries tend to have lower loan portfolio risk. These results are comparable to the results for overall bank risk. Furthermore, banks that have higher profitability ratios and a higher quality of assets (LLP/TA), experience lower loan portfolio risk. A possible explanation for the negative relation between the profitability ratio and loan portfolio risk is that banks with a lower profitability ratio increase the risk of their loan portfolio in an attempt to increase profitability. And the assets of a bank consist for a large part of the loan portfolio, and thus a decrease in asset quality is also a decrease in quality of the loan portfolio, which subsequently may lead to a riskier loan portfolio.

4 _{Note that the instruments are not valid in regression (3) as the p-value of the J-statistic is 0.026, which is below}

20 For robustness, the degree of competition is measured by the HHI and the five bank asset concentration (CR5) in regression (5) and (6), which both have no explanatory power for risk of the loan portfolio in line with results of regression (1) and (2).

21

Table 5. This table contains regressions run to explain variation in the risk of the loan portfolio (NPL/TL) using several

independent variables, including the degree of competition. If the estimation method is OLS, bank and time fixed effects are included with standard errors clustered at the cross-section. The GMM-estimations have a White period covariance matrix to control for heteroskedasticity at the cross-section, and time fixed effects. In regression (3) I use the following instruments for the Lerner index: Financial freedom, Activity restrictions and the percent of foreign owned banks. The instruments for the Boone indicator and bank size in regression (4) are: Financial freedom, Activity restrictions, the percent of foreign owned banks, the fixed assets to total assets of the bank, and the natural logarithm of the market share of the bank. The Hansen's J-statistic tests whether the instruments are valid, and follows a chi-square distribution with the number of instruments less the parameters to be estimated as the degrees of freedom. The First stage F-statistic indicates whether the instruments are relevant. The standard errors are in parenthesis. * indicates significance at the 10% level, ** at the 5% level and *** at the 1% level.

Regression 1 2 3 4 5 6

Estimation OLS GMM OLS

Dependent variable NPL/TL

Competition Measure Lerner Boone Lerner Boone HHI CR5 Lerner 0.182 3.58 -19.053 -15.511 -2.110 -0.013 (0.659) (4.755) (5.934)*** (6.757)** (6.586) (0.014) Bank size 0.129 0.285 0.01 0.085 0.023 0.013 (0.289) (0.200) (0.052) (0.240) (0.281) (0.281) Leverage 0.000 0.005 -0.055 0.004 -0.003 -0.003 (0.015) (0.01) (0.019)*** (0.008) (0.011) (0.01) LLP / TA 75.279 82.919 142.445 89.299 77.087 72.503 (16.085)*** (16.603)*** (24.507)*** (17.741)*** (15.465)*** (14.988)*** Net income /TA -0.602 -0.662 -0.601 -0.2 -0.583 -0.624 (0.240)** (0.236)*** (0.257)** (0.215) (0.227)** (0.222)*** Deposits /TA 0.012 0.075 0.893 0.234 -0.249 -0.137 (1.059) (0.895) (0.644) (0.428) (0.941) (0.987) Country size -15.211 -13.61 -0.485 -0.666 -25.334 -13.164 (9.895) (10.404) (0.144)*** (0.129)*** (9.321)*** (10.427) Economic Development -7.778 -4.817 -0.042 -0.145 -6.211 -8.662 (2.786)*** (2.620)* (0.172) (0.122) (2.726)** (2.794)*** Economic stability 0.058 0.047 0.023 0.042 0.055 0.059 (0.012)*** (0.012)*** (0.011)** (0.012)*** (0.012)*** (0.012)*** Crisis 0.035 -0.648 -1.392 -0.67 0.072 -0.095 (0.397) (0.411) (0.489)*** (0.377)* (0.367) (0.389) Global integration 0.002 0.003 -0.008 -0.018 -0.003 -0.001 (0.007) (0.008) (0.004)** (0.004)*** (0.007) (0.007) Observations 6,836 7,239 6,057 5,496 6,342 6,504 R-squared 0.799 0.803 F-statistic 12.322 12.552 J-statistic 7.338 1.098 P-value J-statistic 0.026 0.578

First-Stage F-test 74.769 n.a.

22

5.3 Impact of competition on bank risk, including regulatory, supervisory and

institutional variables

Table 6 shows the results from the estimation of equation (2), with as dependent variable the Z-score of the bank. Equation (2) includes regulatory, supervisory, and market structure variables, allowing us to assess whether these variables influence the relation between competition and stability, and which of these variables themselves have explanatory power for the Z-score. The estimation method for regression (1) to (4) is OLS with bank and time fixed effects, and standard errors clustered at the cross-section. The results for the Hausman test indicate that fixed effects are preferred over random effects, and the redundancy test confirms appropriateness of both cross-section and time fixed effects. Results for these tests are not reported. In regression (5) to (8), I use an IV technique with GMM estimator to control for potential endogeneity of the Lerner index, and the potential endogeneity of bank size and the Boone indicator. Two variables, Credit information sharing and the Investor protection index, only have data available for the years 2007 to 2011, and therefore, I estimate the same model excluding these two variables (regression (2), (4), (6) and (8)). The coefficients of the control variables are excluded from Table 6 for clarity, and are reported in table E1 of Appendix E. The results for the control variables point out that banks with a lower quality of assets (LLP/TA) and banks situated in economically more developed, and globally more integrated countries, tend to have higher Z-scores.

The results of regression (1) in Table 6 point to a positive relation between the Lerner index and the Z-score: a bank with more market power is more stable. Excluding credit information sharing and investor protection from the regression specification does not change this relation. Note that these findings confirm our previous findings in Section 5.1, that a negative relation between competition and financial stability exists. There is no relation between the Boone indicator and financial stability in regression (3) and (4). Without the supervisory, regulatory and institutional variables, the Boone indicator is assigned explanatory power for variance in bank risk (see Section 5.1). After the addition of these variables, the Boone indicator has no explanatory power anymore. Hence, excluding these variables probably leads to an omitted variable bias, and may lead to wrong inferences.

24

* Table 6. This table reports results of regressions with as dependent variable the Z-score and independent variable the competition measure, bank and country control variables, and regulatory, supervisory and institutional variables. If the estimation method is OLS, I include bank and time fixed effects with standard errors clustered at the cross-section. The GMM-estimations have a White period covariance matrix to control for unobserved heterogeneity at the cross-section, and time fixed effects. In regression (5) and (6), I use the following instruments for the Lerner index: Financial freedom, Activity restrictions and the percent of foreign owned banks. The instruments for the Boone indicator and bank size in regression (7) and (8) are: Financial freedom, Activity restrictions, the percent of foreign owned banks, the fixed assets to total assets of the bank, and the market share of the bank. Hansen's J-statistic tests whether the instruments are valid, and follows a chi-square distribution with the number of instruments less the parameters to be estimated as the degrees of freedom. The First stage F-statistic indicates whether the instruments are relevant. The results for the bank and country level control variables are reported in Appendix E (Table E1). In regression (2), (4), (6) and (8) Credit Information Sharing and the Investor Protection index are dropped from the model, as they only have information available for the years 2007-2011, and thereby limit the sample significantly. Standard errors are in parenthesis. * indicates significance at the 10% level, ** at the 5% level and *** at the 1% level.

Regression 1 2 3 4 5 6 7 8

Dependent variable Z-score

Estimation method OLS GMM

Lerner 1.544 1.769 -3.876 3.566 (0.302)*** (0.253)*** (1.757)** (2.207) Boone 1.486 -0.049 -5.853 -3.507 (2.255) (1.485) (3.621) (4.383) Credit Information Sharing 0.156 0.117 0.006 -0.085 (0.097) (0.093) (0.047) (0.059) Financial structure -0.015 -0.013 -0.015 -0.013 -0.005 0.000 0.002 0.001 (0.005)*** (0.003)*** (0.005)*** (0.003)*** (0.002)** (0.001) (0.002) (0.002) Financial Development 0.002 -0.002 0.001 -0.003 0.003* -0.001 0.000 0.000 (0.003) (0.002) (0.003) (0.002) (0.002) (0.002) (0.001) (0.001) Capital Stringency 0.096 0.024 0.108 0.026 0.168 0.038 0.092 0.093 (0.026)*** (0.021) (0.025)*** (0.020) (0.044)*** (0.025) (0.035)*** (0.033)*** Deposit Insurance -0.229 -0.015 -0.213 -0.012 -0.072 0.045 0.018 0.059 (0.082)*** (0.049) (0.081)*** (0.048) (0.052) (0.028) (0.049) (0.039) Supervisory power 0.100 0.005 0.083 -0.016 -0.008 0.010 -0.013 -0.047 (0.040)** (0.028) (0.037)** (0.027) (0.027) (0.021) (0.025) (0.023)** Private Monitor index -0.135 -0.089 -0.136 -0.065 -0.011 0.006 -0.064 -0.081 (0.101) (0.093) (0.097) (0.089) (0.055) (0.036) (0.061) (0.045)* Moral Hazard index -0.138 0.093 -0.145 0.146 -0.100 0.056 -0.137 -0.100 (0.154) (0.082) (0.146) (0.082)* (0.102) (0.056) (0.106) (0.111) Diversification index -0.057 -0.175 -0.042 -0.152 -0.079 0.018 -0.045 -0.101 (0.126) (0.111) (0.123) (0.110) (0.101) (0.086) (0.113) (0.138) Entry Requirements -0.439 0.318 -0.388 0.413 -0.017 0.111 0.758 0.375 (0.225)* (0.215) (0.221) (0.205)** (0.155) (0.100) (0.302)** (0.162)** Investor Protection index -0.359 -0.413 -0.139 -0.267

(0.167)** (0.161)*** (0.058)** (0.086)*** Observations 5567 8030 6010 8706 4733 6532 6920 R-squared 0.613 0.547 0.612 0.541 F-statistic 3.571 4.062 3.517 3.917 J-statistic 1.669 8.342 2.239 4.247 P-value 0.644 0.015 0.326 0.120

First-stage F-test 11.514 10.82 n.a. n.a

25

5.4 Impact of competition on risk of the loan portfolio, including regulatory,

supervisory and institutional variables

Table 7 shows the results from the estimation of equation (2), with as dependent variable the ratio of nonperforming loans to total loans of the bank. Equation (2) includes regulatory, supervisory, and market structure variables, and allows us to assess whether inclusion of these variables has influence on the relation between competition and loan portfolio risk. Besides, we can determine whether these variables themselves have explanatory power for risk of the loan portfolio. The estimation method for regression (1) to (4) is OLS with bank and time fixed effects, and standard errors clustered at the cross-section. The results for the Hausman test indicate that fixed effects are preferred over random effects, and the redundancy test confirms appropriateness of both cross-section and time fixed effects. Results for these tests are not reported. In regression (5) to (8), I use an IV technique with GMM estimator to control for potential endogeneity of the Lerner index, and the potential endogeneity of bank size and the Boone indicator. Two variables, credit information sharing and the investor protection index, only have data available for the years 2007 to 2011, and therefore, regressions (2), (4), (6) and (8), exclude these variables to increase the sample period. The coefficients of the control variables have been excluded from Table 7 for clarity. They are included in Table E2 of Appendix E. The degree of competition has no explanatory power for the dependent variable, risk of the loan portfolio, in case the estimation method is OLS. This holds for both the Lerner index and the Boone indicator. Hence, the addition of regulatory, supervisory, and institutional variables does not influence this relation. However, if instead of OLS, I use an IV technique with GMM estimator, a significant negative relation between the Lerner index and loan portfolio risk exists. Similar to the findings for equation (1), which excludes supervisory, regulatory, and institutional variables. Banks that have more market power tend to hold relatively less risky loan portfolios, in line with the competition-fragility view. This finding holds if Investor protection and Credit information sharing are dropped from the model. Consistent with the OLS estimation, there is no relation between the Boone indicator and loan portfolio risk in regression (7) and (8). Thus, adding regulatory, supervisory and institutional variables to the model changes the significance of the Boone estimator.

26 therefore banks will extend fewer loans to marginal loans applicants, reducing the risk of their loan portfolio.

Among the bank level control variables, banks that have a high asset quality (LLP/TA) experience less loan portfolio risk (see Table E2 of Appendix E). A possible explanation is that loans of bad quality have a high probability to become a nonperforming loan, and thus a decrease in the quality of the loan is related to an increase in loan portfolio risk. Two other bank level control variables that seem to play a limited role in explaining loan portfolio risk are the profitability of a bank and its leverage ratio. A bank that has a relatively low profitability ratio might extend more loans to marginal loan applicants in order to increase its profitability ratio, and thus increase the risk of its loan portfolio. A bank in the sample with a high risk profile, which is reflected in a high leverage ratio, has lower loan portfolio risk. An explanation can be that banks that increase their leverage ratio have to reduce loan portfolio risk to keep in control of the higher risk profile.

The results of Table E2 indicates that banks located in larger countries that are economically more stable, experience lower risk in their loan portfolio. Especially country size is assigned a lot of explanatory power in the OLS regressions. The coefficient in regression (1) is for example -90.579. However, in the GMM estimations the coefficient of country size, although still significant in regression (5) and (6), is considerably smaller, respectively -1.057 and -0.909.

As robustness, I use the HHI and the five bank asset concentration to infer the degree of competition. These variables are not assigned any explanatory power for variance in loan portfolio risk and I do not report the results.

27

Table 7. This table contains the results of regressions run with as dependent variable loan portfolio risk (NPL/TL) and

independent variable the competition measure, bank and country control variables, and regulatory, supervisory and institutional variables. The OLS estimations have bank and time fixed effects with standard errors clustered at the cross-section. The GMM-estimations have a White period covariance matrix to control for unobserved heterogeneity at the cross-section, and time fixed effects. In regression (5) and (6), I use the following instruments for the Lerner index: Financial freedom, Activity restrictions, and the percent of foreign owned banks. The instruments for the Boone indicator and bank size in regression (7) and (8) are: Financial freedom, Activity restrictions, the percent of foreign owned banks, the bank’s fixed assets to total assets, and its market share. The Hansen's J-statistic tests whether the instruments are exogenous and the First stage F-statistic indicates whether the instruments are relevant. The results for bank and country level control variables are reported in Appendix E (Table E2). In regression (2), (4), (6) and (8) Credit Information Sharing and the Investor Protection index are dropped from the model, as they only have information available for the years 2007-2011, and thereby limit the sample period significantly. Standard errors are in parenthesis. * indicates significance at the 10% level, ** at the 5% level and *** at the 1% level.

Regression 1 2 3 4 5 6 7 8

Estimation method OLS GMM

Dependent variable NPL/TL

Competition measure Lerner Lerner Boone Boone Lerner Lerner Boone Boone -0.010 0.111 9.757 1.911 -12.455 -13.426 1.723 -13.470 (0.658) (0.622) (6.651) (4.381) (3.938)*** (3.870)*** (12.136) (17.399) Credit Information sharing -0.327 -0.322 0.273 0.147 (0.219) (0.225) (0.170) (0.210) Financial structure 0.020 0.016 0.020 0.013 -0.012 0.000 -0.005 0.012 (0.017) (0.011) (0.016) (0.011) (0.008) (0.005) (0.008) (0.010) Financial Development 0.012 0.004 0.009 0.000 0.000 -0.008 -0.006 -0.015 (0.008) (0.006) (0.008) (0.006) (0.005) (0.006) (0.004) (0.005)*** Capital Stringency -0.020 0.126 -0.021 0.084 0.364 0.344 0.112 0.233 (0.050) (0.060)** (0.050) (0.059) (0.131)*** (0.106)*** (0.079) (0.100)** Deposit Insurance 0.277 0.243 0.293 0.328 0.168 0.165 0.234 0.363 (0.150)* (0.142)* (0.143)** (0.134)** (0.140) (0.122) (0.155) (0.138)*** Supervisory power -0.007 -0.059 0.014 -0.041 -0.125 -0.106 0.070 0.052 (0.112) (0.072) (0.108) (0.071) (0.090) (0.082) (0.088) (0.079) Private Monitor index 0.039 0.041 0.030 0.038 -0.022 -0.289 0.201 -0.130 (0.166) (0.168) (0.169) (0.170) (0.227) (0.179) (0.242) (0.227) Moral Hazard index -0.101 0.354 -0.105 0.444 0.328 0.539 0.724 0.417 (0.371) (0.298) (0.357) (0.291) (0.327) (0.306)* (0.370)* (0.539) Diversification index -0.804 -0.606 -0.777 -0.555 0.065 0.064 0.969 0.422 (0.337)** (0.437) (0.322)** (0.414) (0.318) (0.350) (0.367)*** (0.499) Entry Requirements -0.593 -1.125 -0.634 -0.972 -0.506 -0.427 -0.522 -0.339 (0.520) (0.432)*** (0.503) (0.416)*** (0.512) (0.392) (0.943) (0.568) Investor Protection Index 0.232 0.155 0.012 -0.094 (0.288) (0.270) (0.214) (0.234) Observations 4,438 5,787 4,712 6,146 3,863 4,765 3,772 4,505 R-squared 0.876 0.83 0.885 0.83 F-statistic 17.218 15.786 18.687 15.786 J-statistic 6.673 1.673 8.28 6.644 P-value J-statistic 0.083 0.643 0.041 0.084

First stage F-test 11.514 10.82

28

6. Conclusion

Under the competition fragility view, an increase in the degree of competition in the banking sector leads to a decrease in financial stability. The competition stability view assumes that an increase in the degree of competition positively affect financial stability. Both theories find support in the literature. This study examines the relation between the degree of competition and financial stability, using a sample of 5,222 banks located in the 28 member countries of the European Union for the period 2001 to 2011. I employ two measures of bank risk: the bank’s Z-score to measure overall bank risk, and the ratio of non-performing loans to total loans to measure loan portfolio risk. The degree of competition might affect these two risk measures differently (Berger et al., 2009). The Lerner index and the Boone indicator serve as proxies for the degree of competition. Besides the degree of competition, other regulatory, supervisory and institutional variables are taken into account to assess whether they influence the relation between competition and stability. I use both an OLS specification for the model, as an IV-technique with GMM estimator to control for potential endogeneity of the competition measure. The results indicate that the competition measure is probably endogenous, and that the competition-stability relation is not robust to the addition of regulatory, supervisory and institutional variables to the model.

The main results indicate that there is no relation between the degree of competition and the bank’s Z-score for the period 2001 to 2011. Among the regulatory, supervisory and institutional factors, I find that higher capital stringency, less investor protection and fewer alternative sources of funding increase the financial soundness of banks in a country.

For loan portfolio risk, I find a negative relation with the Lerner index and no relation with the Boone indicator. As the Lerner index is at the bank level and the Boone indicator is at the country level, it is possible to obtain diverging results. The results for the Lerner index are thus in line with the competition fragility view of Keeley (1990). Keeley (1990) argues that an increase in competition erodes market power, decreases profitability, and reduces the franchise value of the bank. The reduction in franchise value encourages the bank to take on more risk to increase returns, leading to an increase in loan portfolio risk. These results are in line with the findings from Jiménez, Lopez and Saurina (2013), who report a positive relation between loan portfolio risk and the degree of competition for the Spanish loan market. Although banks in a more competitive environment have higher loan portfolio risk than banks in a less competitive environment, overall risk of the bank does not have to increase. Banks may employ risk management methods in order to coop with the increased risk of the loan portfolio. The results also show that a stronger deposit insurance scheme and stricter entry requirements are positively related to loan portfolio risk.

29 too concerned about this phenomenon. The degree of competition seems to bear no relationship with overall bank risk, and therefore regulators should not be overcautious when evaluating mergers and acquisitions. I would recommend more strict capital requirements to reduce overall bank risk. Reducing investor protection and limiting the sources of funding for banks do not seem to be attractive alternatives, as reducing them might inhibit economic growth.

There are several limitations to this study. Although the Lerner index and the Boone indicator both measure the degree of competition, this study finds no concise results for these two measures. It can be argued that a spurious correlation exists between the Lerner index and Z-score, as both include the profitability ratio in their numerator, and any positive relationship between the two might not have any economical meaning (Beck, De Jonghe & Schepens, 2013). Another complication is that the instruments I use in this study might be ‘weak’. Weak instruments do not correct endogeneity biases and only aggravate problems. Although the instruments are valid in most regressions, the simple correlation between the Lerner index and its instruments is quite low, suggesting a weak instrument problem might be present. Weak instruments do not correct endogeneity biases and only aggravate problems. This problem seems to be less severe if the endogenous variable is the Boone indicator, as the simple correlations between the instruments and the Boone indicator are higher. The Boone indicator has its own limitations, as it is calculated on the country year basis, and the degree of competition might not be marked by country borders (Berger et al., 2009).