Are banks in more concentrated markets less stable? Evidence from the EU-25.

(1)

Are banks in more concentrated markets less stable?

Evidence from the EU-25.

Pieter IJtsma

∗

Supervisor: Prof. Dr. Laura Spierdijk

February 3, 2014

Abstract

The existing empirical literature regarding the relationship between banking market concentration and financial stability has either used a pure country-level or a pure level approach, ignoring the fact that both country-level and bank-level characteristics are important determinants of financial stability. This paper tries to bridge this gap through an operationalization of the notion of systemic loss, which emphasizes the importance of default probabilities of banks, loss given default, and correlation of defaults across banks. The results do not provide clear evidence of a positive or negative association between concentration and financial stability.

Keywords: Financial stability, market structure JEL Classification: G21, G28, G34, L110, L16

∗_{University of Groningen, Faculty of Economics and Business, Groningen, The Netherlands. E-mail:}

(2)

1 Introduction

Since the 1990s, a high level of mergers and acquisitions (M&As) has resulted in increas-ing consolidation in the financial sector of many developed economies (Group of Ten, 2001), with a peak in M&A activity occurring in the period surrounding the introduction of the euro (Berger, 2007). As a result, the financial sector tends to become increasingly concentrated in many countries, resulting in a small number of large banks dominating the financial landscape. Although M&As in the banking sector can increase efficiency and reduce costs through economies of scale and scope (Group of Ten, 2001), the as-sociated increase in market concentration within the financial sector might also have an effect on financial stability at the macroeconomic level. Indeed, proponents of the so-called concentration-fragility view argue that, for a number of reasons, concentrated banking systems can be expected to negatively affect financial stability. Others, how-ever, have come up with arguments supporting the hypothesis that concentrated banking systems result in higher financial stability, and are generally referred to as proponents of the concentration-stability view.

Where the theoretical literature about the relationship between banking market con-centration and financial stability provides arguments for both views, the same can be said about the empirical literature evidence this subject. A weakness of the existing empirical literature, however, is that it does not acknowledge the multi-dimensional na-ture of the concept of financial stability, as existing analyses have either used a pure country-level approach or an approach entirely at the bank level of analysis. Financial stability, however, is a concept which depends on both country-level and bank-level char-acteristics, as has been clearly pointed out by ˘Cih´ak (2007), who criticizes all measures of financial stability which have been used in the empirical literature so far.

(3)

by ˘Cih´ak (2007) as a more appropriate indicator of macroeconomic financial stability. Moreover, to compare the results, a pure country-level and pure bank-level approach will be performed as well. The paper proceeds as follows: in section 2, the existing theoretical and empirical literature about the relationship between banking market con-centration and financial stability will be reviewed. The empirical strategy taken in this paper will be elaborated upon in the third section, after which a description of the data follows in section 4. The fifth section presents the results of the country-level analysis, in which a replication of Uhde and Heimeshoff (2009), who obtained very strong results in favor of the concentration-fragility view, will serve as the starting point. The results of the bank-level and multi-level approach will then be presented in section 6, after which a conclusion follows in section 7.

2 Literature Review

In this section, the theoretical and empirical literature regarding the relationship be-tween banking market concentration and financial stability will be presented. It will be shown that theoretical arguments exist for both the concentration-stability view and the concentration-fragility view, so that, theoretically, the relationship between concen-tration and stability is an ambiguous one. Moreover, empirical evidence regarding the subject is inconclusive as well, with some evidence in favor of a positive relationship be-tween concentration and stability, and other analyses pointing at a negative association. Arguments in favor of the concentration-stability view

(4)

(5)

financial contagion introduced by Allen and Gale (2000). In the model, banks utilize interbank claims for insurance against liquidity shocks, but these claims are also the source of contagion when one of the banks in the system fails. S´aez and Shi (2004) extend the model and argue that when one bank fails, the remaining (healthy) banks may have an incentive to save it from bankruptcy to prevent the negative effects of financial contagion. However, if the number of banks in the system is large, free riding by individual banks might result in a coordination problem, preventing an agreement from being reached. In their model, financial contagion is thus less likely in a more concentrated market.

(6)

as a general problem in all banks with similar characteristics, resulting in bank runs and liquidity problems for solvent banks. Both counter-arguments illustrate that financial stability does not only depend on the variance of banks returns, but also on the corre-lation between them. Hence, this is something that needs to taken into account when empirically investigating the effects of concentration on financial stability.

Arguments in favor of the concentration-fragility view

Many of the assumptions underlying the arguments of the concentration-stability view can also lead to arguments for the concentration-fragility view. First, in the model of Boyd et al. (2006), banks not only compete in deposit markets but also in loan markets. Boyd et al. argue, then, that in less competitive loan markets, banks are able to charge higher interest rates. As a result, entrepreneurs that borrow from these banks will engage in riskier project, so that the riskiness of the bank’s asset portfolio increases as well. They show that this loan market effect dominates the charter value effect, so that a lower level of competition should be associated with more financial fragility. Hence, if competition and concentration are negatively related, higher levels of concentration will lead to increased fragility.

Secondly, Leitner (2005) argues that financial contagion is more likely in a concen-trated system because for large banks, the benefits of preventing contagion might not outweigh the costs of bailing out a failing bank. Moreover, Nier et al. (2007) show that in a concentrated system, the probability that one bank is large enough to make a significant impact on the rest of the system increases with the degree of market con-centration, so that the risk of contagion is higher in more concentrated markets. This is the case even when the size of a liquidity shock is held constant, because in a more concentrated market not only the banks themselves are larger, but so are the interbank connections between them.

(7)

it is often argued that large banks are more prone to moral hazard resulting from too-big-to-fail policies (TBTF) (De Nicol´o, 2000; De Nicol´o et al., 2004; Schaek et al., 2009; Vives, 2010). The argument is that when bank managers know that their bank is considered TBTF, they will engage in risky behavior, as the downward risk is not borne by the bank itself. Dam and Koetter (2012) provide some evidence that bank managers who expect to be bailed out in case of failure indeed engage in more risk-taking behavior. Mishkin (2005) argues, however, that the effects of TBTF policies have not played a large role in most banking crises. Moreover, in many countries, especially LDCs, almost all banks have been considered TBTF. If that is the case, the degree of market concentration should not affect financial stability through this mechanism.

Since theoretical arguments for both views exist, the relationship between concen-tration and financial stability is ultimately an empirical matter. In the remainder of this section, the empirical evidence regarding this relationship will be presented. The first subsection focuses on analyses which have been performed at the country level, whereas the second section presents the empirical results of several bank-level analyses.

Evidence from the country level

(8)

al. (2006) expand the analysis to 69 countries and add a range of control variables to take into account differences in macroeconomic conditions and regulatory policies. Their results confirm those of D-D (2002), although the effect of concentration becomes insignificant when country size is controlled for. Schaek et al. (2009), finally, use a different sample and obtain similar results when controlling for competition. They do this through the inclusion of (estimated) H-statistics, which are equal to the sum of the partial elasticities of a bank’s revenue with respect to factor prices. In case of perfect competition, there should be a one-to-one relationship between revenues and costs, so that the H is equal to 1. As market power increases, the relationship between revenues and costs becomes weaker, and might even become negative. Hence, higher H-statistics should in theory be reflection of more intense competition (Panzar and Rosse, 1987). Note, however, that Bikker et al. (2012) demonstrate formally that the H-statistic is an inappropriate proxy of competition. Schaek et al. also perform a duration analysis to estimate the effect of concentration on the “time until crisis” of a country. In line with the concentration-stability view, they find that higher concentration lowers the probability of a system crisis, while it increases the time until crisis.

(9)

their inclusion makes it impossible to disentangle between the effects of an adverse shock to the banking industry and the effects of policies introduced to restore the stability of the industry. Boyd et al. argue that financial crises are associated with a decline in loans, deposits and bank profits, and with an increase in loan rates. They consequently develop indicators of financial crises which depend on these variables, and show that lagged values of their indicators are able to predict the indicators which depend on policy interventions, including the indicator of D-D (2002). Furthermore, Boyd et al. estimate the relationship between concentration and stability, and show that with their indicators higher levels of concentration are associated with an increase in the probability of a crisis. Hence, the estimated effect of concentration appears to depend on the indicators that are used to proxy a financial crisis.

Another analysis which supports the concentration-fragility view was performed by Uhde and Heimeshoff (2009), who will be referred to as U-H in the remainder of this paper. They do not look at real episodes of financial crises, but focus on the Z-score as a measure of stability. The Z-score is defined as follows:

Z = µ +

e a

σ(µ) (2.1)

(10)

they are negatively affected by concentration.

Overall, then, the country-level evidence regarding the relationship between concen-tration and stability does not paint a clear picture, as some studies indicate a positive association, while the results of other studies point to the opposite. In the next subsec-tion, the empirical evidence resulting from bank-level analyses will be presented. Evidence from the bank level

One of the first bank-level analyses of financial stability was performed by Boyd and Runkle (1993). They investigated whether large banks are less likely to fail compared to small banks, but did not obtain conclusive results. Their work, however, was continued with by De Nicoló (2000), who looked at banks from 21 industrialized countries and found a negative association between the size of banks and their Z-score in the period between 1988 and 1998. His estimates are consistent across separate analyses for the U.S., Japan and Europe. De Nicoló and Kwast (2002) focus on a different source of financial fragility, namely the correlation of banks’ returns. If the returns of different banks are highly correlated, the probability of a systemic crisis will be higher, holding all else constant. Hence, a positive relationship between concentration and banks’ return correlations offers support for the concentration-fragility view. Using a sample of large U.S. banks in the period between 1988 and 1999, De Nicoló and Kwast indeed obtain a positive consolidation elasticity of return correlations, indicating that return correlations are higher the more concentrated is the banking system. The relationship exhibits substantial time variation, however, so that it is difficult to draw strong conclusions from their results.

(11)

concentration and Z-scores when the five-bank concentration ratio (CR5) is used as the

measure of concentration. Their results also suggest, however, that market share itself increases stability.

To our best knowledge, the only bank-level analysis which indicates a positive re-lationship between concentration and stability is a study performed by Berger et al. (2009). They use a sample of banks from 23 industrialized countries, and find a positive association between the Herfindahl-Hirschman Index (HHI) and individual banks’ Z-scores. Interestingly, the analysis looks at both the market for deposits and the market for loans, and finds a similar relationship in both markets. Hence, although the majority of bank-level analyses supports the concentration-fragility view, the paper of Berger et al. (2009) offers some support for the opposite view.

When analyzing the association between banking market concentration and financial stability, it is important to recognize the possibility of reverse causality, as the failure of banks might have strong effects on market concentration in the banking industry. Hence, a strong association between concentration and stability might not be the result of a causal effect running from market concentration to financial stability, but simply reflect the fact that banking market concentration has increased as a result of bank failure. To establish a true causal effect running form concentration to stability, studies should thus in some way control for potential reverse causality. Table 2.1 gives an overview of the empirical analyses performed, and indicates which studies have indeed controlled for the possibility of reverse causality.

Table 2.1: Overview of empirical analyses on association between banking market concentration and financial stability. ‘Association’ refers to the estimated relationship between concentration and stability.

Author Level Dep. variable Association Reverse causality controls D-D (2002) Country Crisis dummy Positive No

(12)

What the bank-level empirical literature so far is lacking, is an analysis which includes a measure of financial stability that translates information about individual banks into a country-level variable. This is important because financial stability is ultimately a country-level phenomenon. A strong theoretical framework in this respect was provided for by ˘Cih´ak (2007), who argues that a proper measure of financial stability incorporates three elements: (1) the probability of failure of individual banks, (2) the loss given default of individual banks, and (3) the correlations of defaults across institutions. He criticizes existing measurements of financial stability, arguing that both the individual Z-score and its weighted (country) average only look at individual institutions separately, without taking into account the fact that defaults might be correlated. This is an important criticism, which will be taken into account in the remainder of this paper.

3 Empirical Strategy

(13)

the market shares of the five largest banks in the market: CR5 = 5 X i=1 Si (3.1)

where i refers to one of the five largest banks in the market, and S is the bank’s market share in terms of assets. The main benefit of the CR5 is that it as a relatively

straightforward measure and that it suffers little from data limitations, as only data of the largest five banks in a country are needed for its calculation. A downside of this measure is that it ignores all information regarding banks other than the country’s largest five, which is a rather arbitrary cut-off point (Bikker, 2004). The HHI is another often-used measure of concentration, and is defined as the sum of the squared market shares of all banks in the market:

HHI =

n

X

i=1

S_i2 (3.2)

where i refers to an individual bank, n refers to the number of banks in the market, and S is the bank’s market share in terms of assets. The benefit of this measure is that it uses information of all banks in the market, thereby preventing the need for a cut-off point. A downside, however, is that the index is relatively sensitive to the entrance of a high number of very small banks (Hart, 1975). Moreover, different bank size distributions can result in the same value of the HHI (Rhoades, 1995). Bikker (2004) argues that, although the correlation between the CR5 and the HHI is generally high, differences

across countries in the CR5 are mainly the result of differences in the skewness of the

distribution of bank size, whereas differences in the HHI mainly result from differences in the number of banks in the market. For this reason, both measures will be used as a proxy of concentration.

Country-level analysis

(14)

and 2005, and estimate the effect of banking market concentration on financial stability by using consolidated Z-scores as a measure of financial soundness at the country level. The Z-score is defined as in Equation (2.1) and is often referred to as a distance-to-default indicator, as it measures the number of standard deviations with which a bank’s return on assets would have to fall in order for the bank to fail. Hence, higher Z-scores indicate a higher level of stability. The calculation of consolidated Z-Z-scores is straightforward, given that these are simply asset-weighted averages of the Z-scores of all banks in a country. Note, however, that, unlike the conventional Z-score, the consolidated Z-score is not directly interpretable as a measure of distance-to-default, because the measure does not take into account the effects of potential contagion of individual failures. Hence, the consolidated Z-score incorporates the first two elements of financial stability in the framework of ˘Cih´ak, but not the third, making it an imperfect proxy of financial soundness at the country level. Nevertheless, it is interesting to analyze the effect of banking concentration on financial stability when using this measure of financial soundness, given the extremely strong results obtained by U-H, who find a large, robust and highly significant negative effect of banking market concentration on consolidated Z-scores. The analysis thus starts with a replication of U-H (2009), after which the results of several variations to their approach will be presented. These consist of the use of more recent data, the inclusion of better defined control variables, and the inclusion of country fixed effects to control for unobserved time-invariant heterogeneity across countries. The latter are included because a Hausman test indicates that the coefficients resulting from the fixed-effects model are significantly different from those of the random-effects model, which indicates that the estimation of the random-effects model results in biased coefficients.

Bank-level analysis

(15)

analysis will be performed as well. The bank-level analyses in the existing literature have generally used individual bank’s Z-scores as a proxy of financial stability, and this will therefore be the starting point here as well. There are two important caveats to this approach however. First, the Z-score does not take into account the size of a bank, and thus attaches equal importance to very large and very small banks. As a result, the loss given default of individual banks is not incorporated in analyses which make use of sample individual Z-scores. Second, the use of individual bank’s Z-scores in a multi-country analysis leads to an overrepresentation of countries with a high number of small banks, and an underrepresentation of countries with a small number of large banks. From Table A.4 it can be observed that Germany hosts by far the largest number of banks, so that the use of simple Z-scores leads to an overrepresentation of that country. To remedy these problems, a relatively simple and appealing approach will be taken: all observations will be weighted according to the size of the respective bank’s assets relative to those of the country as a whole. Hence, all variables will be transformed by multiplying them with the following weight (ωit):

(16)

Although the weighting procedure described above takes into account the loss given default of individual banks, it does not consider the correlation of defaults across banks. As a result, the approach does not take into account the potentially destabilizing effects of financial contagion. For this reason, a second transformation will be performed in which the weights are higher for banks which exhibit a high correlation between their Z-score and the consolidated Z-score of the country in which they are located. The weights are defined as follows:

ωit = ait X i∈j ait !−1 (c + ρZi,Zj) (3.4)

where c ≥ 1 and ρZi,Zj refers to the correlation between the bank’s Z-score (Zi) and the

consolidated Z-score of the country in which the bank is located (Zj), over the period

(17)

4 Data

The country-level analysis replicates U-H (2009) as closely as possible. Unfortunately, a perfect replication is impossible as a result of data limitations. First, U-H use data from the period between 1997 and 2005, while the current BankScope database, from which the data needed to calculate consolidated Z-scores are taken, only includes data from 1998 onwards. Hence, the replication sample covers the years between 1998 and 2005. Moreover, the ECB statistics do not include the three-bank concentration ratio (CR3) anymore, which rules out its use as a measure of concentration, even though

U-H include the measure as variable as the measure of market concentration in one of their specifications. Third, in some cases the (lack of) completeness of the data from the source used by U-H necessitates the use of a different data source. Finally, there are some obvious mistakes in the data, which are addressed either by ignoring the respective observation, or through extrapolation, but it is not clear whether this procedure was followed by U-H as well. Table A.3 in the data appendix gives an overview of the variables and their definition and data sources, while descriptive statistics of all variables included in the country-level analysis are reported in Table A.1

Since the number of banks included in the BankScope database increases significantly between 2004 and 2005, the representativeness of the database in the period before 2005 can be called into question. For this reason, separate estimations will be performed for the periods 1998 - 2005 and 2005 - 2012, as well as for the entire period between 1998 and 2012. Table A.4 of the appendix gives an overview of the number of banks with which the consolidated Z-scores are calculated for every country and year in the period between 1998 and 2012. The sharp increase in the number of banks in 2005 can clearly be observed.

(18)

analysis thus consists of all banks in the EU-25 that are registered in the BankScope database in the years between 2005 and 2012, but only includes banks of which data are available for at least 4 different years. The reason for this is that the Z-score is based on the standard deviation of the bank’s return in the sample period, of which a reliable vale can only be calculated when enough data is available. The exclusion of all banks with less than 4 observations brings the total number of banks to 2791, of which 1216 are German. Note that data availability is not the only cause of the dataset not being balanced, since some banks have failed during the sample period, while others have merged or have been acquired by competitors. Table A.2 in the appendix reports the descriptive statistics of the bank-specific variables. The table clearly shows the large heterogeneity among banks with respect to their Z-score, which ranges between a minimum of -11 and a maximum of no less than 558. In addition, the correlation between individual banks’ Z-scores and consolidated Z-scores ranges from -0.96 to 1, with a mean of 0.43.

Country-specific means of the CR5 and HHI, as well as their correlations, are shown

in Table A.5. The table clearly shows the large variety in both measures across countries. Moreover, the correlation between the two measures is relatively high overall (0.892), and is above 0.5 for every country, with the exception of Finland (where it is negative, surprisingly). The cause of this negative correlation between the CR5 and the HHI in

Finland remains unclear, although Bikker (2004) has pointed out that the CR5 depends

(19)

5 Model specification & results (country level)

U-H (2009) estimate the following random-effects model to study the effect of banking market concentration on financial soundness:

Zit = αt+ β1cit+ βkx

0

it,k + it (5.1)

where Zitis the consolidated Z-score of the banking sector in country i at time t, while cit

is a measure of concentration and xitis a vector of country-level and (consolidated)

bank-level control variables that are likely to affect the stability of banks. Finally, αtis a

time-specific constant, and it is an error term consisting of a country-specific time-invariant

component and a time-varying component capturing the remaining disturbance. Both components of the error term are assumed to be randomly distributed with mean zero, and to be uncorrelated with the explanatory variables.

As measures of concentration, U-H use the CR5, CR3 and HHI. Real GDP growth

(20)

val-ues of the moral hazard index indicate more moral hazard for banks in the respective country, so that its coefficient is expected to have a negative sign.

The (consolidated) bank-level variables that are included in the model are the bank’s net interest margin, loan loss provisions, and cost-income ratio. The net interest margin is included as a proxy of bank profitability, so that its sign is expected to be positive. The level of loan loss provisions, on the other hand, is assumed to be an indication of credit risk, and is therefore expected to have a negative effect on consolidated Z-scores. The cost-income ratio, finally, is a proxy of inefficiency, and its sign is thus expected to be negative. A more thorough discussion of all control variables and the reasons for their inclusion is provided for by U-H (2009).

Results of replication

The analysis focuses on the base model of U-H (2009), the results of which are shown in Table 5 of their paper. Given the aforementioned data limitations, only specifications 1, 2, 4, 5 and 6 can be estimated. The results of the replication are shown in Table 5.1, in which the specifications are numbered in accordance with those of U-H for ease of comparison. Contrary to the results obtained by U-H, the estimated effect of concentra-tion on consolidated Z-scores is not found to be significant in any of the specificaconcentra-tions, irrespective of whether the CR5 or the HHI is used as the measure of concentration. The

(21)

the estimated coefficients of the model which includes country fixed effects are therefore reported in Table 5.2. As can be seen, the results are the same in a qualitative sense, although it must be noted that the time-invariant moral hazard index drops out in this specification. Again, only the net interest margin and aggregate loan loss provisions are estimated to have a significant effect on consolidated Z-scores, and the coefficients of both variables have their expected sign. The estimated effect of concentration on consolidated Z-scores is not significant in any of the specifications.

Table 5.1: Results of replication of U-H (2009). The numbering of the specifications refers to Table 5 of U-H (2009). Standard errors are clustered by country. Data sources and variables definitions are

described in Table A.3 of the appendix.

Dependent variable: consolidated Z-score (1) (2) (4) (5) (6)

CR5 -0.001 0.009 0.001 -0.126 (0.094) (0.093) (0.093) (0.121) HHI 0.028 (0.269) GDP per capita 0.034 (0.115) Real GDP growth -0.006 0.000 -0.006 0.260 0.318 (0.314) (0.309) (0.307) (0.365) (0.370) ∆ inflation -0.177 -0.169 -0.175 -0.210 -0.133 (0.112) (0.112) (0.110) (0.134) (0.277) ∆ real interest ratet−1 -0.011 -0.014 -0.011 0.058 -0.238

(0.115) (0.109) (0.112) (0.114) (0.239) Credit growtht−1 -0.019 -0.019 -0.019 -0.011 -0.025

(0.017) (0.016) (0.017) (0.015) (0.028) Log(margin) 2.340** 2.372** 2.320** 2.331* (1.112) (1.152) (1.083) (1.263) Loan loss provisions -7.46e-07** -7.53e-07** -7.45e-07** -8.13e-07***

(3.66e-07) (3.74e-07) (3.72e-07) (2.05e-07) Cost-income ratio 0.027 0.029 0.027 0.019

(0.029) (0.030) (0.029) (0.035) Moral hazard index 11.004*** 10.851*** 11.062*** 10.021*** 10.792***

(1.880) (1.959) (1.774) (1.843) (1.836)

Time dummies Yes Yes Yes Yes Yes

Observations 144 144 144 144 122

R2 _0.39 _0.387 _0.383 _0.424 _0.514

No. of countries 23 23 23 23 20

Clustered standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 5.2: Results of replication of U-H (2009) with country fixed effects instead of random effects. The numbering of specifications refers to Table 5 of U-H (2009). Standard errors are clustered by

country. Data sources and variable definitions are described in Table A.3 of the appendix.

CR5 0.100 0.100 0.079 0.158

(0.116) (0.117) (0.080) (0.140)

HHI 0.266

(0.260)

(22)

(0.192)

Real GDP growth 0.004 0.004 -0.003 0.245 0.015 (0.299) (0.299) (0.291) (0.346) (0.296) ∆ inflation -0.184 -0.186* -0.185 -0.229 -0.165

(0.126) (0.106) (0.118) (0.151) (0.131) ∆ real interest ratet−1 -0.000 0.001 0.010 0.081 -0.020

(0.102) (0.110) (0.101) (0.102) (0.120) Credit growtht−2 -0.021 -0.022 -0.020 -0.013 -0.022

(0.019) (0.016) (0.019) (0.018) (0.018) Log(margin) 2.479* 2.482** 2.401* 2.438** (1.201) (1.181) (1.201) (1.071) Loan loss provisions -7.98e-07* -7.98e-07* -7.76e-07** -8.09e-07**

(3.87e-07) (3.90e-07) (3.72e-07) (3.65e-07) Cost-income ratio 0.032 0.032 0.029 0.033

(0.028) (0.028) (0.029) (0.026)

Observations 144 144 144 144 141

R2 _0.343 _0.343 _0.345 _0.170 _0.339

Note from Table A.4, however, that the number of banks included in the database is significantly higher from 2005 onwards, which indicates that in the years prior to 2005, missing data are a serious issue, so that the consolidated Z-scores might not constitute a proper representation of financial stability. Moreover, data about the real interest rate in Malta and Cyprus is not available prior to 2008, which explains why the number of countries reported in Tables 5.1 and 5.2 is only 23. Because of this, it is interesting to utilize more recent data in order to find out whether the lack of significant results obtained in our replication is simply a reflection of data limitations. Hence, two additional ‘replications’ with more recent data are performed. In the first, consolidated Z-scores from the period between 2005 and 2012 are used, while in the second, all years from 1998 until 2012 are included in the sample. Note that with an increase in the sample period, all Z-scores are affected, as the standard deviation with which the consolidated Z-scores are calculated depends on the sample period itself. Both estimations include country fixed effects, and the results are shown in Tables 5.3 and 5.4.

In both cases, the coefficient of the CR5 remains insignificant (with the exception of

(23)

the HHI and consolidated Z-scores is robust against the omission of Finland from the sample, as this is the only country where the CR5 and the HHI are negatively correlated.

Tables 5.5 and 5.6 illustrate that the estimated effect is not robust: when Finland is left out of the sample, neither the CR5 nor the HHI have significant coefficients. With

respect to the control variables, the results are comparable with the base model. Again, the net interest margin and aggregate loan loss provisions are the only variables which are found to significantly affect consolidated Z-scores, and their signs are as expected. For both variables, the estimated association is robust against the omission of Finland from the sample.

Table 5.3: Results of replication of U-H (2009) with more recent data (between 2005 and 2012 and country fixed effects instead of random effects. The numbering of specifications refers to Table 5 of U-H (2009). Standard errors are clustered by country. Data sources and variables definitions are

described in Table A.3 of the appendix.

CR5 -0.032 -0.036 -0.094 -0.035 (0.067) (0.066) (0.104) (0.077) HHI -0.344** (0.154) GDP per capita -0.039 (0.090) Real GDP growth -0.034 -0.023 -0.038 -0.029 -0.035 (0.160) (0.167) (0.153) (0.184) (0.149) ∆ inflation 0.130 0.117 0.136 0.116 0.130 (0.108) (0.120) (0.113) (0.133) (0.101) ∆ real interest ratet−1 -0.052 -0.049 -0.062 -0.094* -0.052

(0.044) (0.046) (0.042) (0.050) (0.043) Credit growtht−2 -0.016 -0.016 -0.000 0.019 -0.016

(0.054) (0.054) (0.050) (0.059) (0.051) Log(margin) 5.742** 5.872* 5.282** 5.739** (2.776) (2.936) (1.946) (2.614) Loan loss provisions -5.21e-08** -5.47e-08*** -4.89e-08** -5.19e-08***

(1.90e-08) (1.58e-08) (1.91e-08) (1.83e-08) Cost-income ratio -0.047 -0.047 -0.036 -0.047

(0.049) (0.049) (0.049) (0.046)

Observations 181 181 181 181 180

R2 0.317 0.318 0.356 0.136 0.317

Table 5.4: Results of replication of U-H (2009) with more recent data (between 1998 and 2012) and country fixed effects instead of random effects. The numbering of specifications refers to Table 5 of

U-H (2009). Standard errors are clustered by country. Data sources and variables definitions are described in Table A.3 of the appendix.

(24)

CR5 -0.164 -0.185 -0.110 -0.221* (0.126) (0.134) (0.109) (0.133) HHI -0.492*** (0.141) GDP per capita -0.190 (0.115) Real GDP growth 0.035 0.098 0.082 0.235** 0.023 (0.113) (0.090) (0.082) (0.091) (0.109) ∆ inflation 0.013 -0.056 0.008 -0.147 0.007 (0.053) (0.064) (0.045) (0.120) (0.056) ∆ real interest ratet−1 -0.040 -0.018 -0.036 0.037 -0.041

(0.049) (0.047) (0.049) (0.059) (0.052) Credit growtht−2 -0.057 -0.060* -0.050* -0.016 -0.056*

(0.035) (0.035) (0.027) (0.026) (0.032) Log(margin) 9.053*** 9.246*** 8.912*** 9.172***

(2.580) (2.609) (1.995) (2.424) Loan loss provisions -7.65e-08*** -8.39e-08*** -7.96e-08*** -7.25e-08***

(2.13e-08) (2.25e-08) (1.82e-08) (2.14e-08)

Cost-income ratio 0.106 0.107 0.104 0.108

(0.098) (0.095) (0.084) (0.091) Time dummies

Yes Yes Yes Yes Yes

Observations 303 303 303 303 300

R2 0.371 0.386 0.400 0.110 0.369

Table 5.5: Results of replication of U-H (2009) with more recent data (between 2005 and 2012 and country fixed effects instead of random effect, when Finland is excluded from the sample. The numbering of specifications refers to Table 5 of U-H (2009). Standard errors are clustered by country.

Data sources and variables definitions are described in Table A.3 of the appendix.

CR5 -0.046 -0.043 -0.108 -0.057 (0.069) (0.068) (0.106) (0.082) HHI -0.126 (0.097) GDP per capita 0.032 (0.072) Real GDP growth -0.014 -0.024 -0.013 -0.015 -0.017 (0.163) (0.172) (0.155) (0.186) (0.151) ∆ inflation 0.131 0.143 0.135 0.110 0.131 (0.117) (0.130) (0.117) (0.135) (0.109) ∆ real interest ratet−1 -0.071 -0.074 -0.070 -0.091* -0.073

(0.050) (0.052) (0.046) (0.051) (0.048) Credit growtht−2 0.002 0.003 0.004 0.024 0.003

(0.052) (0.052) (0.049) (0.060) (0.049) Log(margin) 2.490** 2.346** 2.677*** 2.472***

(1.81e-08) (1.63e-08) (1.96e-08) (1.75e-08) Cost-income ratio -0.078* -0.079 -0.073 -0.078*

(0.045) (0.047) (0.046) (0.042)

Observations 173 173 173 173 172

R2 0.329 0.329 0.330 0.181 0.328

(25)

Table 5.6: Results of replication of U-H (2009) with more recent data (between 1998 and 2012) and country fixed effects instead of random effect, when Finland is excluded from the sample. The numbering of specifications refers to Table 5 of U-H (2009). Standard errors are clustered by country.

Data sources and variables definitions are described in Table A.3 of the appendix.

CR5 -0.012 -0.019 -0.034 -0.052 (0.041) (0.040) (0.072) (0.059) HHI -0.132 (0.079) GDP per capita -0.061 (0.055) Real GDP growth 0.161** 0.181** 0.162** 0.233** 0.143** (0.070) (0.072) (0.066) (0.083) (0.067) ∆ inflation 0.015 -0.008 0.013 -0.041 0.020 (0.034) (0.035) (0.034) (0.038) (0.037) ∆ real interest ratet−1 -0.001 0.007 -0.001 0.012 -0.008

(0.027) (0.026) (0.029) (0.035) (0.029) Credit growtht−2 -0.017 -0.019 -0.017 0.005 -0.018

(0.014) (0.013) (0.013) (0.017) (0.014) Log(margin) 5.649*** 5.753*** 5.851*** 5.733***

(1.29e-08) (1.41e-08) (1.32e-08) (1.28e-08) Cost-income ratio -0.023 -0.021 -0.018 -0.021

(0.017) (0.016) (0.016) (0.016) Time dummies

Yes Yes Yes Yes Yes

Observations 288 288 288 288 285

R2 _0.493 _0.498 _0.499 _0.161 _0.489

In summary, the results of U-H (2009) are not replicable when an (almost) identical sample is used. Where U-H find a negative, significant and robust association between banking market concentration and consolidated Z-scores, the replication of their analysis performed here does not indicate a significant relationship. Moreover, even though the use of more recent data gives results which are more in line with those of U-H when the HHI is used as the measure of banking market concentration, these are not robust against the exclusion of Finland from the sample. Hence, the replication performed in this paper puts some doubt on the reliability of their findings.

Some extensions

(26)

The reason for is that the level of loan loss provisions in itself does not provide much information about the riskiness of a bank’s assets, given that larger banks will naturally have higher loan loss provisions than relatively small banks with the same level of credit risk. Moreover, the ratio of loan loss provisions to total assets is a reflection of both a bank’s ratio of non-performing loans to total loans and its ratio of total loans to total assets, which are often-used variables in the empirical literature on this subject (Boyd et al., 2006; Berger et al., 2009). Second, real GDP per capita is included as the measure of a country’s economic development, instead of nominal GDP per capita. Third, with respect to the real interest rate and inflation, levels are used instead of changes, as this is more in line with the majority of the literature on the subject. Fourth, since the net interest margin is already defined in percentage terms, it is included without a logarithmic transformation. Finally, reverse causality issues are addressed by performing IV regressions in which lagged values of the concentration measures are included as instruments of their contemporaneous values. As was explained earlier, an association between banking market concentration and financial stability might simply result from an increase in market concentration following a bank failure. By using lagged values of the concentration measurements as instruments of contemporaneous values, this effect is ruled out so that the true causal effect running from market concentration to consolidated Z-scores can be estimated.

Since the database is far more complete from 2005 onwards, only data from the years between 2005 and 2012 are used to estimate the improved model. Moreover, Finland is dropped from the sample for reasons mentioned earlier in this section. Table 5.7 presents the OLS results, while Table 5.8 gives the results of the IV regressions in which the concentration indices are instrumented by their lagged values. In both tables, the first three specifications show the results when the CR5 is used as the measure

(27)

through 6.

The results again fail to provide robust evidence of an association between bank-ing market concentration on consolidated Z-scores. Both the OLS and IV results only indicate a significant (negative) relationship in specification 5, where the HHI is used as the measure of concentration, while all bank-level control variables are omitted from the model. The significance of the estimate is thus not robust against either the use of the CR5 as the measure of concentration or the inclusion of (consolidated) bank-level

control variables. The only other variable found to be significant loan loss provisions to assets, which has its expected negative sign, and is significant in all specifications. Furthermore, the cost-income ratio is weakly significant in some specifications, having its expected sign in these cases. Note that the F-statistic in the first stage of the IV estimations ranges between 10.86 and 128.7, indicating that the lagged values of the concentration measures can be considered relevant instruments of the contemporaneous values.

Overall, then, a country-level analysis does not offer clear results regarding the rela-tionship between banking market concentration and financial stability. Note, however, that these inconclusive results might be caused by the relatively small number of ob-servations. Moreover, a country-level analysis does not take into account bank-specific information or possible effect of financial contagion. In the next section, therefore, a bank-level analysis will be performed in an attempt to shed more light in this issue.

Table 5.7: Results of extended country-level analysis (OLS regression). Sample consists of the EU-25 without Finland, in the period between 2005 and 2012. Standard errors are clustered by country. Data

sources and variables definitions are described in Table A.3 of the appendix.

Dependent variable: consolidated Z-score (1) (2) (3) (4) (5) (6)

CR5 -0.038 -0.095 -0.048

(0.068) (0.091) (0.073)

HHI -0.139 -0.297** -0.130

(0.098) (0.137) (0.089) Loan loss provisions / assets -0.853*** -1.741*** -0.845*** -1.703***

(0.225) (0.568) (0.223) (0.586) Net interest margin 0.772 0.680 0.897 0.774

(0.623) (0.566) (0.626) (0.574) Cost-income ratio -0.074* -0.056 -0.068 -0.052

(0.041) (0.037) (0.042) (0.037) Real GDP per capita 0.209 0.004 0.231 0.018

(28)

Real GDP growth -0.009 -0.197 -0.020 -0.195 (0.162) (0.187) (0.160) (0.187)

Inflation 0.279 0.360 0.272 0.347

(0.238) (0.278) (0.228) (0.274) Real interest ratet−1 0.010 0.373 -0.019 0.351

(0.180) (0.242) (0.172) (0.237)

Credit growtht−2 0.003 0.002 0.009 0.003

(0.043) (0.039) (0.043) (0.037)

Time dummies Yes Yes Yes Yes Yes Yes

Observations 192 179 179 192 179 179

R2 _0.303 _0.200 _0.341 _0.306 _0.214 _0.342

No. of countries 24 24 24 24 24 24

Table 5.8: Results of extended country-level analysis (IV regression with CR5 and HHI instrumented

by their lagged values). Sample consists of the EU-25 without Finland, in the period between 2005 and 2012. Standard errors are clustered by country. Data sources and variables definitions are described in

Table A.3 of the appendix.

Dependent variable: consolidated Z-score (1) (2) (3) (4) (5) (6)

CR5 -0.011 -0.161 -0.054

(0.089) (0.120) (0.091)

HHI -0.156 -0.382** -0.199*

(0.215) (0.532) (0.210) (0.547) Net interest margin 0.771 0.675 0.913 0.802

(0.599) (0.527) (0.606) (0.532) Cost-income ratio -0.074* -0.056 -0.067* -0.049

(0.039) (0.035) (0.039) (0.035) Real GDP per capita 0.198 0.004 0.233 0.023

(0.220) (0.160) (0.232) (0.168) Real GDP growth -0.015 -0.197 -0.025 -0.195

(0.153) (0.175) (0.151) (0.174)

Inflation 0.294 0.361 0.277 0.347

(0.227) (0.259) (0.215) (0.254) Real interest ratet−1 0.025 0.374* -0.021 0.342

(0.176) (0.226) (0.163) (0.220)

Credit growtht−2 0.007 0.002 0.012 0.005

(0.040) (0.036) (0.041) (0.035)

Observations 192 179 179 192 179 179

R2 _0.302 _0.194 _0.341 _0.306 _0.212 _0.341

No. of countries 24 24 24 24 24 24

6 Model specification & results (bank level)

(29)

the estimation of the following model will be presented in this section: Zit = αit+ γj+ β1cjt+ βbx 0 it,b+ βcx 0 jt,c+ it (6.1)

where Zit refers to the Z-score of bank i at time t, while αit is a bank-specific constant

consisting of a time-invariant bank-specific component and a time-varying component which is equal for all banks. Moreover, γj is a country-specific constant where j refers to

the country in which the bank is located. The coefficient of interest is β1, which measures

the effect of banking market concentration on individual bank’s Z-scores. The vectors x0_it,b and x0_jt,c refer to a range of control variables at the bank level and the country level, respectively, and it is an error term with mean zero. Note that in the base model in

which the variables are not transformed, the country-specific constant drops out, as it is perfectly correlated with the time-invariant component of the bank-specific constant. Given the time-varying nature of the weights with which the variables are transformed in the later part of the analysis, however, the term does not drop out in the transformed model. The control variables are the same as those used in the last specification of the country-level analysis, although the bank-level variables are obviously not consolidated, which was the case in the country-specific model. Again, both OLS and IV regressions are performed, where in the latter, lagged values of the concentration measures are used as instruments of the contemporaneous values.

(30)

the sample, the sign of estimates of β1 becomes negative, and the estimates are again

significant at the 1% level. Hence, the results clearly indicate the need to weight the observations in order to obtain robust and meaningful results from a bank-level analysis.

Table 6.1: Results of bank-specific base model. Sample of all banks from the EU-25 with the exception of Finland, in the period between 2005 and 2012. Standard errors are clustered by bank. Data sources

and variables definitions are described in Table A.3 of the appendix.

Dependent variable: Z-score (1) (2) (3) (4) (5) (6) CR5 0.341*** 0.308*** 0.310***

(0.029) (0.030) (0.030)

HHI 1.226*** 1.069*** 1.104***

(0.458) (0.394) (0.482) (0.397) Net interest margin 0.658*** 0.633*** 0.677*** 0.649***

(0.235) (0.223) (0.233) (0.220) Cost-income ratio -0.002 -0.003 -0.004 -0.005

(0.006) (0.006) (0.006) (0.006) Real GDP per capita 1.036*** 0.915*** 1.078*** 0.952***

(0.102) (0.101) (0.103) (0.101) Real GDP growth 0.178*** -0.270*** 0.234*** -0.220***

(0.061) (0.067) (0.064) (0.068) Inflation 0.635*** 0.208*** 0.641*** 0.203***

(0.069) (0.075) (0.070) (0.075) Real interest ratet−1 -0.103 -0.192*** -0.153** -0.240***

(0.078) (0.072) (0.076) (0.071) Credit growtht−2 -0.027 -0.003 -0.021 0.002

(0.019) (0.018) (0.019) (0.019)

Observations 21,508 21,486 21,358 21,508 21,486 21,358

R2 0.110 0.106 0.120 0.105 0.102 0.117

Number of banks 2,783 2,786 2,783 2,783 2,786 2,783 Clustered standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

(31)

As was the case with the country-level analysis, the results do not provide convincing evidence of an association between banking market concentration and financial stability. The coefficients related to the CR5 and HHI are insignificant in all specifications except

the second, where it is only weakly significant (with a negative sign). The variables found to significantly affect financial stability are the three bank-specific variables (which all have their expected sign), and, juin some specifications, the lagged real interest rate (with a negative sign).

Table 6.2: OLS results of bank-specific model when the observations are weighted according to the size of the bank’s assets as shown in Equation (3.3). Sample consists of all banks from the EU-25 with the

exception of Finland, in the period between 2005 and 2012. Standard errors are clustered by bank. Data sources and variables definitions are described in Table A.3 of the appendix. Constant included

but not reported.

Dependent variable: Z-score (1) (2) (3) (4) (5) (6)

CR5 -0.098 -0.168* -0.118

(0.081) (0.089) (0.084)

HHI -0.099 -0.139 -0.044

(0.500) (0.480) (0.445) (0.468) Cost-income ratio -0.024** -0.018** -0.023** -0.019**

(0.011) (0.009) (0.011) (0.009) Real GDP per capita 0.033 -0.036 0.061 -0.029

(0.133) (0.115) (0.127) (0.108)

Real GDP growth 0.089 -0.141* 0.098 -0.141

(0.065) (0.079) (0.061) (0.089)

Inflation 0.141 -0.087 0.162* -0.084

(0.103) (0.066) (0.097) (0.065) Real interest ratet−1 -0.183*** -0.066* -0.172*** -0.053

(0.067) (0.035) (0.065) (0.033) Credit growtht−2 0.005 -0.005 0.002 -0.010

(0.022) (0.022) (0.022) (0.023)

Country dummies Yes Yes Yes Yes Yes Yes

Observations 21,508 21,486 21,358 21,508 21,486 21,358

R2 _0.498 _0.436 _0.512 _0.496 _0.429 _0.508

*** p<0.01, ** p<0.05, * p<0.1

Table 6.3: IV results of bank-specific model when the observations are weighted according to the size of the bank’s assets as shown in Equation (3.3). Concentration measures are instrumented by their lagged values. Sample consists of all banks from the EU-25 with the exception of Finland, in the period

between 2005 and 2012. Standard errors are clustered by bank. Data sources and variables definitions are described in Table A.3 of the appendix. Constant included but not reported.

(32)

(0.112) (0.107) (0.111)

HHI 0.005 -0.053 0.043

(0.498) (0.482) (0.402) (0.446) Cost-income ratio -0.024** -0.018** -0.025** -0.020**

(0.011) (0.009) (0.011) (0.010) Real GDP per capita 0.033 -0.036 0.045 -0.045

(0.132) (0.115) (0.124) (0.102)

Real GDP growth 0.088 -0.143* 0.099 -0.154

(0.064) (0.080) (0.062) (0.095)

Inflation 0.140 -0.088 0.150 -0.096

(0.065) (0.035) (0.063) (0.034) Credit growtht−2 0.006 -0.006 -0.003 -0.013

(0.021) (0.021) (0.023) (0.024)

Observations 21,507 21,486 21,357 21,507 21,486 21,357

R2 _0.497 _0.436 _0.512 _0.495 _0.428 _0.507

*** p<0.01, ** p<0.05, * p<0.1

(33)

The results are surprisingly similar to those obtained when the correlation of Z-scores is ignored: again, only the coefficients related to the bank-level variables are robustly significant (and have their expected sign). The coefficient related to the CR5

is (negatively) significant in the second specification at the 5% level, but the signifance falls to the 10% level when the CR5 is instrumented by its own lagged value to control

for reverse causality. The results do not appear to depend on the exact manner in which the observations are weighted, as they are approximately equal when c is set to either 1, 2, 3 or 5, as shown in Tables B.3 through B.8 in the appendix.

Table 6.6, finally, gives an overview of the findings of this paper and compares them with those of U-H. Overall, it must be concluded that, the analyses performed do not offer evidence of a relationship between banking market concentration and financial stability, and thus do not support either the concentration-stability or the concentration-fragility view.

Table 6.4: OLS results of bank-specific model when the observations are weighted according to Equation (3.4) with c=2. Sample consists of all banks from the EU-25 with the exception of Finland,

in the period between 2005 and 2012. Standard errors are clustered by bank. Data sources and variables definitions are described in Table A.3 of the appendix. Constant included but not reported.

CR5 -0.089 -0.164** -0.109

(0.078) (0.083) (0.080)

HHI -0.144 -0.184* -0.088

(0.502) (0.467) (0.439) (0.458) Cost-income ratio -0.023** -0.016** -0.021** -0.016* (0.011) (0.008) (0.010) (0.008) Real GDP per capita 0.034 -0.047 0.073 -0.033

(0.131) (0.107) (0.125) (0.103) Real GDP growth 0.108* -0.138* 0.115* -0.130

(0.063) (0.078) (0.060) (0.086)

Inflation 0.112 -0.111* 0.140 -0.105*

(0.067) (0.035) (0.065) (0.033) Credit growtht−2 0.003 -0.002 0.003 -0.005

(0.021) (0.021) (0.021) (0.022)

Observations 21,508 21,486 21,358 21,508 21,486 21,358

R2 _0.456 _0.393 _0.471 _0.456 _0.389 _0.468

(34)

Table 6.5: IV results of bank-specific model when the observations are weighted according to Equation (3.4) with c=2. Concentration measures are instrumented by their lagged values. Sample consists of

all banks from the EU-25 with the exception of Finland, in the period between 2005 and 2012. Standard errors are clustered by bank. Data sources and variables definitions are described in Table

A.3 of the appendix. Constant included but not reported.

CR5 -0.037 -0.186* -0.089

(0.113) (0.105) (0.110)

HHI -0.082 -0.132 -0.039

(0.503) (0.470) (0.398) (0.437) Cost-income ratio -0.024** -0.017** -0.023** -0.017* (0.011) (0.008) (0.011) (0.009) Real GDP per capita 0.035 -0.048 0.061 -0.044

(0.130) (0.106) (0.122) (0.099) Real GDP growth 0.107* -0.139* 0.115* -0.138

(0.063) (0.079) (0.059) (0.091)

Inflation 0.112 -0.113* 0.131 -0.113*

(0.106) (0.063) (0.097) (0.063) Real interest ratet−1 -0.185*** -0.059 -0.172*** -0.048

(0.066) (0.036) (0.064) (0.033) Credit growtht−2 0.005 -0.003 -0.000 -0.007

(0.020) (0.020) (0.022) (0.023)

Observations 21,507 21,486 21,357 21,507 21,486 21,357

R2 _0.455 _0.393 _0.471 _0.456 _0.388 _0.467

*** p<0.01, ** p<0.05, * p<0.1

Table 6.6: Comparison of the results of the country-level and bank-level analyses with those obtained by Uhde and Heimeshoff (2009).

Country level Bank level U-H (2009) Variable Significant? Sign Significant? Sign Significant? Sign CR5 No Sometimes Negative Yes Negative

HHI Sometimes Negative No Yes Negative

Loan loss provisions Yes Negative Yes Negative Yes Negative Net interest margin Yes Positive Yes Positive No

Cost-income ratio No Yes Negative Sometimes Negative

GDP per capita No No Yes Positive

Real GDP growth No No No

Inflation No No No

Real interest ratet−1 No Sometimes Negative No

Credit growtht−2 No No Yes Positive

(35)

7 Conclusion

This paper has investigated the association between banking market concentration and financial stability at the macroeconomic level, through separate analyses at the country level, the bank level and an approach which combines elements of both. The investigation thus incorporates the idea that financial stability depends on both country-level and bank-level characteristics.

First, a country-level analysis was performed in which consolidated Z-scores were used as the measure of financial stability. The starting point of the analysis was a repli-cation of Uhde and Heimeshoff (U-H) (2009), who analyzed the EU-25 in the period between 1997 and 2005 and obtained strong evidence of a negative relationship between banking market concentration and consolidated Z-scores. Surprisingly, however, their results could not be replicated, as no significant relationship between banking market concentration and consolidated Z-scores was found even though the modelling and sam-ple selection decisions of U-H were followed as closely as possible. The only variables which are found to be significantly associated with consolidated Z-scores are aggregate loan loss provisions, which were included in the model as a proxy of credit risk, and the net interest rate margin, included as a proxy of bank profitability, with the coefficients of both variables having their expected sign. Similar results are obtained when country fixed effects are included in the model, as well as with a more recent sample and with the inclusion of better defined control variables.

(36)

macroeconomic financial stability depends on (1) probabilities of default of individual banks, (2) loss given default of individual banks, and (3) the correlation of defaults across banks. By performing bank-level analyses in which the observations are weighted according to the size of the respective banks assets, and additionally, according to both the size of assets and the correlation between individual scores and consolidated Z-scores, all of these elements of financial stability have been taken into account in this paper.

The results again do not provide any evidence of an association between banking market concentration and stability, as the only variables for which a robust significant association with financial stability is found are the ratio of loan loss provisions to total assets (a proxy of credit risk), the net interest margin (a proxy of profitability), the cost-income ratio (a proxy of inefficiency), and, to some extent, the real interest rate. Hence, the analyses performed in this paper do not offer support for either the concentration-stability view or the concentration-fragility view.

(37)

References

[1] Allen, Franklin and Douglas Gale. (2000). Financial Contagion. Journal of Political Economy, 108: 1-33.

[2] Allen, Franklin and Douglas Gale. (2004). Competition and Financial Stability. Journal of Money, Credit and Banking, 36: 453-480.

[3] Baumol, William. (1982). Contestable Markets: an Uprising in the Theory of In-dustry Structure. American Economic Review, 72: 1-15.

[4] Beck, Thorsten, Asli Demirg¨u¸c-Kunt and Ross Levine. (2006). Bank concentration, competition, and crises: First results. Journal of Banking and Finance, 30: 1581-1603.

[5] Berger, Allen. (1995). The profit-structure relationship in banking: tests of market power and efficient-structure hypotheses. Journal of Money, Credit and Banking, 27: 404-431.

[6] Berger, Allen. (2007). International Comparisons of Banking Efficiency. Financial Markets, Institutions and Instruments, 16: 119-144.

[7] Berger, Allen, Leora Klapper and Rima Turk-Ariss. (2009). Bank Competition and Financial Stability. Journal of Financial Services Research, 35: 99-118.

[8] Bikker, Jakob, Sherrill Shaffer and Laura Spierdijk. (2012). Assessing Competition with the Panzar-Rosse Model: the Role of Scale, Costs and Equilibrium. The Review of Economics and Statistics, 94: 1025-1044.

[9] Boyd, John and David Runkle. (1993). Size and performance of banking firms. Journal of Monetary Economics, 31: 47-67.

(38)

[11] Chan, Yuk-Shee, Stuart Greenbaum and Anjan Thakor. (1986). Information reusability, competition and bank asset quality. Journal of Banking and Finance, 10: 243-253.

[12] ˘Cih´ak, Martin. (2007). Systemic Loss, A Measure of Financial Stability. Czech Jour-nal of Economics and Finance, 57: 5-26.

[13] Corvoisier, Sandrine and Reint Gropp. (2002). Bank concentration and retail inter-est rates. Journal of Banking and Finance, 26: 2155-2189.

[14] Dam, Lammertjan and Michael Koetter. (2012). Bank Bailouts and Moral Hazard: Evidence from Germany. The Review of Financial Studies, 25: 2343-2380.

[15] Demirg¨u¸c-Kunt, Asli and Enrica Detragiache. (2002). Does deposit insurance in-crease banking system stability? An empirical investigation. Journal of Monetary Economics, 49: 1373-1406.

[16] Demsetz, Harold. (1973). Industry Structure, Market Rivalry and Public Policy. Journal of Law and Economics, 16: 1-10.

[17] Demsetz, Rebecca and Philip Strahan. (1997). Diversification, Size and Risk at Bank Holding Companies. Journal of Money, Credit and Banking, 29: 300-313. [18] De Nicol´o, Gianni. (2000). Size, charter value and risk taking in banking: An

international perspective. Federal Reserve System International Finance Discussion Papers, 689.

[19] De Nicol´o, Gianni and Myron Kwast. (2002). Systemic risk and financial consoli-dation: Are they related? Journal of Banking and Finance, 26: 861-880.

(39)

[21] Group of Ten. (2001). Report on Consolidation in the Financial Sector. Bank of International Settlements. http://www.bis.org/publ/gten05.pdf.

[22] Hellman, Thomas, Kevin Murdock and Joseph Stiglitz. (2000). Liberalization, Moral Hazard in Banking, and Prudential Regulation: Are Capital Requirements Enough? American Economic Review, 90: 147-165.

[23] Keely, Michael. (1990). Deposit Insurance, Risk and Market Power in Banking. American Economic Review, 80: 1183-1200.

[24] Leitner, Yaron. (2005). Financial Networks: Contagion, Commitment, and Private Sector Bailouts. Journal of Finance, 60: 2925-2953.

[25] Marcus, Alan. (1984). Deregulation and Bank Financial Policy. Journal of Banking and Finance, 8: 557-565.

[26] Martins, Bruno Silva and Leonardo Alencar. (2009). Banking concentration, prof-itability and systemic risk: An indirect contagion approach. Central Bank of Brazil Working Paper.

[27] Mirzaei, Ali, Tomoe Moore and Guy Liu. (2013) Does market structur matter on banks’ profitability and stability? Emerging vs. advanced economies. Journal of Banking and Finance, 37: 2920-2937.

[28] Mishkin, Frederic. (1999). Financial consolidation: Dangers and opportunities. Journal of Banking and Finance, 23: 675-691.

[29] Mishkin, Frederic. (2005). How big is too big to fail? NBER Working Paper, 11814. http://www.nber.org/papers/w11814.

[30] Molyneux, Phil and William Forbes. (1995). Market structure and performance in banking. Applied Economics, 27: 155-159.

(40)

[32] Panzar, John and James Rosse. (1987). Testing for “Monopoly” Equilibrium. Jour-nal of Industrial Economics, 35: 443-456.

[33] Peltzman, Sam. (1977). The Gains and Losses from Industrial Concentration. The Journal of Law and Economics, 20: 229-263.

[34] Ramakrishnan, Ram and Anjan Thakor. (1984). Information reliability and a theory of financial intermediation. Review of Economic Studies, 51: 415-432.

[35] S´aez, Lawrence and Xianwen Shi. (2004). Liquidity Pools, Risk Sharing and Finan-cial Contagion. Journal of FinanFinan-cial Services Research, 25: 5-23.

[36] Schaek, Klaus, Martin ˘Cih´ak and Simon Wolfe. (2009). Are Competitive Banking Systems More Stable? Journal of Money, Credit and Banking, 41: 711-734.

[37] Shaffer, Sherrill. (1989). Pooling intensifies joint failure risk. Federal Reserve Bank of Philadelphia Working Paper, 89-1.

[38] Smirlock, Michael. (1985). Evidence on the (Non)-Relationship between Concentra-tion and Profitability in Banking. Journal of Money, Credit and Banking, 17:69-83. [39] Uhde, Andr´e and Ulrich Heimeshoff. (2009). Consolidation in banking and financial stability in Europe: Empirical evidence. Journal of Banking and Finance, 33: 1299-1311.

[40] Vives, Xavier. (2010). Competition and Stability in Banking. CESifo Working Pa-per, 3050.

[41] Vries, de, Gaaitzen. (2005). The simple economics of bank fragility. Journal of Banking and Finance, 29: 803-825.

(41)

(42)

A

Data appendix

Table A.1: Descriptive statistics of country-level variables and consolidated bank-level variables.

Variable N Mean SD Min Max

Z-score(1998-2005) 200 21.77 18.19 -1.44 112.88 Z-score(2005-2012) 200 14.94 16.41 -2.28 102.20 Z-score(1998-2012) 375 11.98 7.95 -2.77 81.77 CI5 (%) 372 59.19 18.84 18.95 99.36 HHI (%) 372 11.23 7.89 1.33 40.67 Inflation (%) 375 2.77 2.59 -4.64 20.30 ∆ inflation (%) 375 -0.19 2.31 -15.88 10.42

GDP per capita (th. US$) 375 26.83 18.45 2.75 114.21

Real GDP per capita (th. US$) 375 20.51 14.78 2.14 80.69

Real GDP growth (%) 372 2.45 3.57 -17.96 12.23

Credit growth (%) 360 4.51 11.74 -20.12 151.80

Real interest rate (%) 336 1.05 2.87 -9.36 18.44

∆ real interest rate (%) 325 -0.28 2.75 -12.79 20.90

Moral hazard index 375 0.89 1.00 -0.46 3.27

Net interest margin (%) 375 2.54 1.35 0.13 6.68

Cost-income ratio (%) 375 62.94 22.74 26.06 380.34

Loan loss provisions (th. US$) 375 5019068 1.35E+07 -1119433 1.42E+08

Loan loss provisions / assets (%) 375 0.50 0.87 -4.50 6.36

Table A.2: statistics of (non-consolidated) bank-level variables.

Variable N Mean SD Min Max

Z-score(2005-2012) 21706 49.50 47.24 -11.11 558.30

Net interest margin (%) 21668 2.59 1.57 -25.55 80.00

Cost-income ratio (%) 21588 67.96 25.76 0.00 950.00

Loan loss provisions / assets (%) 22360 0.36 0.43 -0.67 6.36

(43)

(44)

(45)

Table A.5: Mean of CR5, HHI and their correlation, per country, in the period between 1998 and 2012

Country Mean CR5 Mean HHI Corr.CR5,HHI

(46)

B

Robustness checks

Table B.1: Results of bank-specific base model. Sample of EU-25 without Finland and Germany, between 2005 and 2012. Standard errors are clustered by bank. Data sources and variables definitions

are described in Table A.1.

Dependent variable: Z-score (1) (2) (3) (4) (5) (6) CR5 -0.162*** -0.189*** -0.175***

(0.028) (0.032) (0.030)

HHI -0.354*** -0.381*** -0.351***

(0.117) (0.128) (0.125) Loan loss provisions / assets -0.381** -1.539*** -0.550*** -1.736***

(0.173) (0.285) (0.178) (0.298) Net interest margin 0.297* 0.264* 0.311* 0.279*

(0.157) (0.152) (0.163) (0.157)

Cost-income ratio 0.003 0.003 0.004 0.004

(0.006) (0.007) (0.006) (0.007) Real GDP per capita -0.243** -0.248** -0.137 -0.152

(0.100) (0.097) (0.096) (0.094) Real GDP growth -0.122** -0.349*** -0.128** -0.385***

(0.059) (0.078) (0.059) (0.079)

Inflation 0.045 -0.043 0.120* 0.011

(0.075) (0.074) (0.072) (0.073) Real interest ratet−1 -0.053 -0.094 -0.045 -0.091

(0.073) (0.064) (0.073) (0.064) Credit growtht−2 -0.095*** -0.086*** -0.103*** -0.094***

(0.018) (0.018) (0.019) (0.018)

Observations 11,923 11,898 11,773 11,923 11,898 11,773

R2 _0.015 _0.015 _0.019 _0.012 _0.012 _0.016

∗ ∗ ∗p < 0.01, ∗ ∗ p < 0.05, ∗p < 0.1

(47)

Table B.2: OLS results of bank-specific model when the observations are weighted according to the size of the banks asset as shown in Equation (3.3). Sample consists of all banks in the EU-25 with the exception of Finland and Germany, between 2005 and 2012. Data sources and variables definitions are

described in Table A.1 of the appendix. Constant included but not reported.

CR5 -0.128 -0.207** -0.152*

(0.079) (0.086) (0.081)

HHI -0.108 -0.145 -0.054

(0.500) (0.477) (0.449) (0.466) Cost-income ratio -0.021** -0.015* -0.021** -0.016* (0.010) (0.008) (0.010) (0.009) Real GDP per capita 0.030 -0.033 0.074 -0.014

(0.134) (0.119) (0.128) (0.110)

Real GDP growth 0.098 -0.129 0.109* -0.129

(0.066) (0.080) (0.061) (0.089)

Inflation 0.120 -0.098 0.148 -0.090

(0.102) (0.065) (0.097) (0.064) Real interest ratet−1 -0.190*** -0.072** -0.180*** -0.060*

(0.067) (0.034) (0.066) (0.033) Credit growtht−2 -0.000 -0.010 -0.003 -0.015

(0.022) (0.021) (0.022) (0.022)

Observations 11,923 11,898 11,773 11,923 11,898 11,773

R2 _0.520 _0.459 _0.539 _0.516 _0.448 _0.532