Competition and financial stability in Banking sector in Eastern European countries

Competition and financial stability in Banking sector in

Eastern European countries

Master Thesis

MSc. International Economics and Business

Groningen, August 2010

Laima Dambrauskaite, s1940961

Thesis Supervisor

Thesis Co-assessor

dr. M. Koetter dr. Robert Inklaar

Abstract

This paper examines the impact of bank competition on bank financial stability using an unbalanced panel data set of 205 banks in 9 Eastern European countries (new members of EU) during the period of 1998 to 2008. Lerner index and Boone indicator are applied as proxies for market power, while banking soundness is measured by Z-score. The results suggest, after controlling for a number of bank-specific and environment variables, a negative link between higher level of competition and banking stability. Moreover, this study also finds that competition has been decreasing in most banking markets of Eastern European countries in the last decade. Furthermore, no evidence has been found that foreign banks are more fragile than domestic ones. Finally, results imply that EU membership might have had some positive impact on financial soundness of CEE countries’ banking sector.

Table of Contents

1. Introduction ... 4

2. Literature review ... 7

2.1. Competition and financial soundness ... 7

2.2. Competition and Concentration ... 9

2.3. Overview of methods of measuring bank competition ... 10

2.4. Overview of methods of measuring banking fragility ... 13

3. Data and methodology ... 15

3.1. Data and measures ... 15

3.2. Empirical model ... 20

4. Results ... 22

4.1. Specification ... 22

4.2. Competition ... 23

4.3. Bank stability... 27

4.4. Testing the relation between bank soundness and level of competition ... 27

4.5. Robustness of the results ... 33

5. Conclusions ... 34

References ... 37

1. Introduction

The ongoing structural change in banking sector in terms of rapid consolidation has been an increasing concern in terms of what is its impact on financial stability. As consolidation might impair competition (Bikker, 2004, p.278), research on the effect of the increase of the degree of banks’ market power has important policy implications. However, the current debate on this topic has not yet reached a consensus and the results of the empirical studies are still ambiguous.

Two opposing views in the literature addressing this concern have emerged. Some studies argue that a negative trade-off exists between competition and bank soundness (Keeley, 1990; Beck et al., 2003, 2006; Berger et al., 2003, etc.). More recent research, on the other hand, provides evidence for the “competition-stability” view (De Nicoló, 2000; De Nicoló et al., 2004; Schaeck and Čihák, 2008a; Schaeck et al., 2009; etc.).

A voluminous amount of literature investigates the introduced relationship in developed countries, focusing on a broad set of countries or Western Europe (Schaeck and Čihák, 2008a), however, only few studies explored the effects of consolidation in Eastern European countries (Turk Ariss, 2010; Uhde and Heimeshoff, 2008). The latter study addresses country specific effects among Eastern and Western European countries. The authors conclude that Eastern European banking sector is less competitive, contains fewer diversification opportunities, and that a higher number of government-owned banks are prone to financial fragility. Bikker and Spierdijk (2008), using Panzar-Rosse approach, find that Eastern Europe has experienced a significant decrease in competition over the past 10 years. Turk Ariss (2010) provides support for the traditional view that increased competition may impair banking stability in developing countries. These issues are even more important in light of adverse consequences of the recent financial crisis for some of those countries.

Another interesting fact about this set of countries is that the emerged banking sector in Eastern and Central European countries has undergone a lot of changes, such as rapid consolidation and privatization. This resulted in large presence of foreign (mainly Western European) banks in these countries1. It is therefore interesting how competition is influenced by these changes and how this affects the overall stability in the banking sector. While the effect of ownership on efficiency has been widely researched (e.g. Lensink, 2008, who finds a negative relationship between foreign ownership and bank efficiency), literature on what implications a foreign bank entry has for financial stability is less abundant. Foreign banks may bring better banking practices that improve safety of

1

the banking system; however greater number of competitors might impede stability (Claessens, et al., 2001) as well be a cause of new sources of risk (exchange risks, etc.).

Next, studies examining the competition – stability relationship use different measures of competition. More recent studies favor new measures of market power, such as Boone indicator (Boone et al., 2005) instead of traditional measures, such as Panzar and Rosse H-statistic or concentration and Hirschmann-Hirfendahl indexes because the new indicator overcomes the problems of the previous measures. However, Boone indicator has only been applied to the set of developed countries (Schaeck and Čihák, 2008a, 2008b).

Furthermore, although the literature on the banking sector in transition economies is vast, little attention 2 has been given to what are the implications of entering the EU for the new member states. Čihak and Fonteyne (2009) analyze changes in the new member states in terms of financial integration and financial stability after they joined the EU. By using financial soundness indicators (FSI3) they find that after joining EU the results for overall soundness of financial sector in new member states are mixed: in terms of profitability (ROA) and returns of equity the situation has improved substantially, while credit growth indicates new potential risks.

The aim of this paper is to investigate what is the effect of competition on banking soundness in 10 new members of European Union4, employing a new way to assess competition – Boone indicator as well as the traditional Lerner index, for no consensus has been reached on what is the best way to measure competition (Cárbo et al., 2006). Based on findings of Turk Ariss (2010) for developing countries, this study hypothesizes that increased competition might negatively affect financial soundness (H1). Another hypothesis tested in this study is whether Eastern European countries have experienced a decrease in level of competition during the ongoing process of consolidation (H2). The third hypothesis suggests that foreign banks are more fragile than domestic ones (H3). Finally, the fourth hypothesis (H4) tested in this paper is whether joining the EU has had any significant positive effect in terms of stability in banking markets of Eastern European countries (by using a more comprehensive measure of stability – Z-score).

This analysis complements previous studies in several different aspects. To begin with, it is the first paper to analyze the effect of competition of banking stability solely on Eastern European countries. Also, while previous scholars used data from developed countries (Schaeck and Čihák, 2008a), it is the first study to apply Boone indicator to a set of transition countries. Finally, it is one

2

Main topics include banking crisis (what triggers financial distress), process of consolidation, entry of foreign banks, and privatization of state banks (Hawkins and Mihaljek, 2001, give an excellent overview of these topics).

3

ROA (return on assets), ROE (return on equity), nonperforming loans to total loans and capital to assets ratios.

4

of the first studies to analyze the effect of EU membership on banking soundness for Eastern European countries.

2. Literature review

2.1. Competition and financial soundness

A growing body of literature is addressing the relationship between competition, concentration in the banking sector and financial soundness. A dominant view is related to a “franchise value” or “charter value” paradigm5. A seminal article by Keeley (1990) started the discussion on this relationship, arguing that competition decreases banks’ charter value, which in turn increases the default risk. The reason for this is that banks that have more market power have a higher capital to assets ratio and lower default risk, which is indicated in lower risk premiums on large CD’s. Decrease in charter values induced a decline in monopoly rents and resulted in growth of bank failures in U.S. in 1980s. The results of this study have been largely supported by other empirical works.

Authors of this strand argue that higher profits act as a safety net against detrimental shocks and increase the franchise value of the bank ( Boyd et. al, 2003), and, as discussed by Keeley (1990), this results as lower incentives for bankers to engage in more risky activities. As franchise value is an intangible asset that can only be present if bank remain in business, opportunity costs of bankruptcy are too high, therefore, banks hold more equity, less risky or smaller portfolios (Berger et al., 2009). Uhde and Heimeshoff (2008) review other supporting arguments for the trade-off between competition and stability, presented by previous studies. In addition to “charter value” argument, large banks not only provide better credit monitoring services, but are also easier to monitor. Furthermore, economies of scale and scope provide better opportunities for risk diversification (Boyd and Prescott, 1986) as well as geographical risk diversification. Jimenez et al. (2007) present a good review of theoretical studies on this topic.

Empirically, a positive relationship between competition and fragility has been largely supported by studies. Keeley (1990) found a positive relationship between market power (measured by Tobin’s q) and solvency ratio for U.S. banks. Jimenez et al. (2007), using data on Spanish banks, find a negative relationship between loan market power and bank risk, when the former is measured by Lerner index.

Other studies provide evidence for competition-fragility view using a cross-country data. Beck et al. (2003), using data from 70 countries over the period 1980-1997, find that crises are more

5

likely in less concentrated banking systems. However, it is also found that higher competition decreases the probability of systemic banking crisis occurrence in a country. Beck et al. (2006) examine 69 countries and, using different concentration measures and controlling for ease of monitoring, indicate a positive relationship between concentration and banking system stability. Banking systems that are more concentrated tend to have larger, better-diversified banks. However, this study also notes that concentration is not always a good proxy for competition and indicates something else besides the competitive environment.

A recent study by Turk Ariss (2010) presents some more evidence on the trade-off between competition and soundness. Analysis of interaction between market power, stability and profit efficiency provides empirical justification for the traditional view that higher degree of market power leads to more stability. This paper also has important implications for emerging economies, as the study is based on data from 60 developing countries.

A growing amount of literature finds a positive relationship between competition and financial soundness. This opposing view of positive relationship between competition and stability was proposed by Boyd and De Nicolò (2005), who made an assumption that competition is allowed in deposit markets but suppressed in loan markets. Boyd et al. (2006) extended the previous study by allowing competition in both deposit and loans’ market and tested the predictions of so-called CHV (modified Allen and Gales’, 2004 model) and BND models, using a sample of U.S. firms. This paper provides evidence for the positive relationship between competition (measured by HHI) and bank risk. Proponents of this strand of literature suggest that increase of the number of large financial institutions is associated with more severe moral hazard problems created by “too-big-to-fail” (Mishkin, 1999), as large banks are more likely to receive public subsidies, thus encouraging managers to engage in more risk-taking. Additionally, more market power may result in higher interest rates charged to loan customers, which in turn makes it harder to repay loans and discourages risk-averse customers. Higher interest rates also may increase riskiness of loan portfolios because of adverse selection problems (Berger et al., 2009). Finally, as opposed to competition-fragility strand, proponents of the competition-stability literature argue that if bigger banks are also more complex, then large financial institutions are not easier to monitor.

Fiordelezi (2009) showed a positive relationship between bank concentration (measured by HHI or C5) and the likelihood of financial distress after controlling for bank specific variables. Jimenez et al. (2007) find that for Spanish Banking system Lerner measures of loan market power have a negative relationship with portfolio risk. However, as they only consider loan portfolio risk, the results cannot be applied for the overall bank risk.

Schaek and Čihák (2008a) apply a large sample of banks in U.S. and 10 European countries during the years 1995-2005 and offer an industrial organization-based explanation for the competition-stability view. This study investigates efficiency as a transmission mechanism by which higher competition increases bank soundness using a two-step approach. First, they establish a positive link between competition (measured by Lerner index) and profit efficiency in banking, and in the second step Boone indicator is employed to show a positive relationship between competition and soundness. Schaek et al. (2009) illustrate the same link by employing the Panzar and Rosse H-statistic as a measure of competition in 45 countries. Additionally, the latter paper confirms Claessens and Laeven (2004), Beck et al. (2006) findings that concentration and competition describe different characteristics of banking systems. Uhde and Heimeshoff (2008) also provide evidence for competition-stability strand by applying Z-score as a proxy for banking stability and HHI and concentration indexes to measure concentration. They use bank balance sheet data from banks across the EU-25 for the period of 1997-2005. Among other findings, they also indicate differences in terms of level of competition among Eastern and Western European countries.

Finally, some studies do not find support for any of the two discussed hypotheses. Berger et al. (2009) show that the two opposing views in the literature sometimes can produce similar predictions on the effects of competition and market power on banking soundness. Their study uses data on banks from 23 developed countries, and the empirical findings show that banks with a higher degree of market power increase loan risk; however, they also have a lower overall risk exposure. Carletti and Vives (2008) review the literature on competition and stability in banking sector and suggest that market power can have some effect on incentives to take risk. Allen and Gale (2004) present a number of theoretical models regarding the question of competition and stability that provide different outcomes. They indicate that this relationship is complex, however sometimes a positive link exists.

2.2. Competition and Concentration

proxy for something else besides market power (Beck et al., 2006). Cetorelli (1999) argues that market structure cannot be used to determine competition, as direct empirical analysis using individual bank data is necessary in order to get accurate results. Claessens and Laeven (2004), Schaek et al. (2009) suggest that concentration and competition describe different characteristics of banking systems because they find independent effects from the two on crisis.

Concentration measures rely on traditional structure-conduct-performance (SCP) paradigm (which is discussed more thoroughly in the following section) that assumes only one way causality from market structure to performance. Thus, studies that have aimed to detect the degree of competition using various measures of concentration (H-statistics, HHI), have found ambiguous results.

2.3. Overview of methods of measuring bank competition

The literature provides a vast number of studies on measuring competition in banking. As competition cannot be observed directly (costs and price data of single banking products are unavailable), indirect measures have to be applied. Several approaches to estimating competition have been developed in the literature.

The traditional Industrial Organization (IO) structural approach includes structure–conduct– performance (SCP) paradigm and efficiency hypothesis. This paradigm analyzes whether greater concentration in the market causes less competitive bank behavior and results in higher performance, while the efficiency hypothesis investigates whether it is the efficiency that increases performance, as greater efficiency in production and managerial organization could result in higher profits.

The SCP hypothesis has been tested by estimating concentration, using such measures as Herfindahl-Hirschman Index (HHI) or n-firm concentration ratio. The latter is the sum of market shares of n (usually three of five) largest banks. As the data to estimate this measure is easy to get, this has been a widely used measure in empirical studies, which assume that banks with larger market shares have more market power and they use that.

concentration may be a result of fierce competition, as Claessens and Leaven (2004) find a positive link between concentration and competition. De Nicolo et al. (2007), Uhde and Heimeshoff (2008) are some of the recent studies that used HHI and other concentration indexes to investigate the competition-stability relationship.

Non-structural models have been developed to deal with deficiencies of the structural ones. These models are the Iwata (1974) model, the Bresnahan (1982) model, Panzar and Rosse (P–R) model (1987), also identified as New Empirical Industrial Organization approaches. They analyze competition without using explicit information about the structure of the market and use some form of price mark-up over a competitive benchmark. Thus, this theory allows a more direct assessment of banks’ behavior and the focus is on the competitive behavior of the banks instead of concentration and the number of banks.

Commonly used model to measure competition in the literature is Panzar and Rosse (1987) H-statistic, which is computed as the sum of elasticities of the reduced form revenue of the bank with respect to input prices. Claessens and Laeven (2004), etc. have applied this method to measure competition. The greater is the transmission of cost changes into price changes, in both directions, the more competitive the market is (Carbo et al., 2006). The H-statistic ranges from -∞ to 1: H<0 indicates monopoly, H=1 implies perfect competition, while a value between 0 and 1 is interpreted as monopolistic competition. As this measure is derived from level data and reflects bank-specific differences in production functions (Claessens and Laeven 2004), assumptions about the market are not necessary, which is an advantage of this measure against the concentration estimates (Schaek et al. 2009). Boone (2007) has identified several shortcomings of this approach. First, if H is less or equal to 0, nothing is actually learned, because a negative sum of elasticities may indicate both - monopoly and oligopoly. Second, in order to calculate H, information on factor prices is needed, which is usually hard to collect.

A number of studies (e.g. Jimenez et al., 2007; Berger et al., 2009; Schaek and Čihák 2008a) have used the Lerner index to determine the degree of market power. This measure represents mark-up of price (average revenue) over marginal costs, which means that higher degree of market power results from rising margin or a higher Lernex index. The index requires a proper estimation of the marginal cost of the product, as it is calculated as output price minus the marginal costs, divided by the price. Under a perfect competition, Lerner index is 06, while values close to 1 indicates

6

monopoly. An empirical approximation of Lerner index used in the literature is the price-cost-margin (PCM), which is calculated as output price minus the price-cost-marginal costs, divided by output price.

Finally, a new alternative measure has been introduced recently by Boone (2005), which, according to the author, is more robust from a theoretical and an empirical point of view than traditional measures. Boone indicator indicates profits elasticity (PE) and is defined as the percentage fall in profits due to a percentage increase in (marginal) costs. The idea is that in a more competitive market, the same percentage increase in costs will lead to a bigger fall in profits, thus firms in more competitive markets loose more in terms of profits for being inefficient.

Boone (2007) argues and explains why PE is a better measure than HHI and PCM. In order to show that, Boone starts with differentiating between two ways in which competition can increase in the market: either there are more firms in a market because of a fall in entry barriers or due to more aggressive behavior of firms already operating in the market. As more firms enter the market, concentration decreases. Therefore, an increased competition as a result of more firms is correctly captured by concentration measures such as Herfindahl-Hirschman Index. However, when firms tend to switch to a more aggressive behavior, inefficient firms are forced out of the market, and consequently concentration increases. This case should not be accounted as a fall in competition. Furthermore, market shares of efficient firms increase at the expense of inefficient firms. This reallocation raises HHI as well, resulting in a positive correlation between HHI and PE, contrary to what is expected. PCM have the similar shortcomings as the HHI when measuring the effect of increased competition. Greater level of competition usually reduces banks’ PCM. However, since more efficient banks have higher PCM, increase of their market shares may result in higher average PCM in the industry, which is again an opposite outcome to what is expected. To sum up, by using simulations, Boone (2005) shows that the competition measures PCM and HHI are consistent when competition increases as a result of fall of entry barriers. However, these two measures fail when firms switch to more aggressive behavior. The paper argues that PE is able to correctly measure both forms of changes in competition.

moderate amount of data is necessary in order to compute Boone indicator. In addition to these advantages, several shortcomings are also indicated. Firstly, it is assumed that banks usually pass part of their efficiency gains on to the customers. Besides, this approach does not account for differences in the quality of bank products (Leuvansteijn et al., 2007). Another study is the one from Schaek and Čihák (2008a), which investigates the relationship between competition and financial soundness, using Boone indicator.

2.4. Overview of methods of measuring banking fragility

Literature on measuring banking fragility is ample. Scholars have proposed a lot of different financial soundness indicators (FSIs). Čihák (2007) provides an excellent overview of broad set of measures of financial distress used in literature and categorizes them into four main groups: probabilities of default for individual banks based on fundamental data; probabilities of default for individual banks derived from market data; distance to default for group of banks (portfolio); and first and Nth bank to default.

Market based measures are the ones that central banks, international financial institutions increasingly encourage to use in order to measure financial soundness (Chan-Lou and Sy, 2006). These measures include distance to default, bond prices, credit default swaps. One of the main advantages of these indicators over the traditional ones (derived form fundamental data) is that securities’ prices include information about the future, while the traditional ones are more “backward-looking”. What is more, prices on securities are available at a higher frequency and on real-time basis, comparing to financial statements’ data, which is reported only every 3 months. Hillegeist et al. (2004) argue that these measures are more reliable than balance sheet information when liquid markets for securities are available. However, this is one of the limitations as well, because of the liquidity and transparency requirement for the market. Another drawback is that explicit information might not be available for the institutions, which securities are not publicly traded. Market-based indicators have also been used at a macro level, e.g. “portfolio DD” (distance-to-default, which is a probability that the market value of a firm’s assets falls below the value of its debt).

others), measuring financial soundness of an individual bank, is Z-score 7 It has also been widely used by scholars analyzing the relationship between competition and/or concentration and financial stability. One reason for its popularity is that the score is inversely related to institution’s probability to default as well as the easiness of calculation. Another advantage is that this indicator enables the comparison of different financial institutions (e.g. commercial and savings banks). Despite its numerous usage in the literature, this indicator also has several shortcomings. First, as it is based solely on accounting data, in cases when banks are able to misreport the data, indicator might overestimate the soundness of a financial institution. Moreover, Z-score does not reflect any correlation between different banks (effects of contagion). Other accounting based measures are various financial soundness ratios (e.g. nonperforming loans to total loans8). Although they are very easy to calculate, scholars have not yet found a clear connection between them and probability of default or systemic stability (Čihák, 2008). Furthermore, some of these measures (nonperforming loans) only measure loan risk, while Z-score accounts for overall bank risk (Berger et al., 2009) Finally, Z-score is a time-variant measure of the bank’s distance-to-default. For the above mentioned reasons, Z-score is preferred measure of stability in this study.

Another way to measure distress has recently been used by Cipollini and Fiordelesi (2009). They calculate a ratio between Economic Value Added and the shareholders capital invested at time t-1, which is defined as Shareholder Value Ratio. Finally, when determining the relationship between competition and financial stability, some studies (Schaek et al., 2009 among others) use logit model, where the crisis is a dummy variable that takes on the value one if a systemic crisis is observed in a particular year. However, this approach does not evaluate each bank separately, as it takes in to account the whole country. Also, it might not be the most reliable measure as banking crises are not described in the same way across countries. Therefore, it is always debatable what is the exact start and the end of the banking failure.

7

z-score for banks is computed as sum of average return on assets and equity to total assets divided by standard deviation of return on assets.

8

3. Data and methodology

3.1. Data and measures

The data consists of bank-level financial statements for the years 1998-2008, obtained from Bank Scope database published by Bureau van Dijk. Sample consists of data of 9 new European Union member states for commercial, savings, cooperative banks. All the data in the study is inflation-corrected. The original sample (555 banks) is filtered by excluding banks for which the main variables are not available. Also, banks that have negative values for total assets, equity and loans are excluded. The remaining sample consists of 267 banks, of which 251 are commercial, 7 savings and 9 cooperative banks. Because some banks have missing values for variables for years 1998-2006, these observations are also dropped. In addition, observations below 1 and over 99 percentiles for ROA, Total Assets, Profits before taxes (PBT) and Total Revenues are deleted as well. Moreover, banks with 1 observation are dropped.

Table 1. Summary Statistics for all countries for 1998-2008 of variables used for calculating Lerner index, Boone indicator and z-score

Note: all the data presented in thousands of Euros (except for ROAA)

Variable Obs Mean Std. Dev. 95 percentile 5 percentile

After imposing all restrictions, 1 country (Estonia) was dropped from the sample in order to get more precise estimations for each country, as it had only 26 observations9. Finally, an unbalanced panel data set that consists of 205 banks and 759 observations is used for estimations. Coverage is much more limited in the years of 1998-2001 and is not available for all countries in the sample. For Slovakia, data is available from 2001, while for Bulgaria, Hungary and Lithuania data is available from 2002. Table 1 displays summary statistics for main variables adopted in the study.

Lerner indicator

Following the new empirical IO approach, this paper uses conventional Lerner indicator to account for competition. As presented earlier, the Lerner index indicates the mark-up of price over marginal costs and is computed at a bank-level. Following Berger et al. (2009), Lerner index can be defined as: TA TA TA MC P P LI =( − )/ (1)

Here, PTA accounts for price of the output and is calculated as the ratio of total revenues to

total assets, following Maudos and Guevara (2005) and Carbo et al. (2006). MCTA is the marginal

cost of producing an additional unit of output.

In order to calculate MC, a translog cost function is estimated first (for each country separately):

(2)

9

where Cost accounts for Total costs (sum of personnel expenses, other non-interest expenses, and interest expenses) and Q is total assets (Carbo et al., 2006, Turk Ariss, 2010, etc.). Wk is a vector

of input prices: W1 is the price for labor, W2 – funds, W3 – physical capital, which are calculated as ratio of personnel expenses to total assets, interest expenses to deposits 10 and other operating and administrative expenses to fixed assets, respectively (Maudos and Guevara, 2004). Zk is vector of

control variables (following Schaek and Čihák, 2008a) – netputs. Z1 is for fixed assets (fixed assets/ total assets), Z2 is for loan loss provisions (loan loss provisions /equity capital), and Z3 is for equity capital (equity capital/total assets). Trend is used to capture the technical change (e.g. Maudos and Guevara, 2007), which is commonly used in econometrical analyses of production which evolve time series data 11(Baltagi and Griffin, 1988). The term Trend 2 is included for consistency with the second order approximation notion of the translog from (Coelli et al., 2005). Finally, Eq (2) includes only two input prices (excluding W1, which is the price for labor), since in order to impose input price homogeneity, costs and all input prices are scaled by W1. Standard symmetry restrictions and linear homogeneity in input prices are imposed:

θ11= θ22, θ12= θ21, ω11= ω22, ω21= ω12

∑

∑

∑

∑

The purpose of a translog model is to link as much as possible the level of output (which is comprised of a number of products) and prices of input factors to total costs (Bikker, 2004). Having more output factors results in greater accuracy, however it also might cause collinearity between the output factors. Therefore, one output (Total assets) is used instead of having several outputs, e.g. loans and other earning assets.

Economic theory states that MC=∂C/∂Q. Because we are differentiating translog Cost function, the outcome of ∂lnC∂lnQ is MC/AC. Therefore, marginal costs are obtained by ∂lnC∂lnQ*AC.

Marginal costs are then computed as follows:

      _{+ + + +} =

∑

∑

β

β

γ

η

λ

In Bankscope database deposits correspond to Deposits and Short term funding.

11

Here, we also include price of labor (W1). γ1 is obtained from requirements for linear

homogeneity in input prices, that is:

γ

γ

γ

Lerner index is calculated for each bank. Also, this indicator is calculated for each country to demonstrate different levels of competition.

Boone indicator

An alternative way to measure competition is used as well in this study. Proposed by Boone (2001), Boone et al. (2005), this novel indicator has been applied to banking industry as well.

Following Leuvensteijn et al. (2007), Leuvensteijn et al. (2008), Schaek and Čihák (2008a, 2008b), demand function for the banking industry, where each bank i produces product qi is:

∑

where a captures size of the market, b is market elasticity of demand, and parameter d captures extent to which consumer sees different products as close substitutes. Bank maximizes its profits and chooses and optimal level of output qi.:

i i i i =(p −c )q

π

where ci is bank’s marginal costs

Assuming that a>ci and 0<d≤b, first order condition (p=mc) can be written as:

∑

Finally, for a banking system with 3 banks 3 first order conditions can be solved as follows:

      ₋ − + +       _{+ −} +       ₋ =

∑

If entry costs are k, then bank will only enter the market if πi ≥k. Competition can increase in

entry costs k decline (Boone et al., 2005). This paper concludes that firms with higher variable costs (or marginal costs relative to price) have lower profits.

Following Schaek and Čihák (2008b), the model can be defined by the equation:

it t T t t it t t T t it)= +

∑

∑

π

α

β

γ

where β is the Boone indicator for country j and time t, dt are the time dummies. The (8) equation is estimated for each country separately. Yearly estimates are obtained from the year dummies and marginal costs interaction variable. Boone indicator stands for profit elasticity and illustrates the change in profits due to a one percent change in marginal costs. The stronger the competition, the greater is the absolute value of Boone indicator. This is an improved version of Boone indicator estimation comparing to previous studies, for it uses profits as in original model proposed by Boone et al., (2005) instead of market shares (Schaek and Čihák, 2008a; Leuvensteijn et al., 2007). Moreover, marginal costs here are not approximated by average variable costs (Boone et al., 2005; Schaek and Čihák, 2008b); therefore, the results are expected to be more precise. Marginal costs are computed from Eq(3). The log specification implies that, for instance, an estimated value of -0.5 suggests that a bank with 1% higher marginal costs than its peer would have 0.5% lower profits comparing with a more efficient bank.

Bank risk

In order to measure bank risk, this study uses a popular measure (Uhde and Heimeshoff, 2008; De Nicoló et al., 2004 among others) – Z-score – of bank’s distance to default, which is a proxy for financial stability. Z-score is defined as follows:

Z = (ROA+ EA) /σ (ROA), (9)

capital ratio and decreases with greater returns’ volatility (σ). Therefore, higher values of Z-core imply lower probability of insolvency risk. Schaek and Čihák (2008a) point out some reasons for the popularity of this measure. Firstly, it combines bank’s capital, profitability as well as the risk it faces. Moreover, this indicator is easy to calculate and empirical studies have shown the usefulness of this measure. σ(ROA) is calculated for each bank, taking observations from all available years into account. Alternatively, some scholars use for instance a three-year rolling time window for σ(ROA). However, since most banks in the sample do not have more than 4-5 observations, this method is not applied. Z-score is computed for each observation as well country averages in year t for inclusion into regressions.

3.2. Empirical model

In order to test the relation between the level of competition on banking stability, two types of regressions are run using different measures of competition, as presented earlier. Model 1 is specified as follows: ijt jt ijt ijt ijt

ijt COMP COMP X C

Z =

α

β

γ

δ

φ

ε

where dependant variable Zijt is Z-score for bank i at time t, COMPijt measures competition (Lerner

index), X ijt is a vector of bank specific controls and Cjt is vector of country-specific controls.

COMPijt2 (quadratic term of Lerner index) is included in order to account for possible nonlinear

relationship between competition and banking soundness (Berger et al., 2009, Turk Ariss, 2010)

Model 2 is specified very similarly to the previous one:

jt jt jt jt jt COMP X C Z =α +β +δ +φ +ε (11)

where dependant variable Zjt is Z-score for country j at year t, COMPjt stands for competition

(measured by Boone indicator), X jt is a vector of bank specific controls and Cjt is vector of

country-specific controls.

Assets), which is included in order to control for bank size, as larger banks might be subject to regulators’ too-big-to-fail policies (Mishkin, 1999).

Equity ratio to total assets is included in order to control for unobserved risk preferences, as risk-aware banks might tend to have higher equity ratios. (Stiroh, 2004). Ratio of loan loss provisions to total assets controls for loan-portfolio quality and is one of the main measures for credit risk (Uhde, Heimeshof, 2008). Bank’s credit exposure is captured by ratio of total net loans to total assets (Turk Ariss, 2010), which also controls for differences in banks’ asset portfolio (Stiroh, 2004).

Macroeconomic control variables are retrieved from World Development Indicator (WDI) database provided by World Bank. Inclusion of the following is necessary as they can affect both financial stability and market power, while also reducing the omitted variable bias. Natural logarithms of per capita Gross Domestic Product and Real interest rate are included in order to control for differences in economic environment, because although all nine countries are transition economies, they are still quite different in terms of the level of economic development (for instance, Slovenia’s GDP per capita is twice as large as Bulgaria’s). More developed economies (with higher GDP per capita) are expected to have more stable banks. In contrast, interest rate is expected to be negatively associated with Z-score, as higher real interest rates indicate lower level of economic stability.

Furthermore, two additional variables are added in order to test the last two hypotheses (H3 and H4). Foreign ownership (dummy variable that takes on value of 1 if total foreign owned shares exceed 50% of total bank ownership) is included in the regression in order to analyze the effect of ownership on financial stability12. Finally, membership in European Union is also controlled by creating a dummy variable that takes on value 1 if the country is a member of EU at the year t. For Bulgaria and Romania it is the case for all observations for the years 2007 and 2008, while for remaining countries for observations for the years 2004-2008.

12

4. Results

4.1. Specification

The first step of the whole estimation process is obtaining marginal costs. Table 7 presents results for translog cost function estimations for each of the nine countries. Explanatory variables are banks’ output (Total Assets), input prices (funding and physical capital, scaled by wages) and additional characteristics (fixed assets/total assets, loan loss provisions/equity, and equity/total assets).

In order to see which estimator (Fixed or Random effects) is consistent, Hausman test is performed for all nine countries13. Hypothesis has been rejected in favor of Fixed effects model for all countries except for Hungary and Lithuania (see Table 3). In cases where the null hypothesis is not rejected, additionally the Breusch –Pagan Lagrange multiplier test 14 is performed in order to test for random effects. Results indicated that Pooled OLS (POLS) is consistent in case of Lithuania, while translog cost function for Hungary is run with random effects estimator. Table 3 reports estimated R2 (in most cases, it is equal to 1, which means that the model fits the data perfectly) and Wald

χ

Using estimated coefficients, marginal costs are calculated for each bank for each year (Eq. 3). Estimated country averages are reported in Table 8. Maudos and Guevara (2007) report that average marginal costs in banking sector of EU-15 countries were 0.012 in 2002, while Eastern European banks have on average higher marginal costs ranging from 0.01 to 0.1 (which also have declined in most of the countries during the sample period). These differences, however, can be partially explained by the fact that the sample in this study comparing to the one from Maudos and Gueavara (2007) is heavily dominated by commercial banks, which tend to have higher marginal costs (Leuvensteijn et. al, 2007). Authors argue that a possible reason for this might be that higher funding costs of commercial banks are a result of fewer attracted deposits comparing to savings or cooperative banks. Also, differences in the level technological development could explain this.

13

It compares estimated coefficients from both models and checks the null hypothesis that the error term is uncorrelated with the explanatory variables. If the null hypothesis is not rejected, both estimators are consistent, while if it is rejected, only fixed effects estimator remains consistent (Hill et al., 2007).

14

4.2. Competition

Results for measures of competition are presented in Tables 2 and 3 (values for each country). Although estimations of the level of competition in Eastern European countries is still very scarce in the literature, several studies have assessed it using different measures, such as Panzar-Rosse approach (Bikker and Spierdijk, 2008; 2009), Lerner index (Turk Ariss, 2010), as well as concentration ratio (Uhde and Heimeshoff, 2008).

Average value of conventional Lerner index across all countries is 0.35. Turk Ariss (2010) reports an average value of conventional Lerner index for some Eastern European and Central Asian countries is 0.29. The latter study uses a very similar approach of calculating Lerner index (by obtaining marginal costs from translog cost function). The difference in results comparing with this study can be mostly explained by slightly different set of countries and by different time span15. Direct comparison with other articles, however, is not possible because of different methodologies applied16.

Table 2. Summary statistics for variables used in regressions

Variable Obs Mean Std.

Dev. 95th percentile 5th percentile Lerner index 759 0.35 0.32 0.90 -0.09 Boone indicator 746 -0.63 4.39 1.33 -1.41 Z-score 719 30.69 38.21 88.98 6.34 ln(Total Assets) 759 13.81 1.42 16.06 11.23

Loan loss provisions/ Total

assets17 759 5.80 6.77 18.02 0.39

Loans/Total Assets 759 0.56 0.19 0.87 0.25

Equity/Total Assets 759 0.11 0.06 0.23 0.05

ln (GDP) 759 8.48 0.56 9.44 7.59

Interest rate 759 3.64 3.72 9.70 -2.90

Results for Western European countries differ among the studies as well. Berger et al. (2009) find an average value of 0.22; Schaek and Čihak (2008a) report Lerner index values ranging from

15

Bikker, Spierdijk (2008) find that competition in Eastern European countries has decreased over the years; thus, that would imply increase in values of Lerner index.

16

Although direct comparison is not possible, indirect comparison can be implemented by simply ranging all 9 countries from most competitive to least and then comparing the obtained lists from several studies that used different measures of competition. The results differ significantly as expected, because all measures of competition rely on different assumptions. Hence, this confirms Carbo et al. (2006) findings that there is little connection between different measures.

17

0.38 to 0.72 for Loan markets; Maudos and Guevara (2007) estimate an average of 0.33 for EU-15 (in loan markets).

Table 3. Summary statistics for Boone indicator, Lerner index and Z-score across countries Country Variable Obs Mean Std.

Dev. p95 p5 Bulgaria Lerner 86 0.41 0.15 0.64 0.19 Boone 86 -0.66 0.43 -0.12 -1.41 Z-score 83 40.48 79.32 65.11 11.36 Czech Republic Lerner 66 0.53 0.23 0.87 0.05 Boone 66 -3.01 14.34 12.48 -33.94 Z-score 59 35.18 41.27 163.86 4.25 Hungary Lerner 82 0.14 0.16 0.36 -0.10 Boone 76 -0.35 0.29 0.22 -0.74 Z-score 81 27.34 13.97 48.75 7.43 Latvia Lerner 80 0.64 0.19 0.96 0.34 Boone 80 -0.33 0.42 0.30 -0.90 Z-score 77 33.95 48.75 120.84 5.37 Lithuania Lerner 50 0.23 0.10 0.38 0.06 Boone 47 -0.45 0.63 0.70 -1.39 Z-score 46 34.82 31.39 136.04 11.25 Poland Lerner 102 0.03 0.17 0.33 -0.20 Boone 102 -0.12 0.60 1.19 -0.92 Z-score 89 31.15 25.88 101.75 7.47 Romania Lerner 125 0.24 0.20 0.51 -0.13 Boone 124 -1.26 1.42 -0.13 -4.48 Z-score 122 16.36 11.91 32.70 4.69 Slovakia Lerner 75 0.07 0.12 0.23 -0.09 Boone 73 -0.55 0.51 0.20 -1.36 Z-score 73 36.28 24.67 101.93 13.07 Slovenia Lerner 93 0.86 0.06 0.96 0.77 Boone 92 0.76 0.30 1.27 0.23 Z-score 89 31.30 19.61 86.17 7.83

be explained by a rapid consolidation process where large banks are created with a higher degree of market power. Another explanation for the decreased competition is the switch from traditional to more modern banking products. Bikker and Spierdijk (2008) argue that more sophisticated and adapted services are likely to become an advantage for banks to use their market power. The latter study finds that over the years competition has been decreasing in Eastern European Markets (on average it decreased by 10% during 1994-2004). Our results indicate that during 1998-2008 competition in banking sector of Eastern Europe has decreased significantly: from an average value of 0.28 to 0.36 in 1998 and 2008, respectively (29% increase in value of Lerner index). Therefore, the H2 stating that this region has experienced a decline in competition is confirmed.

In contrast with the Lerner index, Boone indicator is calculated on the whole industry, not firm level. Results for average Boone indicator values for all countries are presented in Table 3.

Regressions are run (Eq.8) for each country separately in order to yearly estimates for each country. Tests for heteroskedasticity (Breusch-Pagan/Cook-Weisberg) and autocorrelation (Wooldridge, 2003) have been implemented. Heteroskedasticity has been found present in Bulgaria, Czech Republic, Lithuania and Slovakia, while Wooldridge test for autocorrelation indicated that autocorrelation is present in all countries but Latvia and Lithuania. Hence, Generalized Least Squares method (GLS) is chosen because it deals with one or both of the above problems of panel data (variances’ inequality or correlation between the observations). This approach is applied for all countries expect for one (Lithuania), because it has not been diagnosed with heteroskedasticity and autocorrelation. Therefore, Hausman test that checks for any correlation between error and

explanatory variables in random effects model is performed. 2

χ

for thistest is 52.72, with a p-value of 0.00; thus, fixed effects estimator is preferred in Lithuania’s case. All the other countries’ Boone indicators are obtained with GLS estimator.

The problem of endogeneity when estimating Boone indicator has been addressed by some previous studies. It might be possible that profits and marginal costs can be jointly determined (profits are computed as revenues minus costs). Schaek and Čihák, (2008 a,b) and Leuvensteijn et al. (2007) use instrumental variable (IV) approach in order to deal with this issue. However, this study does not employ instrumental variable approach when estimating Boone indicator18.

18

The results presented in Table 3 suggest that competition in the banking sector varies significantly across countries, as indicated by Lerner index previously. Average value of Boone indicator among nine Eastern European countries is -0.63. The estimated β are negative for all countries (except Slovenia), as expected: higher marginal costs are found in less efficient firms with lower profits. Results suggest that the most competitive markets are in Czech Republic (-3.01) and Romania (-1.26) and the lowest level of competition has been detected in Poland (-0.12). The model is statistically significant in almost all cases (except for Latvia, where p-value is 0.42). These findings differ from results found by the previous estimation of Lerner indices and other studies that use alternative measures of competition. For instance, recent findings by Bikker and Spierdijk (2009), suggest that, measured by H-statistic, competition is highest in Hungary, followed by Lithuania and Latvia and lowest in Poland and Slovenia. However, as mentioned previously, Carbó et al. (2006) find that generally competition measures are weakly correlated. The latter study assessed competition among 14 European countries using 5 different measures of market power, including the most popular ones: Lerner, Hirschman–Herfindahl indexes and H-statistic. It concludes that different indicators often produce conflicting predictions across countries, within countries, and over time. Several reasons explain this result: indicators measure different things and also some additional influence on the results comes from differences in GDP, inflation, cost efficiency and fee income levels.

Finally, when comparing Lerner and Boone indicators that have been obtained using the same data, quite a poor relationship can be found between them (see Figure 1). Thus, it is not surprising that the results differ, taking into account the previously discussed weaknesses of traditional measures that Boone indicator does not suffer from.

4.3. Bank stability

Bank soundness (at firm level) is measured by Z-score, which is inversely related to probability of failure. It increases with higher profitability (ROAA) and capitalization levels (ratio of equity to total assets), and is decreased by higher standard deviation of ROA, that is by unstable earnings. The estimated Z-score is presented in Table 3. Average value of the Z-score is 30.69. Berger (2009) reports an average value of 56.76, while Schaek and Čihák (2008 a) report an average Z-score of 27. The results, on the other hand, are quite similar to the study of Turk Ariss (2010), who finds that the average Z-score among selected Eastern European and Central Asian countries is 30.15. As shown in Table 3, lowest level of banking stability in this study has been found in Romania (16.36) and Hungary (27.34), whereas other observed countries had considerably similarly stable banking sectors (ranging from 31.30 in Poland to 36.28 in Slovakia). Bulgaria’s average is 40.48. Z-score is averaged over time for each country for inclusion into regression of Model 2.

Banking soundness dynamics are presented in Figure 2. As previously mentioned, stability for the whole region is reflected from the year 2002. Average banking sector Z-scores have been increasing during 2004-2007, while in 2008 they have slightly declined (however, mainly because Latvia’s banking system’s stability has decreased by about 50% as measured by Z-score).

4.4. Testing the relation between bank soundness and level of competition

In this section we present the results regarding the relationship between competition and stability in the Eastern European banking sectors.

Results for Breusch-Pagan/Cook-Weisberg and White tests for heteroskedasticity (see Table 4), which suggest the presence of heteroskedasticity in model 1. Hence, POLS regression is estimated with robust standard errors as it is one of the most common ways to deal with heteroskedasticity.

sign between market power and stability, the inflection points are calculated (point where the parabolic curve changes direction). This is done by setting the first-order derivative to zero and comparing the obtained value with empirical distribution of the data. For (1), (3), (4) and (5) regressions inflection points lie well above 99% of the data (99% percentile of Lerner index is 0.96). For (2) and (6) regressions inflection point is lies above 61% and 80% of all data. Because the sign of quadratic term of Lerner index is negative in all cases, estimated function is downward oriented and decreases after the inflection point. Therefore, less competition in the market might be desirable as it would decrease banks’ risk potential. In column (1) Lerner index enters only with bank-specific controls, while in column (2) macroeconomic control variables are incorporated as well (the effect of Lerner index on Z-score increases). Finally, in column (3) time and country dummies are included as well. The results remain robust when using Fixed and Random effects estimators (columns 4 and 5).

The findings for Eastern European countries support the traditional “competition-fragility” view and are in line with findings of Turk Ariss (2010), Beck et al. (2003, 2006), Jimenez (2007), Keeley (1990), etc. This also confirms the first hypothesis that the trade-off between competition and stability exists. Lerner index remains positive and significant after controlling for bank specific and environmental factors19.

Among control variables, larger banks (measured by natural logarithm of Total assets) tend to have lower Z-scores (2 and 3 regressions); however, this result is not significant at 95% significance level in the discussed cases. This has also been found by studies of Boyd (2006), Schaek and Čihák, (2008b); while Turk Ariss, (2010), Schaek and Čihák, (2008a) and Berger, (2009); find a positive relationship. Thus it remains unclear whether the process of consolidation (banks becoming larger) has had any direct impact on financial stability. Also, no support the view that regulators adopted a “too-big-to-fail” policy 20(Mishkin, 1999). The results for other bank control variables are as expected: banks with larger percent of loans tend to be fragile (loans to assets ratio is negatively associated with Z-score), banks that hold more equity tend to be more stable (equity to total assets ratio is positive) and higher ratio of loan loss provisions relative to total assets is associated with lower Z-scores.

As for macroeconomic variables, results for GDP are as expected: more developed countries have higher Z-scores (are more stable), while interest rate did not show the anticipated sign: countries with higher interest rates have higher Z-scores. Cross-country comparisons in banking sector are likely to be more accurate when these variables are included (Carbo et al., 2006).

19

Results also generally hold when standard deviations or all explanatory variables are included into regressions.

20

Table 4. Regression results (effect of Lerner index on Z-score)

Dependant variable: Z-score

POLS POLS POLS FE RE GLS

1 2 3 4 5 6 Lerner 38.07* 65.26** 37,91 8.26** 8.49*** 26.21*** (20,03) (28,20) (21,04) (3,87) (3,29) (8,02) Lerner 2 -19,23 -62.77*** -1,59 -2,27 -2,13 -19.94* (14,09) (22,95) (28,24) (5,39) (4,57) (11,35) Inflection point 0.99 0.52 11.92 1.82 1.99 0.66

Sign of the relationship + + + + + +

ln(Total Assets) 1,46 -0,28 -1,21 0,15 -0,05 1.94*

(1,05) (1,47) (2,43) (1,00) (0,79) (1,08) Loan loss provisions/Total

assets -1.17*** -1.27*** -1.16** -0,04 -0,04 -0.59***

(0,35) (0,40) (0,37) (0,05) (0,04) (0,12)

Loans/Total Assets -16,12 -23,47 -25,79 4,17 3,37 -0,83

(21,26) (22,67) (27,67) (3,63) (3,69) (6,61) Equity /total Assets 97.29*** 125.26*** 144.61*** 186.27*** 185.43*** 120.82***

(21,08) (23,48) (25,89) (31,47) (19,67) (14,07) ln(GDP) 16.36*** -1,01 -5,61 -4,09 12.13*** (6,10) (46,29) (5,98) (4,59) (3,28) interest rate 0,31 1,63 0,06 0,07 0.47* (0,80) (1,14) (0,11) (0,11) (0,25) Constant 10,92 -102.88*** 18,85 55,75 54,58 -114.00*** (21,68) (33,11) (424,72) (38,73) (33,29) (22,69)

Country dummies No No Yes No No No

Time dummies No No Yes No No No

Observations 721 721 721 721 721 721 R-squared 0,02 0,02 0,07 0,42 Wald’s _χ2 154,99 106,31 Probability > _χ2 0 0 Breusch-Pagan/Cook-Weisberg test _χ2 615.94 P-value 0.00 Wooldridge’s F test 26.14 Prob > F 0.00 Durbin-Wu-Hausman’s _χ2 2.69 P-value 0.26

Table 5. Regression results (effect of Boone indicator on Z-score)

Dependant variable: Z-score

POLS POLS POLS RE FE GLS

1 2 3 4 5 6 Boone -0,03 0,1 0,08 0.84*** 1.67*** -0.78*** (0,23) (0,21) (0,21) (0,30) (0,28) (0,18) ln(Total Assets) 5.54*** 3.35* 1,58 -1,64 -3.26*** (1,75) (1,81) (2,20) (3,19) (0,53) Loans/Total Assets -2,54 1,7 40.94** 72.06** 8.00** (13,28) (12,38) (19,29) (33,97) (3,18)

Equity /total Assets -1,13 4,47 -20,56 -43.39* 12,88

(17,23) (23,39) (20,92) (25,29) (9,44) Loan loss provisions/Total

assets -2.59*** -2.65*** -2.13*** -1.62*** -2.31*** (0,34) (0,37) (0,31) (0,44) (0,04) ln(GDP) 4.90*** 5.48*** 19.61* 8.23*** (1,46) (2,05) (10,36) (0,70) interest rate 0,29 0.96*** 1.51*** 0.21*** (0,19) (0,27) (0,54) (0,04) Constant 34.82*** -25,24 -39.97** -45.17* -142.26* 13.59** (1,12) (18,21) (16,30) (24,26) (74,19) (6,19) R-squared 0 0,11 0,11 0,13 Wald’s _χ2 64,78 9055,87 Probability > _χ2 0 0 Breusch-Pagan/Cook-Weisberg test _χ2 P-value 0.00 White _χ2 378.85 Prob > _χ2 0.00 Wooldridge’s F test 0.28 Prob > F 0.60

First stage F-test 13.39

Prob > F 0.00 Sargan’s _χ2 20.82 P-value 0.00 Durbin-Wu-Hausman’s _χ2 50.75 P-value 0.00

Note: Robust standard errors in parentheses; OLS – ordinary least squares, FE – fixed effects, RE – random effect, GLS – Generalized least squares estimator. *** p<0.01, ** p<0.05, * p<0.1.

POLS 21did not produce significant estimates of Boone indicator. However, it is positive when controlling only for bank-specific characteristics (column 2) and adding macroeconomic variables (column 3). Positive relationship between Boone indicator and Z-score means that banking sectors with higher values of Boone indicator (and lower level of competition) have also higher Z-scores. Columns (4) and (5) present fixed and random effects estimations. In this case, Boone indicator not only remains positively related with Z-score, but also it is significant at 99% level of significance. Hence, results for Model 2 also seem to support “competition-fragility view” and contradict findings of a number of studies 22(Uhde and Heimeshof, 2008; De Nicoló, 2000; De Nicoló et al., 2004; Schaeck and Čihák, 2008a, 2008b; Schaeck et al., 2009).

Regarding the control variables, results are not entirely as expected. In fact, they are sensitive to adding new variables or using a different estimator. Only ln(GDP) and the ratio of loan loss provisions to total assets have the anticipated sign (positive for level of economic development and negative for loan loss provisions). In contrast, the remaining variables exhibit different relations with Z-scores depending on the estimator used.

In order to further investigate the robustness of both models, other type of estimator is considered in order to deal with heteroskedasticity. In the presence of the former, POLS is still unbiased, but not the best estimator. The best and unbiased estimator in the presence of heteroskedasticity is Generalized Least Squares GLS (Hill et al., 2007). It changes or transforms the model into one with homoskedastic errors. The link between Lerner index and Z-score remains positive and significant. In contrast, Boone indicator enters the regression with a negative sign, and the effect is significant23. Furthermore, Wooldridge (2002) test for autocorrelation indicated that it is present in Model 1 but absent in Model 2 (see Tables 9 and 10). GLS estimator allows correcting both for heteroskedasticity and autocorrelation in model 1.

Thus, Model 1 strongly supports H1, which states that higher competition might impede banking stability. Although Model 2 generally supports this view, however, the effect of Boone indicator on soundness is not always significant and/or positive (some cases suggest a negative link). This can be explained by the fact that Boone indicator and Lerner index indeed measure different things. When existing firms switch to a more aggressive behavior, as discussed previously, Lerner

21

Breusch-Pagan/Cook-Weisberg test indicated the presence of heteroskedasticity, thus the regressions are estimated with robust standard errors.

22

Results also generally hold when standard deviations of all explanatory variables as regressors are included into regressions.

23

index is not able to capture changes in competition, while Boone indicator is. In other words, when using Boone indicator, which accounts for changes in competition more comprehensively (Schaek and Čihak, 2008b), results generally suggest that market power is negatively associated with stability, while increased competition only due to more banks entering the market strongly supports the view that this does impede financial stability in Eastern European countries.

A number of previous studies account for endogeneity of market power using 2SLS and GMM estimators, as market power might be endogenous because less stable banks tend to “gamble for resurrection”. They increase risky loan supply and as a result competition increases (Schaek and Čihák, 2008a). This study uses four candidates 24 for instrumental variables. In the case of Model 1, instrumental variable approach has not been found necessary (Durbin-Wu-Hausman Chi-sq test proposed that Lerner index is an exogenous variable, see table 4). Model 2, on the other hand, contains an endogenous variable (Boone indicator). The necessary tests for explanatory power (First stage F- statistics) and validity (Sargan’s Chi-sq) were performed (see Table 5); nevertheless, the results of Sargan’s test have failed to confirm the validity of the instruments. Hence, IV approach is not used in any of the models.

Finally, Tables 12 and 13 present the results for regressions where foreign ownership and EU membership dummy variables are added. In most estimations foreign ownership dummy variable enters with a negative sign, thus implying that foreign banks are less table than domestic ones. This could be explained by possibility that foreign banks might only provide limited products but also provide them only to home market, which in turn could result in a higher volatility of earnings (ROA) (Berger et al., 2009). However, Table 12 also suggests that generally the link between bank ownership and financial stability is not significant. The results are in line with findings of Beck et al. (2003, 2006); Berger et al (2009); Turk Ariss (2010). Hence, the third hypothesis, stating that foreign owned banks tend to have lower Z-scores, is not confirmed.

Results from Table 13 are mixed: membership in EU did not have significant effect on banking sector’s stability (columns 1-4) when Lerner index is used a measure of competition (results are very similar when implying a linear relationship between competition and stability). On the other hand, when competition is measured by Boone indicator, EU membership has a significant and positive effect on banking sector stability. Hence, EU membership might have had a positive impact on new members’ banking sector stability. Nevertheless, this issue requires a more thorough study, possibly revising the model. Also, the results might not be accurate because the majority of the

24