Does foreign bank penetration affect the risk of domestic banks? Evidence from emerging economies

Transcription

1 Does foreign bank penetration affect the risk of domestic banks? Evidence from emerging economies Minghua Chen Research Institute of Economics and Management Southwestern University of Finance and Economics Chengdu, China Ji Wu Research Institute of Economics and Management Collaborative Innovation Center of Financial Security Southwestern University of Finance and Economics Chengdu, China Bang Nam Jeon School of Economics, LeBow College of Business Drexel University, PA, USA Rui Wang Research Institute of Economics and Management Southwestern University of Finance and Economics Chengdu, China Abstract We investigate whether foreign bank penetration affects the risk-taking of domestic banks in emerging economies. By using bank-level data from 35 markets during the period of , we find significant evidence that the riskiness of domestic banks increases with the presence of foreign banks, and this finding is shown to be consistent in a series of robustness examinations. We also explore various conditions for the heterogeneity of the nexus between foreign bank penetration and domestic banks risk-taking, including: (1) what types of domestic banks are affected more by the presence of foreign banks, and (2) what patterns of foreign penetration exert more pronounced impact. Keywords: Foreign bank penetration, bank risk-taking, emerging economies JEL classification: G21; G15; F36; E44 This version: May 2017 Corresponding author: Bang Nam Jeon, School of Economics, Bennett S. LeBow College of Business, Drexel University, 33and Market Streets, Philadelphia, PA 19104, USA. chenminghua@swufe.edu.cn (M. Chen), wuji@swufe.edu.cn (J. Wu), jeonbana@drexel.edu (B. N. Jeon), wangruiswufe@gmail.com (R. Wang). 1

2 1. Introduction With the process of financial deregulation and liberalization, many emerging and developing countries witnessed a significant restructuring of their banking sector since the 1990s, characterized by a considerably higher presence of foreign banks. Claessens and Van Horen (2014) document that the number of foreign banks increased by 74% and their market share approximately doubled in emerging countries during the period of Considering the economic might and size of the banking industry in host countries, foreign bank penetration is notably more salient in regions such as Central and Eastern Europe, Latin America and Asia. 1 Appendix A and B provide a summary on the extent by which foreign banks are present in selected emerging markets in the above three regions, respectively, measured by the assets (number) of foreign banks as a share of the total bank assets (number) in the host banking sector. 2 The expansion of bank capital across national borders raises interesting questions about the roles played by foreign banks, in particular, whether allowing foreign participation would introduce more stability or volatility into the host banking markets. The answer to this question is not only important for economists to better understand the advantages and disadvantages of financial globalization, but also bears relevant policy implications for regulatory authorities to maximize the net benefits from the opening up their banking sector. Albeit a rich wealth of literature studying the economic impacts of foreign bank penetration on host markets, extant works are yet to arrive at a consensus. The supportive views suggest that the entry of multinational banks stabilizes credit quantity during domestic financial turbulences, fosters the overall efficiency of financial intermediaries and helps small and medium sized enterprises have more access to credit (De Haas and Van Lelyveld, 2006; Goldberg, 2007; Lehner and Schnitzer, 2008; Havrylchyk and Jurzyk, 2011a; Bruno and 1 Foreign banks are also prominent in the banking markets of Central Asia and Sub-Saharan Africa (Claessens and Van Horen, 2014). 2 There are significant heterogeneities across the three regions in terms of foreign bank penetration. Figure 1 and 2 depict the average levels of foreign bank penetration in all selected emerging economies, and that in Central and Eastern Europe, Latin America and Asia, respectively. Central and Eastern Europe is characterized by the highest foreign bank presence on average. More than 70% of total bank assets are possessed by foreign banks and nearly 60% of banks are foreign owned as of 2014 in the region. The average foreign penetration ratio in Latin American countries lies at 36%-47% (47%-53%) in terms of bank assets (bank number). In comparison, foreign banks have only a modest presence in Asian markets, where the average penetration level is 24% (40%) in terms of total bank assets (bank number) at the maximum. The level of foreign presence also varies over years. Consistent with the observation of Claessens and Van Horen (2014), the average market share of foreign banks rose steadily during the period of , but declined in the wake of global financial crisis, driven by the contraction of foreign banks in Central and Eastern Europe and Latin America. 2

3 Hauswald, 2014). The cautionary views, however, warn that foreign banks can be a new source of instability by transmitting external shocks, weakening the potency of domestic monetary policy, and curtailing their credit more greatly when there is an adverse shock in their home country (Goldberg, 2001; Clarke et al., 2003; Arena et al., 2006; De Haas and Van Horen, 2012; De Haas, 2014). How foreign bank presence affects domestic banks performance remains a question that is only partially answered. Extant works mostly center on aspects such as banks profit, net interest spreads, operational cost, efficiency and credit growth (Dages et al., 2000; Claessens et al., 2001; Unite and Sullivan, 2003; Martinez Peria and Mody, 2004; Xu, 2011; among others). For example, Claessens et al. (2001) find that an increased foreign bank presence is associated with lower profitability, non-interest income, and overhead expenses of domestic banks. Barajas et al. (2000) and Unite and Sullivan (2003) find that foreign banks contribute to lower interest rate spreads of domestic banks, but Martinez Peria and Mody (2004) find no supportive evidence. Gormley (2010) documents a market-wide reduction on the loan volume of domestic banks after the entry of foreign banks. Nevertheless, in comparison to the extensive literature studying the impact of foreign entry on the above aspects of domestic banks, whether and how foreign bank presence affects domestic banks risk is investigated by much fewer works. 3 Whether domestic banks risk-taking is affected by foreign bank penetration is ambiguous theoretically and empirically. Some research suggests that the entry of foreign banks may benefit domestic banks stability by spurring a sophistication of domestic financial regulations, rendering a spillover of know-how and expertise from foreign banks to domestic ones, diversifying domestic banks product portfolio, or stimulating domestic banks to increase investment on modern technology and human capital and thus improve their efficiency in the long-run (Levine, 1996, 2001; Lensink and Hermes, 2004; Goldberg, 2007; Kouretas and Tsoumas, 2016). Were these favorable impacts predominant, the soundness of domestic banks is expected to be bolstered. However, there are a number of competing forces that may offset the above beneficial impacts of foreign penetration and lead the risk of domestic banks to increase. First, domestic banks may be adversely affected by the shift of customers after the entry of foreign banks. On one side, foreign banks may focus their credit and other financial services on informationally transparent clients and crowd their domestic counterparts out of this market niche 3 A reduction in profit (or net interest spreads) can be suggestive of an increase of bank risk. However, it is a less proper indicator for the probability of bank failure, since the beneficial effect of higher profit could be offset by the increase of indebtedness and the volatility of profit. As explained later, the Z-score, a composite index that combines banks profitability, leverage and the volatility of profit, is widely adopted as a better measure of the risk of bank failure. 3

4 ( cherry-picking ), leaving only opaque firms to the latter (Dell'Ariccia and Marquez, 2004; Sengupta, 2007; Gormley, 2014). If opaque firms are characterized by lower creditworthiness, and domestic banks advantage on soft information cannot sufficiently shield themselves from borrowers defaults, the asset quality of domestic banks would deteriorate consequently. 4 On the other side, depositors may transfer their savings out of domestic banks but into foreign banks because of the latter s superior service and international reputation ( flight-to-quality ), causing domestic banks to incur higher cost to either attract more deposits or substitute deposits with other sources of funding. Corresponding to higher cost of liabilities, domestic banks increase their lending rates, which may trigger the problem of adverse selection (Mian, 2003; Rashid, 2011). Due to pressure from both sides, domestic banks fragility may increase with the expansion of foreign banks. This effect is probably more profound in less developed countries, because of the limited flexibility of domestic banks to adjust their portfolio and then diversify risk, than the banks in developed countries (Hermes and Lensink, 2001). Second, competition may increase as foreign banks establish their business in host markets (Claessens and Laeven, 2004; Jeon et al., 2011). Traditional theory posits that higher franchise value would limit banks incentive to take excessive risk (Keeley, 1990; Demsetz et al., 1996). However, if foreign bank entry is associated with a higher market competition, it can reduce banks franchise value due to lower profitability, and thus its constraint on banks risk-taking tends to be weakened (Claessens and Lee, 2003; Jiménez et al., 2013). 5 Suggested by the competition-fragility hypothesis (Beck et al., 2006; Berger et al., 2009), more intensive competition leads to lower net interest margins, eroding the major source of bank profit and thus inducing more risk-taking behaviors to search for yield. Dell'Ariccia and Marquez (2006) argue that increased competition from foreign bank entry induces existing domestic banks to relax screening of loan applications to retain their market share and thereby worsen the quality of their asset portfolio. 6 Third, in order to secure their market shares, domestic banks may follow their foreign competitors in providing new services (Xu, 2011), which nevertheless may increase their 4 De Haas and Naaborg (2006) and Beck et al. (2014) document that foreign banks are never limited on transactions lending to transparent clients, but develop their own relationship lending to opaque firms, which may squeeze the market for domestic banks, reduce their earnings and increase their risk. 5 Some works, for example Martynova et al. (2014), suggest that the relationship between franchise value and bank risk-taking needs to be further explored. High franchise value allows a bank to borrow more, thus higher leverage may offset the lower incentives of risk-taking. 6 The competition-stability view argues that competition may strengthen banks stability since more intense competition would reduce the market interest rates and thus lowers borrowers probability to default (Boyd and De Nicoló, 2005). If this effect outweighs the competition-fragility impact, ceteris paribus, domestic banks are expected to be associated with lower risk when there is an increase in the foreign bank participation. 4

5 operational costs and lead to higher risk if the foreign banks own comparative advantage on these services. Analogously, foreign bank entry can compel domestic banks to increase investment into cutting-edge technology and employee training. However, the increased outlay is immediately translated into higher costs but the gains may take some time to emerge. Subsequent losses are, therefore, likely at least in the short-run, even though the supply of new services, together with the investment on modern technology and employee training, may increase domestic banks efficiency and strengthen their stability in the long-run. Existing literature provides very limited empirical results on the impact of foreign bank penetration on domestic banks risk. Unite and Sullivan (2003) indicate an increase of loan loss provision of domestic banks with the expansion of foreign banks, while this association is found insignificant by Claessens et al. (2001). Using non-performing loan ratio as the indicator of loan quality, Barajas et al. (2000) and Degryse et al. (2012) find some evidence that foreign bank entry undermines the soundness of domestic banks. However, Angkinand and Wihlborg (2010) and Agoraki et al. (2011) document only mixed results that a higher foreign bank presence is associated with a higher/lower riskiness of the entire banking sector. The competing theories and limited empirical evidence lead us to explore in this paper whether and how foreign bank penetration affects the riskiness of domestic banks. Our paper differs from existing works in several ways. First, many papers discuss the effect of foreign bank penetration on host financial stability from the perspective of credit quantity, but not credit quality. Foreign banks are usually suggested as a stabilizing (destabilizing) force when the amount of their lending is less (more) volatile than that of domestic banks (Martinez Peria et al., 2002; De Haas and Van Lelyveld, 2006; De Haas and Van Horen, 2012). However, we refer the stability role played by foreign banks by their impact on credit quality. We find that, after having controlled for other potential determinants, domestic banks stability decreases when foreign bank penetration increases, implying a deterioration of domestic banks riskiness. Our result suggests a potential tradeoff between the credit quantity of foreign banks and the credit quality of domestic banks, which can be particularly important for host financial authorities to fully estimate the outcomes of banking sector openness. Second, when addressing the impact of foreign penetration on the domestic banking sector, a conventional practice in earlier works is to study the performance of foreign banks relative to domestic ones by distinguishing these two groups of banks, usually using a foreign/domestic dummy (for example, Berger et al. (2005) and Claessens and Van Horen (2012)). The behaviors of these two groups of banks are implicitly premised as independent. In our paper, under a different presumption, we focus on examining the variation of domestic banks risk and identifying the impact attributable to the penetration of foreign banks. We believe this analysis has a complementary value to the enrichment of extant literature by helping explore if there is connection between the market participation of foreign banks and 5

6 the risk-taking of domestic banks. Moreover, after finding consistent evidence that there is a negative association between foreign bank presence and domestic bank stability, we further explore conditional factors affecting the heterogeneity of this nexus, which are only understudied so far in prior literature. We ask first what types of domestic banks are affected more by foreign bank penetration, and find that the impacts across banks are varying with different bank efficiency, business diversification and private/state ownership. However, banks with any size are affected by the entry of foreign banks. We next investigate what patterns of foreign bank presence exert more pronounced effects. To be specific, we examine whether the different entry modes of foreign banks, the proximity of home country to host market, and the strength of intragroup internal capital markets would matter for the impact of foreign bank penetration on domestic banks. Besides adding to the literature on the economic impacts of foreign bank penetration, our paper also contributes to a growing strand of research on the impact of financial liberalization on the risk-taking of financial institutions. A number of works have warned of some negative effects of financial reforms and capital account openness, from the perspective of either bank risk-taking or profit efficiency (Demirgüç-Kunt and Detragiache, 1998; Cubillas and González, 2014; Luo et al., 2016). As a part of financial liberalization, allowing the participation of multinational banks in host markets is also found to be associated with some undesired effects in this paper. The rest of the paper is organized as follows. Section 2 introduces the data and the construction of main variables. Section 3 describes our econometric model. Section 4 reports the baseline results and the robustness tests. In Section 5, we examine the differential effects of foreign bank entry on different types of domestic banks. Section 6 reports our findings on the heterogeneous impacts associated with the different patterns of foreign bank penetration. Section 7 concludes. 2. Data and variables We build an unbalanced bank-level panel with annual data from 35 emerging economies located in Central and Eastern Europe, Latin America and Asia during the period of Only commercial banks are included in our sample to minimize any possible bias due to the different nature and business scope among banks that have different objectives and conduct businesses in different specializations. In order to avoid selection bias, we 7 Specifically, the selected emerging markets include: Albania, Belarus, Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Moldova, Macedonia, Poland, Romania, Slovakia, Slovenia, Ukraine (Central and Eastern Europe); Argentina, Bolivia, Brazil, Chile, Colombia, Mexico, Paraguay, Peru, Uruguay, Venezuela (Latin America); China, Hong Kong SAR, India, Indonesia, Korea, Malaysia, Pakistan, Philippines, Singapore and Vietnam (Asia). 6

7 include not only existing banks, but also those that have ceased business operation. We collect the data used to measure banks risk and individual banks characteristics from Bureau van Dijk s Bankscope database Bank risk Three Z-score based indicators are employed to proxy the riskiness of domestic banks. Using multiple measures not only exhibits more dimensions of banks financial soundness, but also strengthens the robustness of our finding. Meanwhile, using the Z-score based indicators also differentiates our paper from some earlier ones that adopt the ratio of loan loss provisions or the non-performing loan ratio as the measure of bank risk. As commonly used in extant literature (for example, Laeven and Levine (2009), Demirgüç-Kunt and Huizinga (2010) and many others), our primary indicator of the riskiness of domestic banks is the time-varying Z-score, which is formally expressed as: Z ijt ROA EA ijt ijt (1) ( ROA ) ijt where ROA ijt denotes the return on assets of bank i in country j in year t, EA ijt is the ratio of equity over total assets, and σ(roa) ijt is the standard deviation of return on assets. We follow Schaeck and Cihák (2010) by using a three-consecutive-year rolling window to calculate σ(roa) ijt, rather than the full sample period. 8 Interpreted as the number of standard deviations by which returns must decrease to wipe out all equity owned by the bank (Roy, 1952), the Z-score can be viewed as the inverse of the probability of bank failure. A higher value of the Z-score suggests a higher financial stability of the bank, or put differently, a lower exposure to insolvency risk. 9 Since this Z-score is only calculated based on banks own return on assets and equity-assets ratio, we view it as a measure of the absolute risk-taking of banks. However, a simple comparison between the values of Z-scores across different countries may lead to biased conclusions, since it can be argued that banks Z-scores in some countries may be in general higher or lower than those in other countries, thus a higher figure 8 We also experiment using a five-year rolling window to calculate Z-scores and find that our main results do not change and remain statistically significant. However, using a five-year rolling time will cause a considerable reduction in the number of our observations. 9 Because the Z-score is highly skewed, we apply the natural logarithm to (1+ Z-score) to smooth higher values (Beck et al., 2013). Using 1+ Z-score instead of using simply Z-scores is to avoid the truncation of the Z-score at zero. We denote ln(1+ Zscore) as the Z-score in the latter part of the paper for brevity. Prior to our calculation of the Z-score, we removed the outliers of ROA ijt and EA ijt above the 99th percentile and below the 1st percentile of the sample distribution to rule out abnormality or probable measurement errors. 7

8 of Z-score for Bank A in country 1than that for Bank B in country 2 may not necessarily mean that Bank A is placed at a position relatively less risky than Bank B in their own country. In order to overcome this problem, we normalize Z-scores for each country respectively as follows: Z _ n i jt Zi jt min( Z jt ) max( Z ) min( Z ) jt jt for country j=1, 2 (2) where min(z jt ) and max(z jt ) respectively denote the minimum and maximum value of Z-scores for all banks, including both domestic and foreign ones, in country j over the sample period. The results thus lie in the rage of [0, 1], suggesting the relative level of riskiness that banks are exposed to in their markets. A higher value in Z_n suggests the bank has a relatively greater stability/lower insolvency risk in comparison to its counterparts across countries. We interpret this indicator as reflecting banks relative riskiness. Nevertheless, it can be argued that the banks current stability may be deviated from the potential maximum stability that they can achieve, given the different asset portfolios that banks choose to produce. Thus, we borrow the concept of X-efficiency of stability from Fang et al. (2014) and Tabak et al. (2012), who assume Z-scores as the outcome of banks production choice under the trade-off of return and risk, and suggest that the same Z-scores may be associated with banks different deviation from their implicit highest financial stability. We estimate the X-efficiency of banks financial stability by applying the stochastic frontier approach (SFA) to the following production function: Z 1 1 w 1 w w ln( ) c ln( y ) ln( y ) ln( y ) ln( ) ln( ) ln( ) w w w w m m n ijt h h ijt hk h ijt k ijt m ijt mn ijt ijt 3 2 h 1 2 h 1 k 1 m m 1 n wm 1 1 wm hmln( y h ) ijt ln( ) ijt ln EQijt i ln EQijtln( y h ) ijt m ln EQ ijtln( ) ijt 2 h 1 m 1 w3 2 h 1 2 m 1 w3 2 1T 2T ijt ijt ijt ijt (3) u (4) where y i represents three bank outputs, namely, loans (y 1 ), securities (y 2 ) and other non-interest related operations (y 3 ), w i denotes three input prices, i.e., price of funds (w 1 ), price of fixed capital (w 2 ), and price of labor (w 3 ), and EQ is banks equity, which is included as a netput. T denotes the time trend. 10 The error term, ε ijt, is distinguished into two parts. The 2 first part, u ijt, denotes the random noise, assumed normally distributed ( u ~ N( 0, )), and ijt u 10 We assume a standard production function by following earlier literature. w 1 is proxied by the ratio of interest expenses to total deposits and other funds, w 2 is measured by the ratio of (overhead cost personnel expenses) to fixed assets, and w 3 is calculated by the ratio of personnel expenses over total assets. The normalization by w 3 ensures price homogeneity. 8

9 represents the measurement errors and other uncontrollable factors. The second part, ν ijt, 2 assumed half-normally distributed ( ijt ~ N ( 0, ) ), captures the banks inefficiency to conduct a production that can render an optimal financial stability. Estimating a single frontier for all banks across countries allows the X-efficiency item, ν ijt, to be compared against the same baseline (Tabak et al., 2012). We use the method of Battese and Coelli (1995) to estimate equation (3) and then adopt the Battese and Coelli (1988) estimator to convert ν it into Z_ν ijt = E(exp( ν ijt ε), a term with a similar pattern to Z and Z_n, where a higher value in the range (0, 1) denotes a closer distance to the implicit greatest stability. Given banks different asset portfolio and input prices, a high value in Z may or may not be associated with a high Z_ν. We interpret Z_ν as the excessive risk-taking level of banks Foreign bank penetration In order to measure the degree by which foreign banks are present in host markets, we first need to distinguish foreign banks from their domestic counterparts. In line with the common practice of related works, we define a bank as foreign owned if more than 50% of its capital is held by foreign banks, firms, individuals or organizations. We track the year-by-year domestic/foreign ownership status for each bank in our sample by taking the following steps: First, we check the brief overview of banks documented in Bankscope, which records the ownership information for some banks in the most recent year. Second, we visit banks website to review their historical profile, where important events, such as the establishment and the change of controlling shareholders, are usually documented. Third, we obtain banks mergers and acquisitions (M&A) information from another comprehensive database, the SDC Platinum, which provides relevant information on cross-border banking M&A, including the time and the identity of acquirers. Finally, we resort to various other information sources, such as banks annual reports, central banks publications and news reports from the Internet, to identify the ownership status for remained banks. 12 We measure the level of foreign bank penetration by using two proxies as the standard practice in the literature (Claessens et al., 2001). The first measure is the assets owned by foreign banks as a share of the banking sector total assets, denoted as Pene_assets. The second measure is the number of foreign banks as a proportion of the number of domestic and foreign banks in the host market, represented as Pene_number. 13 Although highly correlated, 11 We lose a large number of observations when estimating Z_ν because of the limited data for some variables. We also experiment estimating Z_ν in each country separately but unfortunately it fails to be implemented in many countries due to the deficiency of observations. 12 At the end we identify 935 domestic banks and 755 foreign banks. 13 It is acknowledged that our measures of foreign bank penetration, like those in many earlier works, are subject to some drawbacks due to the limitation of data. First, virtually only foreign subsidiaries are 9

10 there are different implicit assumptions behind the uses of these two measures. Pene_assets can be a proper measure of foreign penetration if foreign banks exert pressure on domestic banks only if foreign banks are sizable in host markets. For example, the hypothesis that foreign banks may introduce greater competition would be legitimate when they possess a significant share of the market and compete directly with their domestic peers. The use of Pene_number assumes that domestic banks can be affected by the mere presence of foreign counterparts. Foreign banks may probably cherry-pick the scarce best clients or cause a flight-to-quality without having to take over a large segment of the financial market. It is also possible that some domestic banks, such as those in areas remote to where foreign banks are concentrated, may change their behavior amid a potential expansion of foreign banks, even though there is no head-to-head competition yet. However, it is important to note that we do not argue that Pene_assets (Pene_number) captures only the direct competition effect (the cherry-pick / flight-to-quality / the threat of potential entry) of foreign banks. The channels via which foreign banks affect domestic ones may be de facto intertwined with each other, i.e., the cherry-pick associated with foreign entry may also encourage more fierce market competition. Unfortunately, it is empirically difficult to distinguish the clear-cut channels and measure each of them by a particular indicator, thus both Pene_assets and Pene_number may be associated with multiple effects of foreign bank penetration Bank characteristics, macroeconomic conditions and financial regulation Our other control variables are based on a careful review of extant literature on the potential determinants of bank risk. We firstly control for a series of bank characteristics denoted respectively as Size, Liquidity, Growth rate of assets, Income diversification, Funding diversification and Age. Size is a bank s assets as a share of the banking sector total assets. It reflects banks relative scale in their banking markets. 14 Large banks, while owning more advanced risk management skills, may behave in the fashion of moral hazard if they presume they are too big to fail (Afonso, et al., 2014). Liquidity represents the ratio of liquid assets over total assets for individual banks. The abundance of liquid assets may shelter banks from captured by our dataset, since only banks that are incorporated as separate corporate entities publish their own annual reports. As a matter of fact, many multinational banks choose to enter a market abroad by establishing branches without independent status. Second, the foreign shares in domestically controlled banks are not counted. For example, a bank with 51% of capital possessed by domestic owners and 49% by foreigners is still defined as a domestic bank. Unfortunately, it is almost practically impossible to track the specific shareholding structure for such a large pool of banks over more than a decade. Both of these two problems may cause the foreign presence level in emerging economies to be understated. 14 We also tried using the absolute size of banks, measured by the logarithm of bank assets in millions of constant US dollars. Our results are qualitatively unchanged. 10

11 unexpected monetary shocks and deposit runs, yet it is also likely that banks store more liquid assets when they foresee a higher volatility on returns (Alger and Alger, 1999). We also control for the growth rate of banks real assets in terms of constant US dollar, denoted by Growth rate of assets, since credit risk can build up when banks expand their balance sheet too aggressively (Foos et al., 2010). Income diversification and Funding diversification, included as Demirgüç-Kunt and Huizinga (2010), are respectively measured by non-interest income as a share of total operating income and non-deposit short-term funding as a share of the total short-term funding. Conventional wisdom postulates that a higher extent of diversification will translate into lower bank risk and stabilized returns, but many empirical works find conflicting evidence (for example, Stiroh, 2004). Age, which indicates the logarithm of historical length (in years) since a bank has been incorporated into the market, is added to allow for the possibility that there is a start-up period until banks can obtain a desired level of stability after their operation is commenced. We also include the square of bank age, Age 2, since banks risk may subsequently accumulate with time due to the increased managerial complexity and red-tape that are associated with the history of their business. We secondly include a group of variables for macroeconomic conditions in our regressors, namely, GDP growth rate, Inflation, Monetary policy and Crisis. The first is the growth rate of GDP, adjusted by using GDP deflator and the second is the inflation rate based on CPI. These two variables capture the impact of business cycles on financial stability (Marcucci and Quagliariello, 2009). We control for Monetary policy that is proxied using the first order difference of short-term interest rates in each country, where a positive/negative outcome is interpreted as an eased/tightened monetary innovation. Suggested by a growing body of works on the bank risk-taking channel of monetary policy, expansionary monetary policies may increase banks tolerance to risk and/or encourage more behaviors of search for yield, thus increase the riskiness of banks (Borio and Zhu, 2012). The data of above variables are drawn from IMF s International Financial Statistics Database. Since banks would usually incur higher risk during crisis periods, we include in our estimations a binary dummy variable, Crisis, for the episodes of financial crises in emerging economies. Data for crisis periods are selected from Laeven and Valencia (2013). 15 As confirmed by rich evidence presented in as Barth et al. (2004, 2008) and Laeven and Levine (2009), financial regulatory rules are an important factor to affect the stability of the banking sector. We therefore control for the regulatory strength from four aspects, specifically, 15 We assume that financial sectors in all countries are affected by the global financial crisis and let this crisis dummy be equal to one for all countries in We also extend the database in Laeven and Valencia (2013) since it only covers the crises up to

12 the strictness of regulations on capital adequacy (Capital), the restriction on banks activity mix (Activity), the authorities owned by supervisory agencies to intervene banks structure and operation (Supervisory power) and the extent to which banks are exposed to private monitoring and public supervision (Market discipline). Using the survey data provided by Barth et al. (2004, 2008, 2013) and following the methodology suggested by Barth et al. (2004), we build country-level time-series indices for each of the above four regulation aspects for each emerging economies in our sample. A higher score in these indices denotes more stringent regulations Other control variables There are only mixed results in extant literature regarding the impact of market structure on bank soundness (Boyd and De Nicoló, 2005; Beck et al., 2006). We use the Herfindahl-Hirschman Index (HHI), measured as the sum of the squares of individual bank s market share in total banking assets, to proxy the concentration level of host markets. A higher value of HHI indicates that the banking market approaches higher consolidation. A long list of literature has assessed extensively the efficacy of deposit insurance systems on financial stability (Keeley, 1990; Demirgüç-Kunt and Huizinga, 2005). Deposit insurance, designed as a safeguard against bank runs, has also been attributed as a source of moral hazard. Using the data from Demirgüç-Kunt et al. (2013) and following Barth et al. (2004), we construct a composite index, Deposit insurance, by summing up the design features of deposit insurance schemes, such as the coverage limit as a share of GDP per capita, the source of funding, the compulsoriness of membership and others, to measure the strength of the deposit insurance coverage. We also control for Financial depth, measured by the ratio of aggregate deposits over GDP, as a potential determinant of the risk-taking levels of banks. On one side, a higher financial depth could imply a higher sophistication of the banking sector, while on the other side it may also reflect the credit constraints faced by bank clients. Accordingly, the degree of financial depth is expected to impose ambiguous impacts on the stability of banking markets. At last, as La Porta et al. (1998) and many others have argued, institutional environments, including the effectiveness of contract enforcement and the legal protection on creditors, also affect financial development significantly. Following the literature of law and finance, we include Rule of law, as an indicator of the quality of institutions, in our 16 For instance, the index of capital regulations is based on the answers to 9 survey questions such as: whether the minimum capital-asset ratio requirement is risk-weighted in line with the Basel guidelines, whether the minimum ratio varies as a function of market risk, whether the sources of funds to be used as capital are verified by the regulatory authorities, and others. Summing up the answers (1 for yes and 0 for no ) yields a value that denotes the strictness of regulations on the capital requirement. 12

13 regressions. The data are borrowed from the rule of law index in the World Bank s Worldwide Governance Indicators (Kaufmann et al., 2010) Descriptive statistics The definition of variables and the sources of data are presented in Table 1. We also report the main descriptive statistics of these variables. 17 The Z-score of domestic banks is distributed with the mean value of and the standard deviation of Although not reported due to the limited space, Z-scores are ranged between the minimum and the maximum The fairly high standard deviation and the wide range of Z-scores highlight a substantial variation on the level of riskiness across banks. As expected, the mean value of the other two stability indicators, Z_n and Z_ν, is approximately 0.5 since the range of these two indicators is between 0 and 1. This observation seemingly suggests that, domestic banks and foreign banks have comparable riskiness on the whole. With regard to foreign bank penetration in terms of bank assets, the mean value of Pene_assets is and its standard deviation is 0.266, indicating a considerable heterogeneity on the presence of foreign banks across emerging markets. In comparison, the foreign bank presence in terms of number, Pene_number, is relatively less varied, with the mean of and the standard deviation of The mean value of foreign bank penetration level in our sample is largely affected by some countries that own a large number of banks but relatively modest presence of foreign entrants, for example, China and Brazil. 18 We also report the pairwise correlations between key variables in Appendix C. The correlations between the Z-score of domestic banks and both foreign penetration measures are negative and statistically significant. This fact indicates higher risk-taking by domestic banks in markets with more prominent foreign bank participation. The bank characteristic variables, and the financial regulatory variables, are found not highly correlated with each other, implying that a joint inclusion of these variables will not cause serious multicollinearity problems. Probably because that foreign banks consider host economic difficulties as an opportunity to seize more market share, either through new acquisitions or extending outstanding credit lines, the correlation between the presence of foreign banks and real GDP growth rate is found negative (Crystal et al., 2002; Althammer and Haselmann, 2011) For the Z-score and the bank specific characteristics, except Age, we delete the observations that lie below the 1 st percentile and above the 99 th percentile of the sample distribution in order to rule out the possible impact of outliers. 18 Ruling out the countries with the largest amount of observations, i.e. China and Brazil, does not change our results. The estimated effect of foreign bank penetration and its statistical significance are found increased without these two countries. 19 Another possible reason is that foreign bank presence is still modest in Asia, a region of rapid 13

14 Furthermore, foreign banks are seemingly more inclined to enter a market with laxer financial supervision, lower market competition and financial depth, higher coverage of deposit insurance and stronger rule of law. This fact justifies the necessity to control for these variables to better distinguish the impact of foreign bank penetration. [Table 1] 3. Model Our baseline econometric model is described as follows: Risk c Penetration Char Macro Regu Other f ijt jt ijt jt jt i ijt (5) where the dependent variable, Risk ijt, is our indicator of the financial riskiness of domestic banks, i.e., Z, Z_n, and Z_ν respectively. Penetration jt reflects the degree of foreign bank penetration in each country over years, in terms of the assets (number) of foreign banks as a share of the banking sector total assets (number). Char ijt, Macro jt and Regu jt represents the series of bank characteristics, macroeconomic conditions and the proxies for bank regulatory rules, respectively. Other is the vector containing the variables for market concentration, the strength of deposit insurance coverage, financial depth, the rule of law and year dummies. f i is the time-invariant bank-specific effect and ε ijt is the idiosyncratic error.,,, and are the coefficients to be estimated. To mitigate the problem of endogeneity, we use the one-year lag of each of the bank characteristic variables, except banks age. The benchmark model is estimated by using the fixed-effects estimator, which is chosen based on the Hausman test that suggests the fixed-effects estimator is preferable to the random-effects estimator because the regressors are shown correlated with the time-invariant bank-specific variables. We use heteroskedasticity and within-panel serial correlation robust standard errors in estimations. 20 To check the robustness of our main results, we also employ various alternative econometric methodologies later. 4. Results 4.1. Baseline results We report the estimation results for our baseline model in Table 2. The columns (1)-(6) differ with each other by using different dependent variables, i.e. Z, Z_n and Z_ν respectively economic growth in past decades. However, even as we experimentally exclude the observations from Asia, the correlation between foreign bank penetration and economic growth is still negative. 20 Alternatively, we use the number of observations for each bank as the weight of our data and find that our results are not changed qualitatively and their statistical significance remains. The results are available upon request. 14

15 and different foreign bank penetration measures, namely, Pene_assets and Penn_number. [Table 2] First, we find that foreign bank penetration is inversely related to the Z-score based indicators. The coefficients on both Pene_assets and Pene_number are negative and statistically significant in most regressions and only marginally not in the other. Since a higher Z-score suggests more stability and less risk-taking, this result is interpreted as that in general the riskiness of domestic banks increases amid a higher presence of foreign banks. The higher risk profile associated with domestic banks is also evidenced by the decline of their relative stability position vis-à-vis foreign banks, when using Z_n as the dependent variable. The results of Z_ν seemingly imply that domestic banks would allocate their resources less optimally, whereby their current stability is further deviated from the implicit maximum stability. Our findings are in line with the hypothesis that, for domestic banks, the risk-increasing effect attributable to foreign participation outweighs its potential beneficial impact, presumably due to a shift of customers, more intensive competition or disadvantages on innovative financial services. Quantitatively, the impact of foreign penetration is also salient. Using the result in column (1) as an example, domestic banks riskiness tends to increase by 0.86 percent for each percentage that foreign banks increase their market share. Alternatively speaking, were foreign bank presence to increase by one standard deviation (0.266, or alternatively speaking, 26.6 percent), the average stability of domestic banks would decrease by nearly 23 percent in response. Second, we also find some interesting results with regard to the risk impact of other control variables. The growth rate of bank assets is negatively related to the stability of banks, and the result is statistically significant when using Z and Z_n as the dependent variable. It is consistent with prior literature which reports that banks would incur higher risk when they expand loans too aggressively. Nevertheless, our results provide no clear evidence on the impact of income diversification on banks risk since its coefficient is negative when Z and Z_n are regressed while it becomes positive in other regressions. Age and Age 2 are both found to affect bank risk significantly but with opposite signs. The financial soundness of banks tends to strengthen after banks commence their business, probably thanks to the proprietary information obtained in the process of lending, but turns sour if inefficiency and red-tape develop with the continuity of bank operation. We find evidence in favor of the pro-cyclicality of financial stability. The coefficient on real GDP growth rate is positive and highly significant in all regressions, implying a lower risk incurred by banks when the economy is booming. Nevertheless, in line with the growing literature on the risk-taking channel of monetary policy, banks undertake more risk when central banks adopt an expansionary monetary policy, as the positive coefficient on the monetary policy indicator suggests (Borio and Zhu, 2012; Delis and Kouretas, 2011). 15

16 Regulatory rules are found to matter for financial stability but in seemingly different directions. Banks in countries with stricter regulations on capital adequacy and stronger market discipline are characterized by a lower risk profile than their peers in other regions, whereas a more stringent limitation on banks activity mix and a greater authority owned by supervisory officials only create undesirably higher risk to banks. These findings are consistent with works such as Barth et al. (2004) and Laeven and Levine (2009). Additionally, deposit insurance coverage shows a negative and statistically significant impact on the soundness of domestic bank. This result is supportive for the argument that more generous deposit protection exacerbates banks moral hazard problem and fuels their incentive to take higher risk (Keeley, 1990; Demirgüç-Kunt and Huizinga, 2005). However, in contrast to some works (for example, Mannasoo and Mayes (2009)), financial depth is found positively associated with our stability indicators, suggesting an overall beneficial effect of the increasingly prominent banking sector in emerging economies Robustness tests In this section we conduct a series of robustness tests to check if our baseline result holds when using alternative risk indicators, different econometric methodologies and country-level data. First of all, following some other practices commonly adopted in extant literature (for example, Claessens et al., 2001; Altunbas et al., 2007; Laeven and Levine, 2009), we use the ratio of non-performing loans over gross loans (NPL), the ratio of loan loss reserves over gross loans (LLR) and the standard deviation of return on equity (σ(roe)) respectively to replace our original dependent variables. Reported in Part 1 of Table 3, the estimated coefficients on foreign bank penetration, measured by Pene_assets and Pene_number respectively (Panel A and B), are found positive and statistically significant in most regressions. This is interpreted as complementary evidence for our benchmark result that the degree of foreign bank penetration is positively associated with the vulnerability of domestic banks, which witness a rise of non-performing loans, hold more loan loss reserves and receive more volatile returns. [Table 3] Second, we apply different econometric methodologies to examine the nexus between foreign bank penetration and domestic banks risk-taking. It can be argued that, foreign banks may be more inclined to enter markets where domestic banks are more fragile since they incur lower costs for mergers and acquisitions, thus this reverse causality would lead to biased results. We employ the 2SLS instrumental variable estimator to address this problem. Three instrumental variables for the penetration level of foreign banks are selected, namely, the penetration of foreign banks in other markets located in the same region (Central and Eastern 16

17 Europe/Latin America/Asia), the regulatory quality of host governments, and the stock of foreign direct investment in host countries in per capita terms. 21 Foreign banks may establish their operation more likely in areas where their peers cluster. Because the common driving factors of foreign bank penetration, such as the home-host cultural and institutional closeness and the expected profitability based on host countries economic development, could be similar in near emerging countries, foreign banks may assemble in a region and thereby the level of foreign penetration in one country can be correlated with that in other near countries. Our data also confirm that the degrees of foreign bank penetration in emerging economies are roughly similar within regions but are quite different across regions. Meanwhile, foreign banks may prefer entering a market where the government can formulate and implement market-friendly policies, or follow their clients into countries subsequent to the latter s foreign direct investment (Brealey and Kaplanis, 1996; Moshirian, 2001). Nevertheless, it is less likely that these factors would affect the riskiness of domestic banks directly, suggesting that they are proper instrumental variables for foreign bank penetration. The result, presented in Part 2 of Table 3 (Panel C and D), is qualitatively consistent with our baseline findings. 22 We next revise our model to a dynamic version by adding the one-year lagged dependent variable as a covariate and then use the system GMM estimator to estimate this dynamic panel model. Besides the level of foreign bank penetration, bank characteristics are also assumed endogenous even though we have used their one-year lagged value in estimations. As the result in Part 2 of Table 3 (Panel E and F) shows, the coefficients on both Pene_assets and Pene_number are still negative and statistically significant in all but one regression, confirming again our earlier results. However, the Hansen J statistics (not reported for brevity) suggest that the instrument variables used in the GMM estimator, i.e. the lagged levels of the variables used for the difference equation and the lagged differences of the variables used for the level equation, are only weak, casting doubt on the validity of the instruments and the estimates obtained by using the GMM estimator. 21 To be more specific, we calculate the ratio of the assets (number) of foreign banks in all other countries in the same region over the total bank assets (number) of those countries, and use it as the instrumental variable of Pene_assets (Pene_number). The data for the regulatory quality of host countries and the stock of FDI in host countries in per capita terms are from Kaufmann et al. (2010) and the UNCTAD Statistics, respectively. 22 Not reported due to the limited space, the sign on the estimated coefficients of the three instrumental variables in the first-stage regression is positive, as expected, and statistically significant. However, the Hausman test on the endogeneity of foreign penetration only rejects the hypothesis that it may be treated as exogenous when using Pene_assets as the bank risk indicator. We also conduct the Hansen test for overidentification restrictions and find the Hansen J statistic is statistically insignificant in most regressions, implying that our choices of instrumental variables are valid. 17