Re-examining the tradeoff between bank competition and bank financial stability in Europe

1

R

E

-

EXAMINING THE TRADEOFF BETWEEN BANK COMPETITION AND

BANK FINANCIAL STABILITY IN EUROPE

Lilia Marinkova

University of Amsterdam

Amsterdam, The Netherlands

Master’s Thesis

2

Table of Contents

1.

Introduction..

………

3

2.

Literature Review

..………..

5

2.1 Theoretical Literature

……….

5

2.2 Empirical Literature

………..

5

3.

The Empirical Model………..………. 9

3.1 Bank Competition Indices………. 11

3.2 Bank Financial Stability Indicators………. 16

3.3 Other Variables……… 16

3.4 Estimation Model……… 17

3.5 Data……….. 20

4.

The Empirical Investigation ……….… 23

4.1 Lerner Index Results ……….. 23

4.2 Boone Indicator Results………. 24

4.3 Empirical Results………. 24

5. Conclusion……… 37

Bibliography……… 38

3

1. Introduction

The trade-off between bank competition and financial stability has always been a controversial issue and a subject of active debate among academics and policymakers. However, the topic is receiving increased interest since the beginning of the global financial crisis. It is being questioned whether the negative effects of bank competition and the consequent financial innovations were the force accountable for the global financial turmoil. Although greater innovation and efficiency have been the result of increased banking competition, there is still lack of theoretical and empirical consensus on the sign of the effect of bank competition on financial stability.

Two opposing views are at the center of the debate on the trade-off between competition and stability (Keeley, 1990; Boyd and De Nicolò, 2005). On the one hand is the competition-fragility view (Keely, 1990 was the one to develop this ‘’traditional’’ hypothesis) which argues that bank competition results in a reduction of market power, decreased profit margins, and reduced franchise value which in turn leads to increased risk-taking behavior by banks. Consequently, greater market (or monopoly) power in the banking system leads to stability because the greater lending opportunities, higher profits, capital ratios and charter values of incumbent banks make them better placed to withstand demand- and/or supply-side shocks, and this environment provides a disincentive for excessive risk-taking (Keeley, 1990; Allen and Gale, 2000, 2004; Carletti, 2008). On the other hand, however, is the competition – stability view (Boyd and De Nicolo, 2005). Accordingly, its proponents argue that competition results in more stability. This comes as a result of the higher interest rates for borrowers which banks, that enjoy market power, can easily charge and this, in turn, complicates the repayment of those loans. This increases the possibility of loan default and, therefore, the risk of bank portfolios, subsequently making the financial system less stable (Boyd and De Nicoló, 2005). Recently, Martinez-Miera and Repullo (2010) questioned the latter by arguing that lower loan rates also reduces revenues from performing loans, and consequently leads to a nonlinear relationship between bank competition and stability.

Studying the trade-off between bank competition and financial stabilitity can be motivated through in the following ways. Policies concerning competition, regulation and supervisions might need to be reconsidered as a result of the restructuring of the financial landscape. For example, should competitive forces be constrained for financial stability to be sustained? Is there a need for such a heavy regulation of the banking sector? Or is there a type of competition that can enhance the

4 stability of the financial sector? The second motivating argument for studying this nexus is that there seems to be a gap in the literature regarding the relationship bank competition – financial stability. Allen and Gale (2000c, p. 268) nicely put it that: “Surprisingly, the relationship between stability and competition has not been studied as extensively as one might expect. On the one hand, there are many models of competition in the literature ... On the other hand, there is a well-developed literature on bank crises ... But there is little on the impact of competition on stability." This study aims to contribute to the literature by explicitly focusing on this interaction. Therefore, my research question is defined as follows:

What is the effect of bank competition on financial stability across 16 European countries, 12 of which are members of the Eurozone?

I carry out the empirical investigation focusing on Europe as a geographic region. In examining the trade-off some studies focus on single country banking market estimation whereas others choose for a world-wide cross-country analysis. The problem with the former is that it is difficult to draw on more general conclusions. An argument against the later choice of estimation is that there exists a considerable segmentation across countries’ banking sectors which could bias the results. I opt for investigation across most of the European countries with developed banking systems.

The appropriate choice of variables to measure both competition and stability is the other important issue that varies vastly in the empirical literature. There are many measures for both competition and financial stability and for robustness I choose to calculate two indicators for bank competition and two for bank financial stability, namely: Lerner Index and Boone Indicator comprise my competition estimates whereas individual bank stability is measured by the Z-score and the ratio of Non-performing loans to gross loans. By way of preview, I find significant results for both a positive and a negative relationship between bank competition and bank financial stability depending on the competition measure employed. Namely, increased market power using the Lerner Index suggests that bank probability of insolvency decreases whereas the country competition indicator – the Boone Index suggests that increased competition results in lower financial instability.

The remainder of the paper is organized as follows. Section 2 outlines the existing theoretical and empirical literature. Description of the data and related methodology are elaborated upon in Section 3. The obtained results and their discussion are presented in Section 4. Finally, Section 5 concludes and provides recommendations for further research.

5

2. Literature Review

As noted in the introduction, academic literature indicates that there is a trade-off between competition and stability in banking. Nevertheless, it remains a controversial and inconclusive issue as regards to the expected sign of the relationship between the two. The competition-stability hypothesis and the competition-fragility hypothesis are the two contrasting views which are used to explain the opposing results in empirical studies (Zigraiova and Havranek, 2015).

2.1 Theoretical Literature

The competition-fragility hypothesis, also known as the franchise value paradigm, holds that bank competition is harmful to financial stability. Many past theoretical findings support this view (Hellmannet al., 2000; Matutes and Vives, 2000). The idea behind is that competition reduces bank profits and as a result leads to a lower banks' franchise value. Consequently, banks’ incentives to undertake more risk for future profits are increased so as to compensate for the decline in the franchise value. Further theoretical arguments in support of this trade-off suggest that the negative relationship can come as a result of superior borrowers screening (Allen and Gale, 2000) as well as enhanced diversification (Beck, 2008). Keeley (1990) and more recently Allen and Gale (2004) present empirical evidence that increased competition in the U.S. banking industry has led banks to take on more risk through the decline in their franchise value. The negative relationship between competition and stability has also been documented in other empirical studies as well such as Berger et al. (2009); Turk-Ariss (2010); Jimenez et al. ( 2013); Fungacova and Weill (2013).

The opposing stance of the literature, built by Boyd and De Nicolo (2005) states that bank competition leads to increased financial stability. They show that the portfolio risks of banks will increase whenever banks gain market power. A low degree of competition in the banking sector increases the interest rates for loans and this leads borrowers to switch to riskier projects which raise the default possibility for banks. Additional arguments which are being put forward in favor of the competition-stability hypothesis are the financial bailout support provided by the government for the ‘’too-big-to-fail’’ banks (Kane, 1989) and the lack of diversity of diversified bank portfolios (Wagner, 2010). The positive interaction between competition and stability finds a lot of empirical support in recent studies, some of which are Boyd et al., (2006); Schaeck et al., (2009); Uhde and Heimeshoff, (2009); Schaeck and Cihak, (2014); Pawlowska, (2015).

6 In contrasts to the aforementioned contrasting views, Martinez-Miera and Repullo, (2010); Berger et al., (2009); Jimenez et al., (2013) and Liu et al., (2013) show that the relation between competition and risk can also be U-shaped. In particular, the chance of a bank to go bankrupt declines at first and then, as competition goes up, it starts increasing.

2.2 Empirical Literature

As noted above, both theoretical as well as empirical literature provides controversial and inconclusive evidence for the effect of bank competition on financial stability. This can be attributed to the selection of measurements for bank competition and financial stability, bank level or country level analysis, the choice of countries and the time periods examined. This section outlines the differences across indicators and provides an overview of the empirical findings.

2.2.1 Measures of Financial Stability and Bank Competition

2.2.2 Measures of Financial Stability

Bank stability measures can be classified into two categories: systematic and individual measures. The former include systematic banking distress and marginal expected shortfall, however, they have the disadvantage that there is not a lot of publicly available data as opposed to the later for which there is vast amount of sources who supply information about bank accounting statements.

Individual banking distress measures often met in the literature include the non-performing loan (NPL) ratio, Z-score, loan loss provisions, capital ratio, distance-to-default and income volatility.

2.2.3 Measures of Bank Competition

Structural and non-structural indicators are the two subgroups of bank competition measures which have generally been used in the literature1:

1

Leon (2014) provides an excellent critical overview of the different methods used to measure competition in the banking industry.

7 1) The structure-conduct-performance (SCP) paradigm –

i) Market shares; HHI index

2) New Empirical Industrial Organization (NEIO) measures - Price responsiveness to cost shifts ii) Lerner Index; Panzar – Rosses H-statistics; Iwata, Bresnahan and Lau mark-up tests

(simultaneous equations approach); Boone Indicator

Essentially, there are two main approaches in measuring bank competition which have been established in the literature: the structural and the non-structural approach. The former, relies on measures such as the concentration ratios or the HHI and it presumes that a higher concentration in the banking system leads to the execution of a less competitive bank conduct, and this results in a better performance (higher profits). This is known as the ‘’Structure-Conduct-Performance’’ (SCP) hypothesis and it was widely used in the 90s for the U.S banking industry (Leon, 2014). Nevertheless, this view was highly criticized because of the possibility of market structure endogeneity problems, meaning that increased profitability is not the result of higher market power but rather it is the consequence of differences in efficiencies among market players (Demsetz, 1973).

As a result of the critiques of the structural models, new methods emerged in the literature. These came to be known as New Empirical Industrial Organization. The non-structural measures that were introduced, specifically useful for the banking system, were the Lerner Index (1934), Iwata (1974), Bresnahan (1982) and Lau (1982) models, Panzar and Rosse H-statistic (1987) and lastly, the newly emerged Boone Indicator. All of the aforementioned indicators have their own advantages and disadvantages and all provide different estimates.

2.2.4 Related Empirical Findings

The empirical literature that is presented here is concentrated around the relationship between bank competition and bank riskiness. In the past, the focus was mostly on a single country whereas the data availability nowadays has allowed an empirical investigation in a cross-country setting which is particularly useful since banks are increasingly engaging in activities not only in their domestic countries but also in their foreign counterparts. Table 1 provides an overview of the empirical literature that has examined the relationship bank competition-financial stability over the years focusing on different geographical areas and using different measures for both competition as well as financial risk.

8 In a cross-country setting with 69 countries and a time span of more than 20 years, Beck, Demirguc-Kunt and Levine (2006) use concentration (C3) as a measure for competition and find that the more concentrated the banking market is, the higher the bank stability is through a lower probability of systematic banking crisis. Nevertheless, they acknowledge that their choice of measurement might not be the most appropriate one. In another attempt to relate bank concentration as a bank competition indicator and examine the impact on bank risk, Levy-Yeyati and Micco (2007) demonstrate statistically significant negative relation between bank competition and bank risk applied to eight Latin-American countries.

Boyd, De Nicoló and Al Jalal (2006) try to empirically challenge the traditional competition-stability hypothesis and demonstrate a positive relation between competition and bank insolvency. They examine two data samples (a single country setting and a cross-country one) and find statistically significant positive effect of bank concentration (HHI) on bank risk (measured by Z-score, return on assets and the standard deviation of the return on assets). Furthermore, De Nicoló and Loukoianova (2007) demonstrate even stronger results once ownership is accounted for. By means of H-statistics as a measure of competition, Schaeck and Cihák (2010) show that more competitive national banking systems exhibit lower probability of systematic crises. In their setting, the competitive conduct is represented by the newly emerged Boone Indicator and the Z-score serves them as a proxy for bank insolvency. Fiordelisi and Mare (2014) as well as Beck et al. (2012) look at the relationship in a cross-country setting focusing on the either the Euro area or Europe as a region and both find economically meaningful and significant positive effect of competition based on the Lerner Index and Z-score as the dependent risk measure. An interesting study is conducted by Kick and Prieto (2013) who apply both the Lerner and the Boone Indicators to the German banking market and find support for both hypotheses depending on the choice of measure.

Overall, this section shows in more details that the empirical literature is very inconclusive with mixe results depending on the choice of estimates. Therefore, the objective of this thesis is to empirically examine the nature of the relationship between bank competition and bank financial stability - positive or negative - as suggested in theory. I conduct the investigation in the context of the banking system including countries part of Europe as a geographical region as well as separately for the Euro zone members for robustness checks.

The following testable hypotheses based on the existing theories regarding the relationship between bank competition and bank financial stability are established:

9

Hypothesis 0: bank competition decreases bank financial stability.

This hypothesis is in line with the competition-stability view.

Hypothesis 1: bank competition increases bank financial stability.

This hypothesis supports the contrasting competition-fragility claim.

Table 1. Overview of Empirical Studies. Effect represents the effect of bank competition on bank

financial stability

Authors Stability Measure Competition Effect Area Period

Keely (1990) Interest Cost Tobin’s q Negative US Demsetz, Saidenberg, Strahan

(1996) Stock Volatility Market-Book Value Negative US Brewer, Saidenberg (1996) Stock Volatility Market-Book Value Negative US Jayaratne, Strahan (1996, 1998) NPL Deregulation Negative US Salas, Saurina (2003) Loan Loss Tobin’s q Negative Spain De Nicolo, Loukoianova (2005) Z-score HHI Positive Non-industrialized Beck, Demirguc-Kunt, Levine

(2006) Crisis Dummy Concentration Negative Cross-country Yeyati, Micco (2007) Z-Score & NPLs P-R H-Stat. Negative Latin America Boyd, De Nicolo, Loukoianova

(2009) Crisis Dummy HHI/ C3 Positive Cross-country Berger, Klapper, Turk-Ariss (2009) NPL/Z-score Lerner/ HHI Negative Developed

countries 1999-2005 Schaeck, Cihak, Wolfe (2009) Crisis Dummy P-R H-Stat. Positive Cross-country 1980-2003 Jimenez, Lopez, Salas (2010) NPLs Lerner Negative Spain 1988-2003 Schaeck and Cihak (2010) Z-score Boone Positive EU 1995-2005 De Nicolo, Ariss (2010) Z-score DepositMarket

Rent Positive Europe 1999-2005 Tabak, Fazio, Cajueiro (2011) Z-score Boone Positive Latin America 2001-2008 Beck, De Jonghe, Schepens (2012) Z-score Lerner Negative EU - Cross-country 1995-2010 Soedarmono, Tarazi (2015) Loan/Deposit

Growth Lerner Index Positive Asia-Pacific 1994-2009 Kick, Prieto (2013) Z-score Lerner/Boone Negative/Positive Germany 1994-2010 Hope, Gwatidzo, Ntuli (2013) Z-score Lerner Positive Africa 2005-2010 Fiordelisi, Mare (2014) Z-score Lerner Index Positive EU -Cross-country 2005-2012 Fu et al. (2014) Z-score Lerner Negative Asia-Pacific 2003-2010

10

3. Empirical Model

This paper uses data on banks from 16 European countries. It covers the period 2004 – 2013 and includes Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain (all those being members of the Euro area ) and Denmark, Sweden, Switzerland and United Kingdom in order to enrich the analysis, summing up to a total of 16 European countries. The main focus of the thesis analysis is to assess the extent to which competition impacts bank stability. Following recent papers (Berger et al., 2009; Fiordelisi and Mare, 2014; Beck et al., 2011, Schaeck and Cihak, 2012), this thesis uses a series of market structure and stability variables to estimate the relationship between competition and stability. The interaction that I want to test can be expressed in mathematical terms as follows:

Financial stability = f (Market structure, Bank controls, Macroeconomic variables)

In order to be able to answer my research questions, three steps are undertaken. The first step involves measuring competition. Then, in the second one, I calculate financial stability indicators. The final stage comprises of statistically examining the trade-off between bank competition and financial stability through running regressions to test the hypothesis that were specified at the end of the previous section. The analysis involves a fixed effects (FE) and a Generalized Methods of Moments (GMM) approaches.

3.1 The Empirical Method

3.1.1 Bank Competition Indices

In this section two measures of bank competition are calculated - Lerner Index and Boone Indicator. Because marginal costs are not observable, I first estimate a cost function to derive them. Then I calculate two competition indices using the estimates marginal costs.

3.1.2 Measuring competition: the Lerner Index

I would like to test my hypothesis in a cross-country setting which varies over time. In order to be able to examine the relationship bank competition – financial stability I need indicators of both that can be estimated at the bank-level and can differ over time. In addition to market share, the Lerner Index is the only competition measure that differs at the bank level and allows one to study the

11 progress of pricing power over time. Another advantage of this non-structural measure is its flexibility (no relevant market definition is required). Furthermore, it can be computed for any type of products.

The Lerner Index represents the degree to which firms (banks in this case) can set a price above marginal cost and it is computed as follows:

𝐿𝑒𝑟𝑛𝑒𝑟𝑖,𝑡=

𝑃𝑖,𝑡− 𝑀𝐶𝑖,𝑡

𝑃_𝑖,𝑡 (1)

where the subscript i denotes each individual bank and t each individual year; P is the price of the output and MC is the marginal cost. Higher values of the index correspond to greater market power. Marginal costs cannot be observed and that is why Schaeck and Cihak (2010) proxy those by the ratio of average variable costs to total income, whereas Van Leuvensteijn et al. (2007) derive them from a translog cost function. My approach is similar to the later, which is an improvement with respect to the former not only because it is closely related with the theory but also because it provides one with the chance to calculate marginal costs of any output (like the loan market in this case) whose costs are not directly observable (Van Leuvensteijn et al., 2007). Following Demirguc-Kunt and Martinez-Peria (2010), Berger et al. (2009) and Beck et al. (2013) methodology, I first estimate, a Translog Cost Function (TCF) using individual bank observations. I run pooled OLS regression over the period 2003-2013 to derive individual bank marginal costs. In a TCF function a bank’s technology constitutes of a production function with many products. I follow the intermediation approach in specifying bank outputs and inputs which presupposes that banks use deposits (as inputs) in order to produce their outputs. Following closely the cost derivation by Berger et al., (2009), Fernandez de Guevara et al., 2005, Hermes and Meesters (2015) and Andries and Capraru (2014) the subsequent translog cost function specification is used comprised of three outputs and three inputs:

12

Equation (2)

(2) where:

 log(TC) is the dependent variable and it denotes the total cost of an individual bank i in a

certain year t measured as the total interest and non–interest expense of a certain bank.

 Qi are output variables

 Q1 - Gross Loans

 Q2 - Other earning assets

 Q3 – Off-balance sheet activities  Wi are input prices variables

o W1 is price of funding measured as – Interest Expense/Total Deposits, Money Market

and Short-Term funding

log

𝑇𝐶𝑖𝑡

𝑊_3,𝑖𝑡

= 𝛽

0

+ 𝛽

1

log 𝑄

1,𝑖𝑡

+ 𝛽

2

log 𝑄

2,𝑖𝑡

+ 𝛽

3

× log 𝑄

3,𝑖𝑡

+

𝛽4 2

log(𝑄

1,𝑖𝑡

)

2

₊

𝛽5 2

log(𝑄

2,𝑖𝑡

)

2

₊

𝛽6 2

log(𝑄

3,𝑖𝑡

)

2

_{+ 𝛽}

7

log

𝑊1,𝑖𝑡 𝑊3,𝑖𝑡

+ 𝛽

8

log

𝑊2,𝑖𝑡 𝑊3,𝑖𝑡

+ 𝛽

9

log 𝑄

1,𝑖𝑡

× log

𝑊1,𝑖𝑡 𝑊3,𝑖𝑡

+

𝛽

10

log 𝑄

1,𝑖𝑡

× log

𝑊2,𝑖𝑡 𝑊3,𝑖𝑡

+ 𝛽

11

log 𝑄

2,𝑖𝑡

× log

𝑊1,𝑖𝑡 𝑊3,𝑖𝑡

+

𝛽

12

log 𝑄

2,𝑖𝑡

× log

𝑊2,𝑖𝑡 𝑊3,𝑖𝑡

+ 𝛽

13

log 𝑄

3,𝑖𝑡

× log

𝑊1,𝑖𝑡 𝑊3,𝑖𝑡

+

𝛽

₁₄

log 𝑄

_3,𝑖𝑡

× log

𝑊2,𝑖𝑡 𝑊3,𝑖𝑡

+ 𝛽

15

log 𝑄

1,𝑖𝑡

× 𝑙𝑜𝑔 𝑄

2,𝑖𝑡

+ 𝛽

16

log 𝑄

1,𝑖𝑡

×

𝑙𝑜𝑔 𝑄

_3,𝑖𝑡

+ 𝛽

17

log 𝑄

2,𝑖𝑡

× 𝑙𝑜𝑔 𝑄

3,𝑖𝑡 + 𝛽18 2

log

𝑊1,𝑖𝑡 𝑊3,𝑖𝑡 2

+

𝛽19 2

log

𝑊2,𝑖𝑡 𝑊3,𝑖𝑡 2

+

𝛽

20

𝑊1,𝑖𝑡 𝑊3,𝑖𝑡

×

𝑊2,𝑖𝑡 𝑊3,𝑖𝑡

+ 𝛽

21

log 𝑒𝑞_𝑡𝑎

+

𝛽

22

log

𝑊1,𝑖𝑡 𝑊3,𝑖𝑡

×

log 𝑒𝑞_𝑡𝑎 + 𝛽

23

log

𝑊2,𝑖𝑡

𝑊3,𝑖𝑡

× log 𝑒𝑞_𝑡𝑎 + 𝛽

24

log 𝑄

1,𝑖𝑡

× log 𝑒𝑞_𝑡𝑎 +

𝛽

25

log 𝑄

2,𝑖𝑡

× log 𝑒𝑞_𝑡𝑎 + 𝛽

26

log 𝑄

23𝑖𝑡

× log 𝑒𝑞_𝑡𝑎 + 𝛽

27

T +

𝛽27

2

T

2

_{+ 𝛽}

28

log 𝑒𝑞_𝑡𝑎 × T + 𝛽

29

log 𝑄

1,𝑖𝑡

× T

+

𝛽

30

log 𝑄

2,𝑖𝑡

× T +

𝛽

₃₁

log 𝑄

_3,𝑖𝑡

× T + 𝛽

₃₂

log

𝑊1,𝑖𝑡

𝑊3,𝑖𝑡

× T + 𝛽

32

log

𝑊2,𝑖𝑡

𝑊3,𝑖𝑡

× T +

13 o W2 is the price of labor – Personnel Expense/ Total Assets (since the number of

employees per bank per year is not fully available)

o W3 is the price of capital – Other Operating Expense/Fixed Assets

ln(eq_ta) is the equity ratio and it is included as a control variable to correct for the difference in loan

portfolio risk across banks. To account for the potential effect of technological change over time I include a time trend (T = YEAR – 2004) and I further interact it with the output as well as input variables. Moreover, I include a control variable at the country level, namely the nominal value of the

LaborCostIndex (in % of GDP)of each country c as suggested by (Fiordelisi and Mare, 2014).

Cost functions have two properties: input prices linear homogeneity and cost-exhaustion (Van Leuvensteijn, 2007). For the former to hold, the sum of the three input prices must equal one whereas their square values and cross terms sum to 0. Following the recent empirical literature I normalize the dependent variable (TC) and all input prices by the price of physical capital (W3) in

order to ensure the properties are being satisfied (Berger et. al. (2009) and Beck et al. (2013).

Having estimated the above equation, I use the coefficients to calculate the marginal cost for banki in

year t for output category loans as follows:

𝑀𝐶𝑖,𝑡 = 𝜕𝑇𝐶𝑖,𝑡 𝜕𝑄𝑖,𝑡 = 𝑇𝐶 𝑖,𝑡 𝑊 3,𝑖𝑡 𝑄𝑖,𝑡 × [ 𝛽1+ 𝛽4log 𝑄1,𝑖𝑡 + 𝛽9log 𝑊1,𝑖𝑡 𝑊2,𝑖𝑡 + 𝛽10log 𝑊1,𝑖𝑡 𝑊2,𝑖𝑡 + 𝛽15log 𝑄2,𝑖𝑡 + 𝛽16log(𝑄3,𝑖𝑡) + 𝛽24log 𝑒𝑞𝑡𝑎 + 𝛽29𝑇 ] (3)

I enclose in the Appendix the estimated results of the translog cost function. Moreover, table 2 displays the derived marginal costs which are averaged per county and per year.

Finally, I obtain the Lerner Index with the price of output P calculated as total revenues (interest plus non-interest income) divided by total assets.

3.1.3 Measuring competition: the Boone Indicator

A new competition measure has been put forward in the literature, as first introduced by Boone (2008). The idea behind this indicator is that competition enhances the performance of efficient firms while it is harmful for less efficient ones. In particular, it estimates the effect of performance (in

14 terms of efficiency) on market share and profits. Following Boone (2004 and 2008), Van Leuvensteijn et al. (2007), and Schaeck and Cihak (2010), I estimate the following equation per country for the whole time period 2004-2013:

ln(𝑀𝑆𝑖,𝑞) = α +β ln (𝑀𝐶𝑖,𝑞) (4)

Where MSq,i and MCq,i denote market share and marginal cost of bank i in output category q (loans in

this case). The slope coefficient  is the Boone Indicator which is expected to be negative (β<0). A bank with low marginal costs is expected to increase its market share and the stronger the competition, the larger the value of β will be (the more negative it becomes). Equation (4) uses a log-linear transformation to account for heteroskedasticity and because it allows for an easy interpretation (β as an elasticity) (Van Leuvenstijn, 2010). One of the assumptions is that banks’ outputs are substitutes which can be viewed as an advantage since an increase in product substitutions will lead efficient banks to gain market share and this, in turn, will raise the competition in the market. Generally, the choice of dependant variable in equation (4) has two alternatives: profits or market shares. I choose the later because profits are the result of market shares and profit margins. Furthermore, they are always positive whereas banks can record losses on their balance sheets and empirical literature adjusts those negative values in various ways in order to be able to use them in a log-linear estimation.

However, the Boone Indicator has some disadvantages as well. In particular, banks may choose to decrease their prices in order to gain market share or just to translate it into an increase in their profits. In addition, this approach does not account consider the level of bank product quality, design or innovations.

While having an estimate of competition over the whole sample period might be interesting to see, I want to look at the evolution of bank competition over time. For this purpose, I follow recent empirical works (Schaeck and Cihak, 2010, Tabak et al., 2011) and modify the main equation (4) with the inclusion of time dummies. The modified Boone Indicator is based on the following equation:

ln(𝑀𝑆𝑖,𝑞𝑡) = α + Σ β*𝐷𝑡* ln(𝑀𝐶𝑖,𝑞,𝑡) + 𝑡=1,…,(𝑇−1)𝑦𝑡𝐷𝑡 +𝜇𝑖,𝑞,𝑡 (5)

Market share is calculated based on total loans. The marginal cost of each bank is obtained from the previous estimation of the translog cost function in (3), d is a time dummy variable from 2004 to

15 2013. Equation (5) thus allows me to examine the evolution of competition over time by looking at the rate at which banks that have low marginal costs in output loans gain market share on a year-by-year basis.

Van Leuvensteijn et al. (2007) and Schaeck and Cihak (2010) point out that endogeneity might be a problem when estimating the specification of equations (4) and (5) because cost and performance are determined concurrently. Therefore, I first test whether endogeneity (MC being correlated with the error term) is present in the two models by means of Hansen statistics. The Sargan-Hansen J-test is testing for overidentified restrictions. Essentially, under the null hypothesis the instruments are valid (uncorrelated with the error term) or exogeneous. Accordingly, if endogeneity is found to be an issue (a rejection of the null hypothesis at the 1%, 5% or 10% level), I use a two-step GMM estimator where the first lag of MC is used as an instrument. Otherwise, I use a within fixed-effects OLS method. In order to control for endogeneity, I use the kernel-based Newey and West’s (1987) heteroscedasticity and autocorrelation consistent (HAC) variance estimation.

3.2 Measuring financial instability

The main bank financial stability indicator used in this thesis is the Z-score which represents the overall risk of failure (Laeven and Levine, 2009; Beck et al., 2013). The Z-score is calculated at the bank level as follows:

Z-score = 𝑅𝑂𝐴𝑖𝑡 +𝐸/𝐴𝑖𝑡_{𝜎𝑅𝑂𝐴𝑖𝑡} (6)

where the subscript i indicate the individual bank, t is the corresponding year; ROA is the return on assets; E/A denotes the equity to total assets ratio; σ(ROA) is the standard deviation of return on assets for bank i using a three-year rolling time window (used to smooth it over the time span). The Z-score indicates by how many standard deviations return on assets need to decline so as to exhaust the equity of a bank. A higher Z-score implies a lower degree of insolvency and thus it gives a direct measure of bank financial stability. The natural logarithm of the Z-score is considered in the analysis in order to smooth out higher values of the distribution. As robustness check, I calculated the Non-Performing loans to total loans weighted by the market share of loans of each bank per each specific year. Table 3 below shows the evolution of the Z-score over time. In addition, table 4 in the Appendix shows the evolution of the alternative measure of bank risk, the NPL ratio.

16

Table 3 The profression of the Z-score per country per year. It is computed as the sum of the Return on Assets and Equity to Total Assets ratio divided by the standard deviation of the Return on Assets using a three-year rolling window

3.3 Other variables

In order to control for the differences across banking markets in European countries, I employ additional bank-specific and macroeconomic control variables to be able to examine the potential impact other factors might have on the relationship between bank competition and financial stability. The bank level-fundamentals included in the analysis are: loan loss provisions to interest income as a key measure for asset quality, non-interest income to total assets to account for revenue composition, cost-to-income ratio to control for banks’ efficiency, asset composition (fixed assets to total assets, loans to total assets) (Beck et al., 2011), liquidity, asset growth to account for different risk preferences and size. Liquidity is built as liquid assets to deposits and short term funding and it gives and information on how many resources a bank can quickly utilize to cover cash outflows (Fiordelisi and Mare, 2014). The size variable, computed as the natural logarithm of bank total assets, indicates the extent to which a bank can increase its solvency through the ability to diversify its business. Bank lending behavior, captured by the ratio of loans to total assets (LA), is expected to be negatively related to bank stability, as credit is one of the riskiest areas of banking business.

Country 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Mean AUSTRIA 3.103 2.202 2.424 30.951 3.358 3.303 4.130 5.658 4.384 4.725 6.424 BELGIUM 1.642 2.929 3.259 3.338 3.281 3.549 4.280 3.988 4.176 3.619 3.406 FINLAND 3.588 2.242 2.797 4.548 4.837 4.445 4.491 5.058 5.767 5.664 4.344 FRANCE 3.236 3.166 2.506 2.998 2.746 2.719 3.407 3.322 3.089 3.144 3.033 GERMANY 31.582 31.723 31.934 35.094 36.430 36.238 36.375 36.505 36.751 36.559 34.919 GREECE 0.540 4.119 3.125 3.860 3.888 3.766 3.790 1.398 1.302 1.641 2.743 IRELAND 3.188 3.882 4.000 4.278 3.003 3.148 4.979 4.571 4.389 5.594 4.103 ITALY 1.720 4.628 3.661 3.620 4.867 33.155 4.505 5.032 5.106 5.026 7.132 LUXEMBOURG 5.342 4.388 4.966 4.848 2.899 3.897 4.826 4.114 4.350 4.080 4.371 NETHERLANDS 6.810 4.300 5.203 5.179 5.037 4.468 4.673 4.837 5.725 5.928 5.216 PORTUGAL 1.908 2.588 2.582 3.693 2.842 3.325 3.830 3.952 4.035 3.959 3.271 SPAIN 2.309 3.226 2.705 3.612 3.747 3.602 3.799 4.505 3.847 3.849 3.520 EU 12 5.624 6.108 6.067 6.824 6.689 9.301 7.178 7.026 7.140 7.188 6.874 DENMARK 3.974 4.871 4.701 4.653 3.250 4.392 4.573 4.387 4.535 4.516 4.385 SWEDEN 5.207 4.143 4.356 3.610 2.987 2.846 4.233 4.682 4.010 4.467 4.054 UNITED KINGDOM 2.908 3.710 5.120 4.943 5.155 3.409 3.435 5.255 5.764 4.204 4.390 SWITZERLAND 3.228 3.348 3.210 3.327 3.278 3.084 3.640 4.099 4.787 5.072 3.707 Europe 16 5.018 5.342 5.409 7.660 5.725 7.459 6.185 6.335 6.376 6.378 6.189 Z-score

17 Furthermore, the financial freedom index and the market share interacted with loan growth are used as instruments. The financial freedom index measures the degree of openness of the financial system in a specific country and ranges between 0 (not open) to 100 (very open). It is used to instrument the Boone Indicator because the level of interference by the government can affect competition. Moreover the interaction term of market share and loan growth is used because both of them have an impact on the Boone Indicator.

In addition, I control for the influence of the macroeconomic environment using the GDP growth rate, the total long term unemployment rate, inflation as an indicator of macroeconomic imbalances, credit growth to private sector by financial institutions (in % of GDP) and HHI index of concentration to control for market structure effect (Schaeck and Cihak, 2014). The rate of growth of real GDP is a control variable because there might be a correlation between business cycles and investment opportunities for banks (Uhde and Heimeshoff, 2009). 2 Consequently, a positive sign is expected of the coefficient when economic growth increases the opportunities for investment. In addition, borrowers’ solvency may be higher under increasing economic performance which in turn raises the banks’ asset quality. Inflation rate is included because whenever interest rates rise, this increases inflation which results in higher net interest margins and profitability by banks. However, funding costs might also increase and the net effect is ambiguous. The credit growth variable is incorporated because excessive provisions of credit cause a decline in capital ratios and therefore, affect financial soundness. Since concentration has been proven to be a poor proxy for competition, I include it on the country level and include the natural logarithm of the total asset size to control for the banking system size.

Due to the big amount of variables incorporated in my empirical model, I include the evolution of the values of all indicators over 2004-2013 in Table 5 below for one of the Eurozone members: the Netherlands. Moreover, Figure 1 indicates the progression of specified macroeconomic factors over 2004-2013.

2 _{It must be noted that GDP might not be the most appropriate macroeconomic control variable for the}

purpose of this study since different countries in Europe exhibit different banking sector size in terms of GDP (e.g. in the Netherlands it is very high)

18

Table 5 Key variables values over time

2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Mean Z-score 6.81 4.30 5.20 5.18 5.04 4.47 4.67 4.84 5.73 5.93 5.22 EQ/TA ratio (%) 9.93 9.87 6.00 6.49 14.63 16.01 16.09 12.89 14.64 12.71 11.93 STDEV(ROA) 0.01 0.01 0.00 0.00 0.01 0.01 0.02 0.01 0.01 0.01 0.01 ROA 0.01 0.01 0.01 0.01 0.02 0.01 0.01 0.00 0.01 0.00 0.01 NPL 0.18 0.13 0.21 0.23 0.24 0.25 0.23 0.20 0.23 0.24 0.21 Boone -0.05 -0.11 -0.10 -0.07 -0.07 -0.05 -0.08 -0.10 -0.02 -0.01 -0.07 Lerner 0.43 0.37 0.38 0.43 0.51 0.40 0.20 0.71 0.35 0.45 0.42

Average price of banking activities 0.06 0.05 0.05 0.05 0.05 0.08 0.05 0.10 0.04 0.04 0.06

MC 0.03 0.03 0.03 0.03 0.04 0.04 0.04 0.03 0.03 0.02 0.03 MS loans 0.07 0.07 0.07 0.07 0.06 0.06 0.05 0.05 0.05 0.06 0.06 MS Assets 0.13 0.13 0.13 0.13 0.10 0.09 0.08 0.08 0.08 0.10 0.10 Loans/TA 0.51 0.53 0.52 0.58 0.64 0.64 0.61 0.63 0.65 0.64 0.59 ln(TA) 14.22 14.73 14.82 14.88 14.79 14.80 14.77 14.81 14.75 14.66 14.72 Cost-Income Ratio 0.63 0.56 0.60 0.69 0.67 0.81 0.62 0.72 0.72 0.65 0.67

Loans Loss Provisions / Interest Income0.10 0.14 0.21 0.27 0.16 0.20 0.15 0.07 0.05 0.03 0.14

Non Interest Income / TA 0.02 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01

FA/TA 0.01 0.01 0.01 0.08 0.07 0.06 0.06 0.06 0.06 0.07 0.05

Liquid Assets / Deposits & ST Funding (%)90.26 124.11 90.19 80.28 45.03 79.58 30.95 44.07 58.55 72.16 71.52

Growth of Total Assets (%) 3.79 10.68 8.01 33.38 0.76 -6.04 20.38 7.02 1.15 -0.24 7.89

Growth of Loans (%) 9.07 14.57 8.80 34.39 1.38 3.38 0.61 2.77 5.92 -0.59 8.03

HHI (Assets) 0.17 0.18 0.18 0.19 0.22 0.20 0.20 0.21 0.20 0.21 0.20

GDP per capita growth (annual %) 1.50 2.01 3.65 3.97 1.68 -3.79 0.55 1.19 -1.95 -1.02 0.78

Inflation (annual %) 0.99 1.62 2.16 1.87 2.32 0.46 1.16 0.14 1.25 1.08 1.31

Unemployment (% of total labor

force) 5.00 5.30 4.30 3.60 3.00 3.70 4.50 4.40 5.30 6.70 4.58

Domestic credit provided by

financial sector (% of GDP) 160.23 167.73 167.41 185.58 183.26 207.20 197.79 197.91 201.70 193.01 186.18 Financial Freedom Index 90.00 90.00 90.00 80.00 90.00 90.00 80.00 80.00 80.00 80.00 85.00

Total Bank Size (ln Total Assets) 16.56 17.23 17.45 17.57 17.65 17.57 17.64 17.68 17.66 17.58 17.46

Netherlands

Number of Banks: 24

19

Figure 1 Macroeconomic indicators over time for the Netherlands

One can make several inferences based on the information in table 3. First, bank stability has been increasing ever since the start of the global financial crisis whereas both the Boone and the Lerner Index indicate a decreasing trend of bank competition. Furthermore, the market is highly concentrated (HHI is increasing) starting from 2008 onwards which coincides with increasing unemployment and relatively low GDP per capita growth. The non-interest income to total assets is relatively low (around 2%) meaning that banks are primarily focused on carrying out the traditional activities (loans, etc.).This can be further seen from the loans to total assets ratio, which is gradually increasing. However, the cost-income ratio has been increasing pointing to possible banking system inefficiency. Furthermore, non performing loans ratio has been gradually increasing suggesting that banks who lend out more money (high loans to total assets ratio) also tend to have riskier loan portfolios. On the other hand, the loan loss provisions in terms of interest income is a relatively small percentage in the last couple of years which can be a sign of a banking system with a relatively low chances of loan defaults.

3.4 The Empirical Model

Empirical literature takes on two main approaches when estimating the relationship bank competition – financial stability: a cross-country or single country setup. The cross - country setting supplies useful information for the average relation between between competition and stability for the set of countries under investigation (developed countries as in Berger, Klapper, and Turk Ariss (2009), the European Union as in Schaeck and Cihak (2010)), while controlling for other

country-3.00 6.70 1.08 1.68 -3.79 -1.02 -10.00 -5.00 0.00 5.00 10.00 15.00 20.00 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Unemployment (% of total labor force) Inflation (annual %)

GDP per capita growth (annual %)

20 specific factors such as macro-economic conditions. Nevertheless, single country studies (such as Keeley (1990), Jimenez, Lopez, and Saurina (2010), Boyd, De Nicolo, and Jalal (2006)) document a large degree of variation in the competition-stability relationship. I use the following setup:

𝑅𝑖𝑠𝑘𝑖,𝑐,𝑡= 𝛽𝑖 + 𝛽1𝐶𝑜𝑚𝑝𝑒𝑡𝑖𝑡𝑖𝑜𝑛𝑖,𝑐,𝑡−1+ ξ𝑛𝑘 𝑘𝑋𝑖,𝑐,𝑡−1 + ƴ𝑛𝑘 𝑘𝑀𝑐,𝑡−1 + ηt + ϵi,j,t (7)

where Risk is the measure for financial stability, in particular the natural logarithm of the Z-score or alternatively, the ratio of non-performing loans to total loans. Competition is represented either by the Lerner Index or the Boone Indicator; X is the vector of controls at the bank-level. M is a vector of controls for the macroeconomic conditions. βi is the time invariant component of the error. ε

indicates robust standard errors clustered at the bank level. β, ξ and ƴ are the parameters to be estimated. Each independent regression is lagged by one-year to minimize the issue of endogeneity bias. A panel-fixed effects model is used with robust standard errors clustered at the individual bank level to examine the relationship between individual bank stability, competition and. However, investigating the effect of market power on bank risk-taking behavior might be subject to an endogeneity bias. Schaeck and Cihak (2008) suggest that there could be a reverse causality, meaning that the degree of risk-taking by banks could impact the level of competition in the market and consequently the market power measure. If banks are faced with high probability of default they might engage in more risk-taking activities or, alternatively, decrease the level at which their products are priced, which would increase their market power. In order to account for this problem, I consider instrumental variable approach using the GMM estimator. The GMM estimator is more efficient than a standard 2SLS because it accounts for heteroskedasticity and it does not require any distributional assumptions on the error terms (Berger et al., 2009). Following the precedents from previous studies, I consider three instrumental variables for each of the competition measures: the lagged values of the Lerner Index, interaction term between market share and loan growth and cost-income ratio (for the Lerner measure) and financial freedom index (for the Boone Indicator).

3.5 Data

Bank level financial information is obtained from Bankscope database covering the time period 2004-2013. It is worth mentioning that the data availability improves significantly after 2006. This thesis analysis focuses on 16 European Countries (due to lack of data I exclude Estonia, Cyprus, Malta, Lithuania, Latvia, Slovenia and Slovakia). My final dataset includes the countries: Austria, Belgium, Germany, France, Spain, Luxembourg, Netherlands, Italy, Sweden, Denmark, Switzerland, United

21 Kingdom, Finland, Greece, Ireland and Portugal. Considering data sufficiency, I include all types of bank in the data set (commercial banks, cooperative banks, savings banks, real estate & mortgage banks, investment banks, non-banking credit institutions and bank holding companies whenever their subsidiaries are not included) except central bank and clear & custody bank in each investigated country. I consider consolidated data where possible (a consolidated statement includes the data of its controlled subsidiaries) and unconsolidated if the former is unavailable. I go through each individual bank to ensure that no double-counting exist. For example, looking at the Netherlands, the bank ABN AMRO Bank has data available from 2009 onwards. However, a careful examination provides further insights into how one needs to operate with the data extracted from Bankscope. The bank was re-established in 2009 following the acquisition and break-up of the original ABN AMRO by a banking consortium consisting of Royal Bank of Scotland Group, Santander Group and Fortis and for this entity there is historical data available. Table 6 below presents the number of banks per country which are used in the analysis as well as the % presence in terms of the total number of institutions considered. Table 7 in the Appendix provides a full overview of the variables used, their definition as well as the source from which they were obtained.

22

Table 6 Description of the total number of banks which are used in this thesis per country. The third column shows a country’s percentage contribution to the full sample

4. The Empirical Evidence

4.1. Lerner Index Results

The results derived from the estimation of equation (1) are presented in Table 8 below. The intuition behind Lerner Index is that greater competition in the market pushes the price of loans close to its marginal cost resulting in a smaller Lerner Index. Generally speaking, Lerner indices fluctuate significantly across countries, more precisely it declines somewhat in 2006-2008 and increases again in 2009 through 2013. Financial globalization has been increasing since the global financial crisis start in 2007 and this might have led to the increase in market power in the last years. Furthermore, banks are increasingly engaging in cross-border merger and acquisitions activities and thus rising their efficiency at the expense of lending rates being left uncut. Furthermore, the start of the global financial crisis coincides with a decrease in the market power. This can be attributed to the large

Austria 284 7.2% Belgium 36 0.9% Finland 18 0.5% France 298 7.6% Germany 1708 43.4% Greece 11 0.3% Ireland 22 0.6% Italy 557 14.1% Luxembourg 37 0.9% Netherlands 24 0.6% Portugal 35 0.9% Spain 141 3.6% EU 12 3171 80.54% Denmark 87 2.21% Sweden 94 2.39% United Kingdom 210 5.33% Switzerland 375 9.53% Europe-16 3937 100% Country Number %

23 capital losses suffered by banks as well as losses from non-performing loans which in turn has reduced their efficiency. Looking into the numbers we can infer that Sweden, Austria, Belgium, Germany and Greece seem to have the most competitive banking systems whereas Ireland, Finland, Italy, Luxembourg, the Netherlands and France are characterized by the least competitive banking markets.

Table 8 Lerner Index time series progression. The estimates below are calculated based on equation (1). Bank market power is first estimated on the bank level, then it is averaged per country and per year. The last column summarizes the average across 2004-2013.

4.2 Boone Indicator Results

The results of the Boone Indicator are presented in Table 9 and 10. The first table presents the estimation for the whole period 2004-2013, while the second one displays the time specific evolution of the Boone measure. For none of the countries have marginal costs been classified as endogenous and that is why OLS is run instead of GMM estimation. I expect that banks with low marginal costs gain market share (β < 0). Competition tends to increase this effect, since more efficient banks surpass less efficient ones in terms of performance. The more negative is β, the higher is the competition level in a banking market. Nevertheless, as Van Leuvensteijn et al. (2007) show, it is also possible to derive positive values. According to the estimates obtained Greece, Ireland, Finland,

COUNTRY 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Mean AUSTRIA 0.215 0.136 0.149 0.387 0.321 0.313 0.225 0.227 0.162 0.312 0.245 BELGIUM 0.520 0.283 -0.089 0.118 0.095 0.130 0.020 0.035 0.454 0.522 0.209 FINLAND 0.243 0.394 0.134 0.433 0.497 0.628 0.703 0.624 0.501 0.490 0.465 FRANCE 0.495 0.202 0.406 0.278 0.179 0.314 0.151 0.293 0.361 0.708 0.339 GERMANY 0.195 0.573 0.212 0.168 0.168 0.451 0.280 0.476 0.169 0.184 0.288 GREECE -0.243 -0.160 0.121 -0.137 -0.094 -0.124 -0.053 -0.108 0.232 0.391 -0.017 IRELAND 0.252 0.125 0.254 0.244 0.216 0.722 0.825 0.809 0.768 0.679 0.489 ITALY 0.983 0.633 0.296 0.664 0.036 0.221 0.209 0.299 0.056 0.125 0.352 LUXEMBOURG 0.207 0.442 0.386 0.337 0.349 0.446 0.384 0.234 0.218 0.410 0.341 NETHERLANDS 0.428 0.367 0.381 0.428 0.509 0.404 0.198 0.715 0.353 0.447 0.423 PORTUGAL 0.480 0.200 0.284 0.161 0.516 0.184 0.046 0.538 0.187 0.290 0.289 SPAIN 0.452 0.500 0.052 0.265 0.250 0.206 0.230 0.312 0.206 0.095 0.257 EU 12 0.352 0.308 0.216 0.279 0.253 0.324 0.268 0.371 0.306 0.388 0.307 DENMARK 0.469 0.464 0.431 0.294 0.097 0.187 0.168 0.271 0.169 0.223 0.277 SWEDEN 0.342 0.348 0.415 0.356 0.254 0.665 0.646 0.637 0.797 0.542 0.500 UNITED KINGDOM0.334 0.290 0.692 0.261 0.157 0.430 0.318 0.318 0.318 0.318 0.344 SWITZERLAND 0.227 0.225 0.245 0.250 0.113 0.290 0.232 0.161 0.275 0.146 0.216 Europe 16 0.350 0.314 0.273 0.282 0.229 0.342 0.286 0.365 0.327 0.368 0.313 Lerner Index

24 Belgium, Luxembourg and Spain have the most competitive banking sectors while Italy, Portugal, Germany, Austria, the Netherlands and Switzerland the least competitive ones. However, I also acknowledge that this sort of comparison should be made carefully, since the estimation of the Boone Indicator depends on how it was initially modeled.

Table 9 Boone Indicator estimated for the whole sample period per country based on equation (4)

Country

Boone

t-value

Austria

0.039

1.100

Belgium

-0.267

-2.140

Finland

-0.372

-1.180

France

-0.035

-1.040

Germany

0.741

10.210

Greece

-0.385

-1.360

Ireland

-0.141

-0.860

Italy

-0.072

-1.380

Luxembourg

-0.265

-1.570

Netherlands

-0.031

0.650

Portugal

0.034

0.610

Spain

-0.075

-2.330

EU 12

-0.069

Denmark

-0.032

-1.250

Sweden

0.109

1.890

United Kingdom-0.086

-1.560

Switzerland

-0.004

-0.120

Europe-16

-0.054

25

Table 10 Boone Indicator estimation results based on equation (5) per country and per year

YEAR

Boone z-value Boone z-value Boone z-value Boone z-value

2005 -0.046 -1.800 -0.075 -0.320 -2.625 -15.6 -0.085 -3.170 2006 -0.065 -2.58 -0.235 -1.97 -0.250 -1.64 -0.100 -4.08 2007 -0.055 -2.34 -0.253 -2.43 -0.214 -1.31 -0.080 -2.93 2008 -0.004 -0.17 -0.331 -3.01 -0.307 -2.53 -0.091 -3.29 2009 -0.039 -1.48 -0.381 -3.71 -0.256 -2.21 -0.082 -2.75 2010 -0.029 -1.05 -0.403 -3.71 -0.267 -2.55 -0.050 -1.86 2011 -0.020 -0.76 -0.322 -2.87 -0.303 -2.71 -0.058 -2.23 2012 -0.043 -1.64 -0.435 -1.75 -0.269 -2.78 -0.054 -2.17 2013 -0.036 -1.44 -0.343 -2.63 -0.256 -2.83 -0.071 -2.89 Observations 1475 165 68 1724

Austria Belgium Finland France

YEAR

Boone z-value Boone z-value Boone z-value Boone z-value

2005

0.452 17.3

-4.314 -0.430 -0.689 -1.410

0.039 0.730

2006

-0.058 -2.21

-3.358 -0.43

-0.715 -1.50

-0.048 -1.29

2007

-0.061 -2.43

-3.412 -0.43

-0.620 -1.52

0.001 0.03

2008

-0.051 -2.04

-4.830 -0.42

-0.690 -2.41

-0.026 -0.83

2009

-0.056 -2.41

-4.659 -0.43

-0.648 -1.96

0.026 0.99

2010

-0.050 -1.96

-5.169 -0.42

-0.401 -1.16

0.038 1.47

2011

-0.071 -2.87

-5.491 -0.43

-0.511 -1.74

0.043 1.59

2012

-0.078 -3.08

-5.036 -0.43

-0.397 -0.95

0.008 0.32

2013

-0.075 -2.96

-5.782 -0.43

-0.749 -0.75

0.001 0.06

Observations

12181

81

95

3070

26

YEAR

Boone z-value Boone z-value Boone z-value Boone z-value

2005

-0.561 -5.280 -0.048

-1.89

0.285 1.18

-0.260 -2.82

2006

-0.671 -6.26

-0.109 -0.300

0.248 1.45

-0.218 -4.89

2007

-0.685 -6.60

-0.097 -1.60

0.069 0.42

-0.068 -2.01

2008

-0.554 -4.55

-0.083 -0.55

-0.020 -0.14

-0.340 -2.69

2009

-0.418 -3.06

-0.069 -2.16

-0.066 -0.49

-0.115 -2.21

2010

-0.556 -4.15

-0.070 2.22

-0.024 -0.20

-0.193 -3.39

2011

-0.688 -4.28

-0.018 -2.66

-0.092 -0.56

-0.114 -1.98

2012

-0.640 -4.23

-0.052 -3.04

-0.136 -0.67

-0.197 -3.00

2013

-0.668 -4.69

-0.010 -3.40

-0.186 -0.96

-0.233 -3.06

Observations

185

112

149

542

Spain

Luxembourg

Netherlands

Portugal

YEAR

Boone

z-value Boone

z-value Boone

z-value

Boone

z-value

2004

2005

-0.108 -1.670

-0.077 -2.160

-0.024 -1.330

0.141

1.470

2006

-0.128

-2.93

-0.077

-2.19

-0.034

-1.76

0.109

1.20

2007

-0.151

-3.54

-0.080

-2.60

0.005

0.23

0.163

1.28

2008

-0.153

-3.30

-0.071

-2.47

-0.025

-1.24

0.102

1.01

2009

-0.124

-3.11

-0.100

-3.48

0.030

1.34

0.080

0.85

2010

-0.112

-2.94

-0.066

-2.55

0.033

1.57

0.050

0.62

2011

-0.102

-2.10

-0.075

-2.19

0.030

1.34

0.072

0.98

2012

-0.082

-1.91

-0.041

-1.41

0.024

0.99

0.010

0.08

2013

-0.120

-2.87

-0.047

-1.47

0.010

0.48

-0.030

-0.30

Observations

495

512

2618

538

27

4.3 Lerner Index, Boone Indicator and the Z-score

Putting the two competition measures and the financial stability indicator together can provide a preliminary source of interdependence. Since both the Lerner Index and the Z-score are built using profitability, a positive relation might not receive economically meaningful interpretation due to measurement bias (Beck et al., 2012). Therefore, I investigate the relationship between the Lerner Index and Z-score over time in Figure 2 which provides information on the time series evolution of the Lerner Index and the Z-score. Both variables are averages for the whole sample of 16 European countries per year. The graph is built on two axes – the let-hand axis represents the Lerner Index while the right-hand axis shows the Z-score respectively. The time period during which we can observe a close interdependence between the two measures is 2008-2010, which corresponds to the first years after the start of the global financial crisis. According to Figure 2 it is characterized by an increase in market power and lower default risk (2008 to 2009) and then a decline in both (2009-2010). This time period has seen a lot of banks withdrawing their foreign operating subsidiaries and vast amounts of government interventions such as guarantees, recapitalization measures, asset reliefs and other liquid measures. In the subsequent years, we cannot infer anything on the relationship between the two measures because it is rather ambiguous. Figure 3, 4 and 5 provides further information on the relationship between the second competition measure which is used in this paper, the Boone Index, and the Lerner Index (Figure 3); between the Boone Index and the Z-score (Figure 4) and all the three measures in one graph (Figure 4). Because Boone is inversely related to competition, I take the opposite (-β) and the graphs reflect the positive values of the index in order to make directly proportional inferences about competition. Contrary to the Lerner Index, we can observe that until 2010 the newly emerged competition measure and the proxy for bank insolvency move in opposite directions. In fact the same holds when we just look at the two bank competition measures suggesting that using alternative approaches to determine a market’s competitive landscape can yield different results.

4.4 Empirical Results

Table 11 presents some correlation between the variables of interest.

Regression-based evidence on the relationship between bank market power and bank soundness is reported in Table 12 and 13. In this pooled cross-country setup, I regress the natural logarithm of the

28 Z-score on the Lerner Index or the Boone Indicator and a set of bank and macroeconomic control variables, as described in regression Equation (7).

The results in column 1 of Table 12 consider only bank-specific variables. A positive and significant relationship is found between market power and bank soundness indicating that an increase in market power (reduced competition) leads to lower risk-taking by banks and lower probability of insolvency. This result is in line with the competition-fragility hypothesis that the higher the level of competition in the banking sector, the lower the charter value of banks which further spurs them to take on more risk. The results are also consistent with the findings of other existing literature that also uses the Lerner Index as a market power proxy (Beck et al., 2012; Kick and Prieto, 2013). In column (2) I add macroeconomic control variables and the effect is again positive and significant at the 1% level. Furthermore, the point estimate of the concentration indicator (HHI) is also found to be positive and significant proving that concentrated markets exhibit lower chance of insolvency. GDP and inflation are both significantly impacting the stability measure in a positive way implying that developments in the general business cycle improve the investment opportunities by banks and hence their stability while inflation increases bank solvency through higher interest margins. As mentioned before, there might be concerns of potential endogeneity bias of the Lerner Index. Faced with increased chances of insolvency banks can potentially engage in aggressive undercutting of competitors in terms of prices and, in turn, affect the market competitiveness. Even though I use lagged versions of the explanatory variables, endogeneity might still persist in the case when bank default events are well anticipated by bank managers. Consequently, in columns (3) and (4) I use the GMM estimation method. I use three instruments: Lagged Lerner Index (t-2), interaction term between market share and loan growth and cost-income ratio. The choice of instruments stems from the dynamic panel data literature (Blndell and Blond, 1998) as well as other past empirical literature (e.g. Koetter, Kolari, and Sperdijk, 2012; Beck et al., 2012). Market share directly impacts competition and the interaction with loan growth is included because of their interdependence – the former increases with the rise of the later. Increased cost-income ratio, on the other hand, translates into inefficient performance which could result in a loss of market share. I find again significant positive effect of market power on financial stability. In order to test the validity of the instruments I present the p-value of the Hansen J-test. The null hypothesis is that the model is overidentified and a rejection would mean that the set of instruments can be excluded from the model. In both settings (4) and (5) the p-value is larger than 10%. Moreover, I present the Breusch-Pagan test of

29 heteroskedasticity. Looking at the output, other interesting results can be noticed. Larger banks (higher natural logarithm of total assets) are more risky and this effect is significant at the 1% level in settings (1), (3) and (4). In both settings (FE and GMM), concentrated markets (HHI) are found to be significantly impacting stability, however, with two opposing signs: a positive and a negative. GDP and unemployment both enter with the expected positive and negative signs respectively.

In table 13, an alternative measure of competition is employed: the Boone Indicator. Here the results become even more interesting. Since the Boone Indicator is the inverse of competition (the negative of competition) the negative sign found can be interpreted as evidence that a decrease in competition would lead to declining Z-scores and thus increased insolvency risk. This supports the competition-stability view and is in line with other studies that use the Boone Indicator as a proxy for competition (Schaeck and Cihak). Higher values of the Boone Index show that banks that are more inefficient, are punished harder in terms of market share, and profits respectively. A large Boone Indicator shows that a banking market is letting inefficient institutions to gain relatively more market share, thereby signaling that the market is characterized by low degree of competition. Larger banks are found to be negatively affecting stability and thus are considered more risky. Whether I use Boone or the Lerner Index to measure bank competition, I find significant negative effect of non-interest income to total assets in the GMM estimator. This means that the more a bank engages in non-traditional banking activities, the less stable it becomes as it might be engaging in more risky activities.

Generally, since the measures of competition are supposed to estimate the degree of market power, a natural question arises concerning the magnitude and type of the effect each index has on bank risk. According to my results so far, they seem to have an independent impact on financial stability, either positive (competition-fragility view) or negative (competition-stability view), with both indexes having a significant effect at the 1% level. However, the results that increased market power enhances financial soundness (Lerner) and low competition improves probability of default (Boone) might seem unusual at first. Nevertheless, both indexes measure different scopes of bank competition. On the one hand, the Lerner index represents banks’ ability to generate profits purely stemming from monopoly power mark-ups. On the other hand, the Boone Index decrypts how harsh banks are being punished if they operate inefficiently. The later provides an alternative way of interpreting the results concerning bank stability. If a banking market is characterized by more

30 competition, in specific, by higher values of the Boone Index (more negative), less efficient players are driven out by more efficient ones (Turk-Ariss, 2010 and Schaeck and Cihak, 2010 show that competition is efficiency-enhancing). There is both theoretical as well as empirical literature that provides evidence that efficient banks are associated with lower level of risk. Efficient institutions could have improved screening and monitoring systems (Petersen and Rajan, 1995). Furthermore, Berger and DeYound (1997) empirically prove the positive relationship between efficient banks and their non-performing loans. Based on the presented arguments, taking the positive relationship of high competition (low Boone Index) can be interpreted as stability improving through an efficiency channel. Looking at the Lerner index, an increased competition which reduces bank risk is in line with the hypothesis that the franchise value will reduce the risk-taking incentives and improve financial stability. At the same time, an increase in the competition between banks leads to less incentives for risk-switching at the borrower level, because bank that are more efficient can expand their monitoring and screening systems (Kick and Pireto, 2013).

The two opposing sings of the impact of competition on financial stability should not be considered as being the result of potential estimation bias. It should be noted that the Lerner Index is measured at the individual bank level and thus it can account for potential inefficiencies across each institution whereas the Boone Index is a country measure and it is derived from a regression. Furthermore, the later comprises of the inefficiencies across banks in terms of their costs and the resulting gains in market share.