Does sectoral concentration lead to bank risk?

Transcription

1 TILBURG UNIVERSITY Does sectoral concentration lead to bank risk? Master Thesis Finance Name: ANR: T.J.V. (Tim) van Rijn s Date: Department: Supervisor: Finance dr. O.G. de Jonghe Session chair: dr. O.G. Spalt Faculty: Tilburg School of Economics and Management

2 Does sectoral concentration lead to bank risk? Abstract Does sectoral concentration lead to bank risk? This question has been studied on a single-country level in Italy by Acharya et al. (2006), in Germany by Hayden et al. (2007) and Brazil by Tabak et al. (2011). These studies come up with different relations between bank risk and level of diversification of the lending portfolio. To this end, this study uses manually collected data about the individual corporate lending exposures from the annual reports of the 325 largest listed banks worldwide over the period Additional to the measure of diversification derived from the reported diversification, a measure of diversification is derived from the stock price-returns. Based on the accounting data, there is statistical significance that in developed countries banks with concentrated loan portfolios are less risky. In the emerging economies, there is no significant relation. For the entire sample, concentration is negatively correlated to market risk. Based on the stock price-return data, sectoral concentration is positively related with bank risk and market risk for the entire sample. The effect is somewhat stronger in developed countries than in emerging countries. Both methods give opposite results.

3 Table of Contents 1. Introduction Literature review Data Methodology Measuring Banks' risk Measuring Banks' lending portfolio concentration Control variables Results Summary statistics Results of the accounting-based concentration measures A. Global Analysis B. Analyses on banks in developed and emerging countries Results of the stock price return-based concentration measure A. Global analysis B. Analyses on banks in developed and emerging countries Robustness and limitations Robustness Limitations Conclusions and recommendations References Appendix 1: Elaborated description and example of hand-collected sectoral exposure data Appendix 2: List of countries Appendix 3: Allocation of the sectors... 40

4 1. Introduction The financial and banking crises have increased the global demand to bank stability. Nearly all the people around the world have relations with at least one bank. Commonly expected from banks is that money is safe if it is in the hands of a bank. Therefore, it is important that banks manage their risks well. Although banks and other financial institutions have shifted their activities to nontraditional banking activities over the last decades, corporate lending is still a core activity of banks. Banks could either diversify their corporate lending portfolios or focus on some industries. The question whether diversification or focus is better has been intensively discussed in many fields in finance literature. However, as Acharya et al. (2006) stated, this question is not covered for banks' corporate lending portfolio frequently. Nevertheless, a large proportion of bank failures are caused by concentration of banks' lending portfolio (Basel Committee on Banking Supervision, 1991). A main reason for the lack of studies in this field is that information on the corporate lending portfolios of banks around the world is not easily accessible. Therefore, for this study data is manually collected from the largest listed banks worldwide. This study examines if sectoral concentration leads to bank risk. Furthermore, the data collected allows to test for cross-country variation. This study focuses here on the ten largest economies in the world and investigates if the effect of concentration on bank risk differs for developed and emerging countries. Characteristics and benefits of diversification and focus strategies have been discussed in many finance fields. However, not many studies investigate the diversification of banks' corporate lending portfolio. Acharya, Hassan and Saunders (2006) studied the effect of focus and diversification on the risk/return tradeoff in Italy over the period This study analyzes individual bank loan exposures to different sectors and shows that diversification is only inefficient for risky banks. In contrast to these findings, Hayden et al. (2007) showed that diversification is only beneficial for highrisk banks. This is a study on individual loan portfolios of German banks for the period from 1996 to Furthermore, Tabak et al. (2011) found that in Brazil, one of the largest emerging markets in the world, loan concentration improves bank performance. So three studies with data from three different countries show different relations between the diversification of the lending portfolio and bank performance. The current study differs from the ones discussed above since it contains not only data on a single country level, but data from the largest banks worldwide. The data is manually collected and since there are no global guidelines for banks how to report their sectoral exposure, the data might be less accurate than the data used on a single-country level. However, the data allows us to do a global 1

5 study and to investigate if there are cross-country differences. To find cross-country differences, this study focuses on the difference between developed and emerging markets. Imbs and Wacziarg (2003) found a u-shaped relation between sectoral concentration and income of a country, measured by the income per capita. The income per capita in emerging countries is lower than the income per capita in developed countries. Therefore, it is interesting if the effect of concentration on bank risk differs for banks in developed countries and in emerging markets. For this research, data from the 430 largest listed banks worldwide are used. From the annual reports, the sectoral corporate lending portfolios are manually collected over the period The sample consists of listed banks which makes it able to construct a risk measure from the stock price-returns. Furthermore, the annual reports of listed banks are publicly available and therefore data on the sectoral exposures are more facile to collect. Since the banks in the dataset are the largest banks globally, it can be assumed that the degree of diversification is at least partially the choice of the bank. Additionally, stock price-returns are used to calculate another measure of banks' concentration. This study is interesting for banks and bank regulators as the Basel Committee on Banking Supervision. Banks play a huge role in modern society and banking stability is important to have financial stability. Globalization allows large banks to expand into new markets and thus to diversify. On the other hand, banks could have the policy to concentrate their lending to a few markets. This study investigates if this leads to bank risk and if there are differences for banks from emerging and developed countries. In the next chapter, theories about diversification and focus and how these theories could affect bank risk are discussed. Section 3 describes the data used in this study and how the data is collected. Thereafter, in section 4, the empirical results are presented and discussed. First, the results on the accounting-based concentration measures are presented, followed by the results based on a stock price-return based concentration measure. Finally, this research ends with the conclusions and recommendations for further research. 2. Literature review Banks can diversify in many ways. The first way is geographically, by expanding into new areas. The second way is functionally, by adding non-traditional banking activities. This study focuses on the diversification and focus of banks' corporate lending portfolios. 2

6 Traditional portfolio theory states that well-diversification optimizes the risk-return tradeoff. Following this theory, risks will be minimized by optimizing correlations among different sectors. Default correlations are higher in specialized portfolios. By adding more low correlated investments, the correlations on default will be lower. Traditional portfolio theory argues that an efficient diversified portfolio eliminates the risk from idiosyncratic shocks. According to this theory, Diamond (1984) argues that diversifying a lending portfolio minimizes monitoring costs and that it resolves incentive problems between borrowers and lenders. Furthermore, Düllmann and Masschelein (2006) found that economic capital is larger for banks with concentrated portfolios than for diversified portfolios for continental European banks. On the other hand, diversification can be costly. If all banks are well diversified, all banks will have similar exposures. If one bank fails, it is likely that the other will fail simultaneously and systemic crises are likely to occur. Hence, a lot of banks with a high degree of diversification is undesirable (Wagner, 2010). Stiglitz and Weiss (1981) argued that there exists an optimum level of interest rates that maximizes banks' expected returns. If interest rates increase, riskier investments will be made to pay back the loans, which is called risk-shifting. Banks with concentrated lending portfolios might have better monitoring skills and could therefore be able to prevent for making too risky investments by borrowers. Additional to this theory, Winton (2000) showed that diversification is only effective if monitoring incentives are low. Otherwise, diversification decreases the average returns, it weakens the monitoring incentives and the bank's chance of failure might increase. Furthermore, Acharya, Hassan and Saunders (2006) found that industrial loan diversification reduces bank return while producing riskier loans, where the effect is the largest for high risk banks. Moreover, they argue that diversification forces a bank to grow in size and this leads to weakened monitoring incentives and therefore to diseconomies of scope. This could be explained by larger monitoring power of focused banks, since they have more knowledge and expertise about the sector they lend to and are better able to react to changes. Böve, Düllmann and Pfingsten (2010) showed in a study on German banks that sectoral specialization improves monitoring for cooperative and saving banks. In contrast to Winton (2010), Hayden et al. (2007) found in a study on German banks for the majority of their dataset almost no performance benefits from diversification across different industries and geographical regions. Only for high risk banks, diversification increases profitability, which is in contrast to the findings of Acharya et al. (2006). 3

7 In a study on one of the largest emerging economies in the world, Brazil, Tabak, Fazio and Cajuiero (2011) found that Brazilian banks' loan portfolios are on average more diversified than loan portfolios in developed countries as the U.S., Germany and Italy. Their overall conclusion is that specialization increases return and risk of default for Brazilian banks. However, Bebczuk and Galindo (2008) found opposite results for Argentinean banks. In the Argentinean crisis in 2001 and 2002, diversified banks faced less non-performing loans than banks with a concentrated loan portfolio. 3. Data This study uses data on the 430 largest listed banks from 60 countries over the period (Appendix 2). A database with manually collected data has been merged with data gathered from Bankscope, Datastream, Barth Caprio and Levine database and a few other databases. The manually collected database contains annual data of 325 of the largest listed banks in the world. For each bank that reports extensively, loan sectoral allocation data is collected from annual reports and categorized into ten economic sectors based on the Standard Industrial Classification (SIC). Personal/consumer loans, loans to central governments and interbank loans are excluded. In some cases, items could be allocated to more than one category or some other items were not a clear fit to one of the ten categories. Assumptions are made for these items and these are put in a file, which can be found in the appendix 3. A more elaborated description of the data collection and an example can be found in appendix 1. Sometimes, there is a huge change in the way of reporting for a single bank in the sample period, which resulted in an unrealistic change in sectoral allocation at one bank. In that case, the data from the most recent years is kept for accuracy. Besides that, occasionally banks do not report a sectoral breakdown of their loan portfolio at all. In those cases, the banks are excluded from the dataset. For the manually collected data, data is available for 325 banks. However, the panel dataset is unbalanced due to the unavailability of sectoral exposures for all banks for the whole time span. 4. Methodology The purpose of this study is to examine if there is a relation between the concentration of banks' corporate lending portfolio and banks' risk. To measure bank's risk, the dependent variable in the model, accounting-based data and stock price return-based is used. Moreover, different methods are used to measure the concentration of banks' loan portfolio. These measures are based on both accounting data and on stock market return data. Since a bank's risk is not only determined by the 4

8 concentration of the lending portfolio, control variables will be added to the model. The following model is estimated: y i,j,t = X 1 i,j,t-1β 1 + X 2 i,j,t-1β 2 + tyear t + jcountry j + ε i,j,t. (1) In this formula, i, j and t stand for bank, country and time respectively. The dependent variables are lagged one year to deal with endogeniety. Furthermore, country-year dummies are included for each regression for fixed effects to control for time and country heterogeneity that cannot be observed, such as changes in the macroeconomic conditions. Since the variance in degree of concentration on bank-specific level is small, bank fixed effects are not included in this model. The standard errors are robust and clustered at the country-year level for the accounting-based measures for concentration, since the variance in concentration at the bank-level is very low. Since the variance in concentration with the return-based measure is larger, standard errors are robust and clustered at the bank-level here, to make the significance levels more accurate. 4.1 Measuring Banks' risk The dependent variable y i,j,t in formula (1) is banks' risk. Two ways to measure banks' risk are used; risk is calculated on accounting-based data and on stock price return-based data. First, banks' daily stock returns are used to calculate the market-based risk. The excess returns of a single bank consists of the correlation with the market return and a bank-specific shock: R i,t = β i,tr m,t + e i,t. Based on this model, the total risk (volatility) can be calculated: σ i,t = β i,tσ m,t + σ,e,t, where σ i,t represents the total risk, and σ m,t and σ,e,t the systematic and idiosyncratic risk, respectively. The larger the total volatility, the larger the bank's risk. Second, the effect of concentration on a bank's systematic risk is tested. The expectation is that banks with a diversified lending portfolio are more exposed to market-wide changes and economic shocks. Contrary, more concentrated banks will face less market risk, which is interesting since some banks went bankrupt in systemic crises and other banks needed government intervention. Market betas (β m,i), are used as dependent variable to measure the market risk. Third, banks' risk based on accounting-based data is measured as the natural logarithm of the z-score (as in Beck et al.(2012), Demirguc-Kunt and Huizinga (2010) and many others). The z-score equals the amount of standard deviations that the return on assets (ROA) has to fall for a bank to get insolvent. The z-score is calculated as follows: 5

9 Zi,t=. (2) In this formula, the ROA equals the return on assets, E/A the equity over assets and σ(roa) the standard deviation of the return on assets. The larger the z-score, the lower the probability of insolvency. Thus, the z-score has to be interpreted the other way around as the volatility of stock returns; a high z-score means low bank risk. Since the panel data is unbalanced, the z-score is calculated over three years to avoid that the measure is calculated over different time horizons. 4.2 Measuring Banks' lending portfolio concentration The explanatory variable of interest in formula (1), X 1,is the concentration of banks' corporate lending portfolio. In this research, two methods are used to measure banks' sectoral exposure. First, data from banks has been gathered, as described in the data part of this paper, to come up with two accounting-based measures of concentration of banks' lending portfolio. The corporate loan exposures have been allocated to a data template that consists of ten industry sectors (as in table 1). From this data template, two different measures are calculated; the concentration ratio and the Herfindahl-Hirschman Index. Second, sectoral exposures are derived from banks' stock price returns and this measure is called dispersion. The concentration ratio that is used in this research is calculated by summing the three largest exposures to sectors a bank has lent to each year, and this ratio will be called CR3 from here. A large CR3 means that a bank has a concentrated corporate lending portfolio. Another measure is used to calculate concentration using all the sectoral categories, which is the Herfindahl-Hirschman Index (HHI). The HHI is calculated per firm each year and is the sum of the squared shares of each loan category relative to the total corporate loans and the formula is as follows: = i 2 (3) When calculating the HHI and the CR3, sector 10 (S10) is ignored. S10 stands for all the other loans and is a pool of loans that cannot be allocated to the nine sectors. Therefore, it is a collection of (a large number of) small exposures and if the exposure in S10 is large, it means that the bank's corporate lending portfolio will be diversified. When using this large exposure in the calculations of the HHI and CR3, it literally states that the lending portfolio is concentrated, while it is actually not. By omitting S10, the HHI and CR3 in this model are more realistic. 6

10 Additionally to the accounting-based way of determining the concentration of banks' corporate lending portfolio, the concentration is calculated by a stock price return-based approach. This method is based on the method used by Beck and De Jonghe (forthcoming). As discussed earlier, perfectly diversified banks are only exposed to systematic risk. This should mean that the stock returns of those banks should co-move with the stock returns on a broad market-wide index. If a bank is more exposed to a certain sector, the stock returns are more affected by news from that specific sector. This causes the bank to be not only exposed to systematic risk, but also to idiosyncratic risk. The model used here is based on the model used by Beck and De Jonghe (forthcoming). This model tests if banks are mainly exposed to the market index and are therefore very well diversified, or if banks have large exposures to specific sectors and can therefore be treated as more focused banks. For this model, the following equation is estimated for each bank: R i,t = c +β i,t R m,t + + (4) Here, banks' stock returns (R i,t) are regressed on the returns of a broad market-wide index (R m,t and on ten indices that are exposed to returns on specific sectors (R s,t). The residual (ε i ) is the bankspecific news component. The level 2 decomposition of the Industry Classification Benchmark (ICB) is used to create the different sectors. This method divides the total market into the following ten sectors: Oil & gas, Basic materials, Industrials, Consumer goods, Healthcare, Consumer services, Telecommunications, Utilities, Technology, and Financials. The returns on the ten sectors are orthogonalized with respect to the market returns and the financial sector returns. After othogonalizing (R s,t), the returns on the ten sectors are only exposures to sector-specific shocks and not influenced by market-wide news or shocks in the financial sector. Furthermore, the sectoral exposures are standardized allowing comparison between the different sectors. Negative exposures to specific industries are allowed in the model, which means that banks can have a short position in an industry. From the estimated coefficients, the standard deviation in the estimated sectoral exposures, the dispersion, is calculated to construct a measure for the concentration of the banks. For the construction of this measure, only the exposures with a t-statistic larger than 1 in absolute values are used. Furthermore the sector Financials is excluded, assuming that a large part in this exposure consists of interbank loans and this is also excluded when calculating the measure based on accounting data. The larger the dispersion, the more it differs from the market-wide index. The bank is therefore more exposed to idiosyncratic news and thus treated as a bank with a concentrated lending portfolio. 7

11 An important remark is that this measure might have different outcomes than the outcomes based on the accounting-based measures. The exposures based on stock market returns may be underestimated if a bank uses derivative contracts to hedge some exposures. For example, it could be that a bank has a large exposure to the automobile industry. If the bank has used hedging instruments to limit the risk on this exposure, the estimated results are an unrealistic exposure to the automobile industry. Furthermore, investors have not all information, so stock prices will not reflect precisely banks' sectoral exposure. 4.3 Control variables Balance Sheet and Income Statement variables are included as control variables, X 2 in equation (1). The control variables bank capital, size growth, the share of deposits funding, the return on assets, cost inefficiency, the share of loans to assets, loan loss provisions, share of non-interest income and bank size describe the bank's business model. The equity-to-assets ratio, bank capital, is included to control for the effect of leverage on banks' risk. The share of deposits funding to total funding is included as measure of the fundmix. Cost-to-income, the ratio of all operating expenses to total income, is included as efficiency measure. The loan-toasset ratio illustrates the asset mix of the bank. The loan loss provisions reported in the annual reports give a signal of the loan quality and is included as measure of credit risk. The return on assets is included as performance measure. Furthermore, banks have shifted into financial activities that generate non-interest income. These revenues are volatile and growing rapidly (De Young and Rice (2004), Stiroh (2006)). Therefore, the share of non-interest income to total income is included as measure of the income mix. Finally, the size and the growth of size are included as the natural logarithm of total assets and the total assets growth. Since bank size is correlated with a lot of the other control variables and with the concentration measures, bank size is orthogonalized. This means that bank size is regressed on the other variables and that the residual is used as a pure size effect (Baele et al., 2007). 8

12 5. Results 5.1 Summary statistics Table 1 presents the summary statistics on the sectoral allocation. On average, the shares of exposures to 'manufacturing', 'wholesale and retail trade' and 'real estate' are quite large with exposures of 16.0%, 13.8% and 12.9%, respectively. On the other hand, banks have on average small exposures in 'agriculture, forestry and fishing' and 'public administration'. The exposure to the category 'other industries' is large, but this can be treated as a pool of all kind of small exposures, so a large exposure to other industries characterizes a diversified bank instead of a bank with a concentrated lending portfolio. An important note to these results is that there are globally no strict guidelines how to present the sectoral allocation of the corporate lending portfolio. In the data collection part is described how an accurate template was designed that creates data as optimal as possible for this research. Table 1: Summary statistics on sectoral allocation Sectors Obs Mean Std.Dev. Min Max S1 Agriculture, Forestry and Fishing S2 Mining & Construction Manufacturing Transportation, communication, S4 Electric, Gas and Sanitary service S5 Wholesale trade and Retail trade S6 Finance and Insurance S7 Real estate Services S9 Public administration S10 Other industries This table presents the summary statistics on the sectoral allocation. Corporate lending portfolios of the largest global listed banks are divided among 10 categories. Data is collected from 2007 to However, not from all banks data are available for the whole time span. For the collected observations, for each sector, the mean, standard deviation, minimum and maximum of the exposures are calculated. The summary statistics of each year of the sample period of the dependent variables are presented in panel A of table 2. The average total risk increases a lot from 2007 to 2008 and decreases after This peak is caused by the global financial crisis in the late 2000s. This peak cannot be observed in the averages of market risk and the z-score. Since risk is high when the z-score is low, banks' risk is 9

13 on average at the largest level in 2009, one year later than based on the total risk. The same holds for banks' market risk. Panel B of table 2 contains the correlation matrix between de dependent variables total risk, the market beta and de z-score. The z-score measures risk in the opposite direction as total risk and the market beta, this explains the negative coefficients for the z-score. Furthermore it is remarkable that the z-score and total volatility are both measures for firm risk and that they are not highly correlated. Table 2: Summary statistics risk measures Panel A: summary statistics dependent variables σ i Mean Stand. dev Min Max β m Mean Stand. dev Min Max Z-score Mean Stand. dev Min Max Panel B: Correlation matrix dependent variables σ i β m Z-score σ i β m (0.000) Z-score (0.000) (0.000) Panel A of this table presents summary statistics on the dependent variables for risk. It contains information on the total risk of the banks, σ i, the market beta, β m, and the accounting-based risk measure, the z-score. For the variables the mean, standard deviation, the minimum and the maximum are presented for each year. Panel B of this table presents the correlations between the dependent variables. The table contains the correlation coefficients, and in parentheses the p- values, of the total risk, market beta and z-score. The summary statistics of the measures of concentration of banks' lending portfolio are presented in table 3. The measures CR3 and HHI are calculated based on the manually-collected accounting-based data and the dispersion is calculated from stock return-based data. The mean and the standard deviation from both the CR3 and the HHI are quite constant over the sample period. Since the variation in concentration for a single bank is very small in the sample period, including bank fixed 10

14 effects is not useful and the standard errors are not clustered at bank-level when CR3 and HHI are used in the regressions. For dispersion, the mean and the standard deviation are more volatile. Therefore, when regressing bank risk on dispersion, the standard errors are clustered at bank-level. The correlation matrix of all independent variables is presented in table 4. The HHI and CR3 are strongly correlated and this is in line with expectations. Remarkably, the dispersion is weakly and even negatively correlated with HHI and CR3, but these correlation coefficients are not significant. Bank size has significant correlation coefficients with a large share of other independent variables and is therefore orthogonalized in all the regressions. Table 3: Summary statistics of the concetration measures CR3 Mean Stand. dev Min Max HHI Mean Stand. dev Min Max Dispersion Mean Stand. dev Min Max This table presents summary statistics on the independent variables that measure the concentration of banks' corporate lending portfolio. It contains information on the concentration ratio, CR3, the Herfindahl-Hirschman index, HHI, and the standard deviation on the sectoral exposures based on stock return data, dispersion. For the variables the mean, standard deviation, the minimum and the maximum are presented for each year. 11

15 Table 4: Correlation matrix independent variables Growth in total assets Fund mix Return on assets CR3 HHI Dispersion Equity-toassets Cost-toearnings Loans-toassets Loan loss provisions Non-interest income share Ln(size) CR HHI (0.000) Dispersion (0.595) (0.562) Equity-to-assets (0.000) (0.000) (0.000) Growth in total assets (0.000) (0.013) (0.353) (0.000) Fund mix (0.000) (0.000) (0.928) (0.000) (0.000) Return on assets (0.182) (0.880) (0.000) (0.000) (0.000) (0.075) Cost-to-earnings (0.192) (0.015) (0.302) (0.000) (0.001) (0.000) (0.000) Loans-to-assets (0.000) (0.000) (0.471) (0.760) (0.027) (0.000) (0.001) (0.017) Loan loss provisions (0.000) (0.007) (0.000) (0.000) (0.011) (0.007) (0.000) (0.000) (0.000) Non-interest income share (0.000) (0.001) (0.566) (0.000) (0.632) (0.000) (0.000) (0.000) (0.000) (0.713) Ln(size) (0.000) (0.000) (0.000) (0.000) (0.555) (0.000) (0.002) (0.042) (0.001) (0.229) (0.000) This table presents the correlations between the dependent variables. The table contains the regression coefficients and in parentheses the p-values. The first three variables are the different measures for concentration of banks' corporate lending portfolio. The other variables are the control variables. Size is the Natural Logarithm of size and is here not orthogonalized as used in all the regressions. 13

16 Table 5 shows the results from the regression of different measures of bank risk on the control variables to show the relations between the control variables and the dependent variables. After discussing these relations, the coefficients of the control variables will not be reported anymore. In line with expectations, the outcomes of total risk are positive when the z-score is negative and vice versa. However, the significance level are not the same for all the variables. The equity-to-asset ratio is negatively related to all the risk factors, which is in line with the results of Stiroh (2006). It means that banks with a higher capital buffer are less risky than banks with a low capital buffer. Growth of total assets is positively related with risk, but not significant for the z-score. If deposits funding to total funding increases, risk decreases regarding to the z-score. The market beta is positively related to the fundmix, so if retail deposits funding gets larger with respect to wholesale funding, banks' market risk gets larger. The return on assets is not significantly related to banks' risk. The efficiency Table 5: Regressions of control variables on Bank Risk σ i β m Z-score Equity-to-assets *** *** 0.097*** (0.016) (0.005) (0.025) Growth in total assets 0.006*** 0.002** (0.002) (0.001) (0.002) Fundmix *** 0.647** (0.295) (0.155) (0.328) Return on assets (0.0858) (0.0320) (0.112) Cost-to-earnings 0.813* *** (0.492) (0.213) (0.495) Loans-to-assets *** (0.284) (0.149) (0.390) Loan loss provisions 33.94*** 13.22*** *** (10.01) (4.254) (8.980) Non-interest income share 0.700* * *** (0.414) (0.145) (0.417) Ln(size) ** 0.136*** (0.037) (0.019) (0.045) Constant 2.367*** 1.107*** 4.788*** (0.480) (0.235) (0.524) Observations 1,459 1,459 1,172 R-squared This table presents OLS regressions of different measures of risk on the control variables. The first, second and third column show the coefficients of the control variables on total risk, the market beta and the z-score, respectively, over the period Bank size is the residual of a regression of Ln(size) on all the other control variables, so that a bank size is made orthogonal and here the pure size effect. Country-year dummies are included in all regressions, but not reported. All independent variables are lagged one year. Standard errors are clustered and robust at the country-year level. ***, **, * show the statistical significance at the 1%, 5% and 10%, respectively. The standard robust errors are reported in parentheses. The amount of observations for the z-score is smaller, since the independent variables are lagged one year and accounting data on the exposures of concentration is only collected for the 5-year period,

17 of banks, measured by the cost-to-earnings ratio, is positively related to risk, but not to market risk. The loans-to-assets ratio is negatively related to the market beta, which is consistent with Baele et al. (2007). If banks have more traditional banking activities, the market risk is smaller. In contrast, the total risk is not significantly related with the loan-to-assets ratio in this sample. On all risk measures, loan loss provisions has a positive coefficient as expected. The non-interest income share is negatively related to the z-score like in Beck et al. (2012). But in contrast to Baele et al. (2007), noninterest income share is weakly significant negatively related to market risk. In line with this study it is weakly and positively related to total risk. Finally, the pure size effect is in line with Baele et al. (2007) positively related to total risk and market risk and not significant to the z-score, which is in contrast to Beck et al. (2012) and Stiroh and Rumble (2006). The amount of observations for the z- score is smaller, since the independent variables are lagged one year and accounting data on the exposures of concentration is only collected for a 5-year period. 5.2 Results of the accounting-based concentration measures A. Global Analysis Table 6 presents OLS regressions of concentration of banks' corporate lending portfolio on total risk. Column 1 shows there is no significance (a p-value of 0.351) for a linear relationship between CR3 and total volatility. The second table shows the non-linear relationship between total volatility and CR3. This relation is not positive, but de p-value increases to a joint significance of CR3 and (CR3) 2 of Statistically, both coefficients are not significant. Moreover, the negative coefficient of the linear relationship indicates that specialization of the corporate lending portfolio results in less risk. This is exactly the opposite to the expectation that more concentrated portfolios are riskier. Since the regression where a quadratic term is included greatly increases the significance, the function is drawn in figure 1 to show the relation. A first look at the graph indicates that there exists a u-shaped relation between total risk and concentration. However, the summary statistics in table 3 panel A show that the mean CR3 equals on average The standard deviation equals approximately The turning point of the parabola is at a total volatility of Summarizing, only 17.3% of the observations are more concentrated than the optimal amount of concentration for the minimum amount of total risk. So if the joint significance in column 2 would have been significant, it would have been true that at a certain level of concentration, banks face more risk. However, the hypothesis that concentration of banks' corporate lending portfolio leads to bank risk cannot be accepted due to insignificant results. 17

18 Total Riks Column 3 and 4 of table 5 show the results for the linear and quadratic effect of the HHI on total risk, respectively. Although the HHI is calculated from the same data as the CR3, the results are not significant. Since the results are strongly insignificant and the sign is even negative for both the linear and the quadratic function, the results do not support the hypothesis that concentration leads to bank risk. Table 6: Total risk regressions σ i σ i σ i σ i CR * (0.177) (0.564) (CR3) (0.444) HHI (0.298) (0.914) (HHI) (1.881) Constant 2.435*** 2.668*** 2.383*** 2.385*** (0.466) (0.449) (0.462) (0.450) Joint significance Nr. of banks Observations 1,459 1,459 1,459 1,459 R-squared This table presents OLS regressions of total risk, σi, on measures of banks' corporate lending portfolio concentration in the period The first column show the coefficient of CR3 on total risk. In the second column, a quadratic term of CR3 is added for non-linearity. The third column shows the coefficient of the HHI on total risk and in the fourth column, the quadratic term of HHI is added for non-linearity. The control variables as presented in table 4 are used, but not reported. All independent variables are lagged one year. Country-year dummies are included in all regressions, but not reported. Standard errors are clustered and robust at the country-year level. ***, **, * show the statistical significance at the 1%, 5% and 10%, respectively. The standard robust errors are reported in parentheses. The row Joint significance reports the p- values from the Wald test of the linear and quadratic term in column 2 and 4 of CR3 and HHI, respectively Non-linear relation CR3 Figure 1: Non-linear relation between CR3 and total risk This figure shows the non-linear relationship between total risk, σ i, and concentration, CR3, as shown in column 2 of table 6. 18

19 Next, the regressions on market risk are discussed. The results of the regressions are presented in table 7. It is hypothesized that banks with a well diversified corporate lending portfolio have more systematic risk and are thus more affected by market shocks. Column 1 and 3 of table 7 show that CR3 and HHI are indeed negatively related to market risk. Moreover, column 2 and 4 show that the linear and quadratic term are jointly significant. Table 7: Market beta regressions β m β m β m β m CR *** * (0.0933) (0.284) (CR3) (0.236) HHI *** (0.162) (0.400) (HHI) (0.816) Constant 1.208*** 1.285*** 1.117*** 1.096*** (0.239) (0.238) (0.232) (0.225) Joint significance 0.002*** 0.026** Nr. of banks Observations 1,459 1,459 1,459 1,459 R-squared This table presents OLS regressions of market risk, βm, on measures of banks' corporate lending portfolio concentration in the period The first column shows the coefficient of CR3 on the market beta. In the second column, a quadratic term of CR3 is added for non-linearity. The third column shows the coefficient of the HHI on market risk and in the fourth column the quadratic term of HHI is added for non-linearity. The control variables as presented in table 4 are used, but not reported. All independent variables are lagged one year. Country-year dummies are included in all regressions, but not reported. Standard errors are clustered and robust at the country-year level. ***, **, * show the statistical significance at the 1%, 5% and 10%, respectively. The standard robust errors are reported in parentheses. The row Joint significance reports the p-values from the Wald test of the linear and quadratic term in column 2 and 4 of CR3 and HHI, respectively. Both the linear and the quadratic functions of the CR3 and HHI are shown in figure 2 and 3, respectively. In figure 2 as well as in figure 3, the quadratic function and the linear function show almost the same relation. Both measures show a negative relation between concentration and market risk. However, the quadratic function in figure 2 is convex and in figure 3 the quadratic function is concave. The negative relation in figure 2 decreases from the mean CR3 of So more concentrated banks are less affected by market shocks, but the slope decreases slightly. A first look at figure 3 indicates an increasing negative effect. But the summary statistics in table 3 show a mean of 0.18 and a standard deviation of The graph shows that the quadratic line starts to deviate from the linear line from approximately a HHI of 0.38, which is 2 standard deviations from the mean. Therefore, it is hard to say that the negative effect of concentration on market risk is increasing. It is statistically and economically significant and in line with expectations that banks' corporate lending portfolio concentration is negatively related to the market beta. 19

20 Market beta Market beta Non-linear relation Linear relation CR3 Figure 2: Linear and non-linear relationship between CR3 and the market beta. This figure shows the linear and non-linear relationship between the concentration of banks' corporate lending portfolio, measured by the CR3, and banks' systematic risk, βm. The figure is drawn based on the regression information presented in table 7, column 1 and HHI Non-linear relatiion Linear relation Figure 3: Linear and non-linear relationship between HHI and the market beta. This figure shows the linear and non-linear relationship between the concentration of banks' corporate lending portfolio, measured by the HHI, and banks' systematic risk, βm. The figure is drawn based on the regression information presented in table 7, column 3 and 4. 20

21 Next, table 8 shows the results of the relation between the concentration measures, CR3 and HHI, and the z-score. The z-score is based on accounting data and measures the distance to insolvency. More specifically, a low z-score means high risk. Therefore a negative relation is expected for the hypothesis that concentration of banks' lending portfolio leads to bank risk. Column 1 and 3 present the results for a linear relation presented between the z-score and CR3 and HHI, respectively. The results are not significant. The possibility for non-linearity is presented in column 2 and 4 of table 8. The coefficients for CR3 and the quadratic term of CR3 are significant at the 10% level, but the variables are not jointly significant. However, the significance increases from a p-value of for the linear term to a p- value of for the joint significance. This indicates that there might be a trend between CR3 and the z-score. The non-linear relation between CR3 and the z-score is drawn in figure 4. The average CR3 equals 0.61, which is on the left-hand side of the top of the parabola of With a standard deviation of approximately 0.17, it seems that from a certain level of concentration, concentration indeed leads to risk. Table 8: Z-score regressions z-score z-score z-score z-score CR * (0.240) (1.403) (CR3) * (1.082) HHI (0.391) (1.468) (HHI) (2.891) Constant 4.731*** 4.013*** 4.782*** 4.711*** (0.571) (0.720) (0.544) (0.576) Joint significance Nr. of banks Observations 1,172 1,172 1,172 1,172 R-squared This table presents OLS regressions of bank risk based on accounting data, measured by the z-score, on measures of banks' corporate lending portfolio concentration in the period The first column shows the coefficient of CR3 on the z- score. In the second column, a quadratic term of CR3 is added for non-linearity. The third column shows the coefficient of the HHI on bank risk and in the fourth column the quadratic term of HHI is added for non-linearity. The control variables as presented in table 4 are used, but not reported. All independent variables are lagged one year. Country-year dummies are included in all regressions, but not reported. Standard errors are clustered and robust at the country-year level. ***, **, * show the statistical significance at the 1%, 5% and 10%, respectively. The standard robust errors are reported in parentheses. The row Joint significance reports the p-values from the Wald test of the linear and quadratic term in column 2 and 4 of CR3 and HHI, respectively. 21

22 Z-score non-linear relationship CR3 Figure 4: Non-linear relationship between CR3 and the z-score. This figure shows the non-linear relationship between the concentration of banks' corporate lending portfolio, measured by the CR3, and banks' risk based on accounting data, the Z-score. The figure is drawn based on the regression information presented in table 8, column 2. B. Analyses on banks in developed and emerging countries Table 9: Summary statistics developed countries and BRICs Panel A: Developed countries CR3 Mean Stand. dev Min Max HHI Mean Stand. dev Min Max Panel B: BRICs CR3 Mean Stand. dev Min Max HHI Mean Stand. dev Min Max This table presents summary statistics on the independent variables that measure the concentration of banks' corporate lending portfolio for the 6 largest developed countries in Panel A and for the BRICs in panel B. It contains information on the concentration ratio, CR3, and the Herfindahl-Hirschman index, HHI. For the variables the mean, standard deviation, the minimum and the maximum are presented for each year. 22

23 Z-score The data collected for this research makes it possible to examine if there are different relations between risk and banks' sectoral concentration across countries. The focus here is on the ten largest economies in the world in 2012, based on GDP in US dollars: The United States of America, China, Japan, Germany, France, United Kingdom, Brazil, Russian Federation, Italy and India. In this sample, Brazil, the Russian Federation, India and China are the so called BRICs, that are large emerging market economies. The other countries mentioned above are the six largest developed economies in the world. Summary statistics of the developed countries and the BRICs are presented in table 9 panel A and B, respectively. The mean CR3 and HHI are higher for the BRICs than for the developed countries. So on average, banks' corporate lending portfolios are more diversified in the developed countries, as in Tabak et al. (2011). Table 10 shows the results of the relation between risk and banks' lending portfolio concentration in the developed countries. Panel A, column 1 and 2, show a negative relation between total risk and CR3 at a 10% significance level. Furthermore, column 3 and 4, show a negative relation between the market beta and CR3 at a 10% significance level. For the z-score, the coefficients are not significant. In Panel B, the HHI is used as concentration measure. The coefficients for total risk, market beta and z-score are more or less in the same direction as with the CR3. However, for the HHI, only the coefficients on the z-score are significant at the 10% level. The relationship here is non-linear as shown in figure 5. From a HHI of 0.14, which is on the left-hand side of the mean of approximately 0.15, the line is upward sloping. So concentration leads to less risk from this point. The turning point is approximately 1.5 standard deviations from the mean. Far from the mean, it seems that more concentration leads to more risk, according to the z-score. Overall, the results are not highly significant, but the results show that banks with diversified corporate lending portfolios are riskier in the developed countries non linear HHI Figure 5: Non-linear relationship between HHI and the Z-score in developed countries. This figure shows the non-linear relationship between the concentration of banks' corporate lending portfolio, measured by the HHI, and banks' risk based on accounting data, the Z-score. The figure is drawn based on the regression information presented in table 10, panel B, column 6. 23