Threshold effect of economic openness on bank risk-taking: Evidence from emerging markets

Journal Pre-proof Threshold effect of economic openness on bank risk-taking: Evidence from emerging markets Tung Duy Bui, Hoai Thi Mai Bui PII:

S0264-9993(19)30053-7

DOI:

https://doi.org/10.1016/j.econmod.2019.11.013

Reference:

ECMODE 5071

To appear in:

Economic Modelling

Received Date: 10 January 2019 Revised Date:

9 November 2019

Accepted Date: 12 November 2019

Please cite this article as: Bui, T.D., Bui, H.T.M., Threshold effect of economic openness on bank risktaking: Evidence from emerging markets, Economic Modelling (2019), doi: https://doi.org/10.1016/ j.econmod.2019.11.013. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Published by Elsevier B.V.

Threshold effect of economic openness on bank risk-taking: Evidence from emerging markets Tung Duy, Bui

∗ 1

and Hoai Thi Mai, Bui

† 1

1

University of Economics Ho Chi Minh City School of Public Finance Addresses: 279 Nguyen Tri Phuong, Ward 5, District 10 Ho Chi Minh City, Vietnam November 8, 2019 Acknowledgments: The authors would like to thank the editors and two anonymous reviewers of this journal for their very constructive comments that contributed to the improvement of the paper. We are solely responsible for any error that might yet remain. This research is funded by the University of Economics Ho Chi Minh City, Vietnam.

∗ †

Email: [email protected] Email: [email protected]

1

Threshold effect of economic openness on bank risk-taking: Evidence from emerging markets

Abstract This paper investigates the non-linear effects of two aspects of economic openness, namely, trade openness and financial openness, on banking system stability. We use a panel of 42 emerging markets from 2000 to 2014 to test whether bank risk-taking behaviour varies with the level of openness. We find that a higher degree of trade openness promotes bank stability linearly. Conversely, the non-linear effect of financial openness on bank risk-taking is evident. When the financial system is not sufficiently open, the impact of financial openness on bank stability is insignificant. However, as the domestic financial market becomes more open, financial openness can help discipline the behaviour of banks, making them more stable. We also find evidence that these effects are transmitted through the market discipline channel. Our findings highlight the importance of strengthening the domestic regulatory framework and transparency as the economy becomes more integrated. Keywords: Financial openness, Trade openness, Panel smooth transition regression, bank risk-taking. JEL Codes: G21, F30, G38.

Preprint submitted to Economic Modelling

November 13, 2019

1. Introduction Orthodox banking operations still dominate the banking systems in emerging countries due to less-developed financial markets and more pronounced asymmetric information problem compared with their counterparts (Vives, 2006; Agca and Celasun, 2012). Although leading emerging economies have made their way into the Basel Committee as proof of their reforming commitment1 , the next step in their transformation process can be troublesome as the Basel regulations are meant, at first, for developed banking systems. These markets have yet to improve the accounting data quality, auditing capability, and risk measurement (Bourgain et al., 2012). The reforming processes of the banking industries in emerging economies are indeed fragmentary and result in sizable market friction (Bhaumik et al., 2018). The continuing heavy reliance on lending activities, thus makes emerging market banks susceptible to credit risk. This type of risk is considered as the primary cause of various banking crises in emerging countries (Y¨ uksel, 2017). The outbreak of the global financial crisis of 2008–2009 had its origin in the extreme risk-taking lending activities of banks before the crisis. Since then, the problem of bank risk-taking has become progressively prevalent to academic researchers and policymakers (Ashraf et al., 2016; Agoraki et al., 2011). Improving efficiency and stability of the banking systems is relevant to the emerging world to ensure the optimal allocation of financial resources because their capital markets are yet to mature, and banks still play an important role in channelling funds to businesses and infrastructure projects. Moreover, in a context of continuing globalisation with borderless financial markets, deregulation, and breakthrough technologies, it is exceedingly important to understand the bank risk-taking behaviour (Mirzaei et al., 2013). The first step of the reforming process in these emerging markets would begin with the opening of their financial systems. This policy is believed to improve the efficiency 1

There are 14 emerging economies in the Basel Committee, of which 11 economies have the full membership status ( Argentina, Brazil, China, India, Indonesia, Korea, Mexico, Russia, Saudi Arabia, South Africa, and Turkey) and three observers: Chile, Malaysia, and United Arab Emirates.

2

of the financial systems and foster growth (Klomp and de Haan, 2014). According to the openness theory, financial development can relate to both integrating into the international goods markets (trade openness) and the international financial markets (financial openness) (Rajan and Zingales, 2003). Recently, several authors have investigated how bank risk-taking behaviour responded to economic openness, including trade openness and financial openness (Ashraf et al., 2017; Bourgain et al., 2012; Cubillas and Gonz´alez, 2014; Luo et al., 2016). However, there is no consensus among researchers regarding the impact of openness on bank risk-taking. On the one hand, opening the domestic market would enhance the risk management of emerging market banks by tightening lending requirements and reducing firms’ reliance on bank credits. On the other hand, financial openness can induce extreme risk-taking due to deregulation and competition (Klomp and de Haan, 2014; Agca and Celasun, 2012). Given this context, the paper aims to examine the possible non-linear impact of economic openness on bank risk-taking behaviour in emerging markets. In the lower quantiles of the emerging world, the reforming processes of their banking industries are still at an early stage. Thus, the findings of the paper can have important policy implications for these governments. The second motivation comes from the fact that most of the previous studies concentrate on the linear impact of financial openness or trade openness on bank stability while paying little or no attention to non-linearity and heterogeneity. The paper contributes to the existing literature of the openness theory by looking at both non-linearity and heterogeneity. First, we believe that the impact of financial openness and trade openness on bank risk-taking behaviour can be non-linear. The previous literature (Ashraf et al., 2017; Bourgain et al., 2012; Cubillas and Gonz´alez, 2014; Luo et al., 2016) only examines the linear effect and thus cannot fully reflect the controversy of the theoretical literature. Even though our sample excludes the advanced economies, there is still heterogeneity among the chosen emerging countries. For instance, there are leading emerging markets with competent levels of banking development, which are members of the Basel Committee. At the same time, some countries 3

found themselves in the early stages of the banking reform process. Thus, heterogeneity is relevant and has different empirical outcomes. To the best of our knowledge, this is the first study to incorporate both non-linearity and heterogeneity to examine the impact of economic openness on bank risk-taking. We use the method called panel smooth transition regression (PSTR) model developed by Gonzalez et al. (2017). In this model, the coefficients of the regressors can be different among individuals and across time. Hence, the model is capable of correcting heterogeneous bias and instability. Our findings indicate that bank risk-taking increases at low levels of financial openness, whereas it decreases at higher levels. These findings suggest that when there is a sufficient level of openness, banks will behave more prudently as the financial markets become more open. Our findings support the positive effect of successfully opening the financial market. The results are robust to various tests, including alternative measures of financial openness and different empirical specifications. The rest of the paper is organized as follows. Section 2 provides a brief review of the literature on the issue and develops the main hypotheses. Section 3 introduces the data and variables. Section 4 presents the PSTR model and its two specification tests. The empirical results are discussed in Section 5, and Section 6 concludes the study. 2. Literature review Existing literature has not reached a consensus regarding the impact of economic openness on bank risk-taking. On the empirical ground, the transmission channels were not fully understood, and researchers have yet to find the exact channel (Cubillas and Gonz´alez, 2014). In general, there are two main opposing streams of literature explaining how economic openness can affect bank risk-taking. In the first strand of literature, opening the domestic economy may induce banks to behave more riskily, through higher exposure to either risk or competition. On the contrary, there are other authors (Boyd and Nicol´o, 2005; Klomp and de Haan, 2014) arguing that opening the domestic markets can help reduce bank risks. For example, bank risk-taking can be reduced through the market discipline effect. Opening the domestic market requires 4

that governments in developing countries strengthen their regulatory framework. Banks would take these opportunities to develop better governance and reduce risk in the process. First, economic openness can have a destabilising effect on bank stability for several reasons (Hellmann et al., 2000; Repullo, 2004). Opening the goods or the financial markets would increase competition and render these markets more unpredictable. According to this view, banks’ market power shrinks in high competition, leading to a decrease in their profit margins and charter values. To tackle the situation, banks would have tendencies to build up their total loans to make up for their lower profit margins. In such aggressive competition, banks are forced to lower their credit standards to accumulate more loans, thus resulting in a higher amount of nonperforming loans. On top of that, the more open the country becomes, the more probable that a surge in nonperforming loans can occur. This is because trade openness can make the domestic economy more exposed to the global economic cycle. Higher trade openness can cause wide fluctuation of domestic macroeconomic variables, including income, prices, employment, and wages. Thus, a negative foreign shock would be detrimental to the domestic business sector and subsequently affect their ability to repay debts. As a consequence, highly open markets are more susceptible to income volatility and unpredictability (Ashraf et al., 2017). Moreover, under the context of an unrestricted financial system, banks can broaden their activities into other markets and unconventional operations. Given that these banks do not have sufficient expertise in such markets, their offshore operations can become more exposed to risk (Cubillas and Gonz´alez, 2014; Fang et al., 2014). This argument is similar to the market risk hypothesis (Berger et al., 2017). This hypothesis states that high-risk foreign assets caused by different market fundamentals can generate high risk levels in multinational banks, which can only be alleviated through extreme diversification. For instance, foreign exchange can be an important source of overseas risk because financial openness often requires removing restrictions on capital flows. 5

There are also other factors that international banks must overcome to establish a strong foundation in foreign markets. They must compete with existing local opponents to earn a reasonable market share and spend time to understand specific foreign traditions and customs. Thus, the problems of imperfect market and asymmetric information become more relevant. The other mainstream of the literature has its foundation in the positive effect of a successful economic openness. Opening the domestic market allows the inflow of foreign capital. These funds can enhance domestic banks’ liquidity and strengthen their capabilities to allocate more capital into high-yielding projects. This argument is in line with the so-called diversification hypothesis (Berger et al., 2017). Diversification is also possible when banks become less restrictive in an open economy, which allows them to follow the economy of scale and scope (Fang et al., 2014). Diversifying their investment portfolios can help multinational banks lower their risk. The degree of risk buffer increases with the correlations of cross-country assets. This effect is similar to opening the goods markets since banks can extend their operations to both overseas and domestic firms. Banks whose clients are multinational firms can survive longer in a negative shock. In fact, highly integrated industries have strong capacities in diversifying internationally and are less vulnerable to domestic shocks. Similarly, there is evidence that the efficiency and survivability of international firms are higher than those of domestic firms (Ashraf et al., 2017). Furthermore, Boyd and Nicol´o (2005) argue that shrinking bank market power in an open market can control the reckless behaviour of domestic banks. In this line of thought, banks with prominent market power can charge their clients with high lending interest rates, which makes repayment more difficult. This situation worsens the moral hazard problems since the clients need to seek riskier investments to compensate for the higher interest rate. Hence, reducing bank market power leads to a decline in the interest rate and lessens the moral hazard problem since the clients would be less interested in risky investments. However, it is essential to note that market power can hurt bank stability. In a recent study, Ahamed and Mallick (2017) find that banks with 6

higher market power might have produced more risky assets due to the moral hazard problem. Hence, a linear concentration–stability relationship in banking literature is controversial, which allows for possible non-linearity. For instance, IJtsma et al. (2017) investigate the bank’s market power–stability nexus using the Z-score at both individual and aggregate levels for 25 European economies. They find that concentration has insignificant effects on bank stability at both levels of analysis. The authors only consider linear relationships. The opposite effects of concentration on bank stability can cancel each other out. As aforementioned, bank stability can be strengthened through the market discipline effect. Trade openness and financial openness require emerging market governments to commit to reform the domestic regulations, such as removing the restrictions on financial markets, improving law, and enforcing the law. Governments also take pressure from the global organisation (WTO, the Basel Committee) and must discipline their domestic markets in accordance with the international standards. Members of WTO must comply with the General Agreement on Trade in Services, which put restrictions on bank size, and capital management procedures (Ashraf, 2018). Stronger regulations enhance banks’ financial transparency and governance, which is beneficial for bank stability (Fang et al., 2014). ). Several authors also believe that financial fragility can be attenuated through reinforcing banking sector regulation and supervision (Klomp and de Haan, 2014). If these measures are not sufficient, the government can rely on similar market discipline in the banks’ regulatory framework (Mostak Ahamed and Mallick, 2017). Lapteacru (2017) finds that better regulatory framework helps to stabilise bank with high market power in Central and Eastern Europe. He shows that larger banks in highly concentrated banking markets will have reduced risk if there are strict regulations for bank supervision. Benbouzid et al. (2017b) provide another supporting evidence for the market discipline effect. They show that banks can have lower credit default swaps (CDS)’s spreads by improving economic and legal institutions. The negative shock of the recent global financial crisis has fewer effects on banks in countries with a stronger regulatory framework. 7

Lately, the impact of economic openness on bank risk-taking has received attention from several researchers (Ashraf et al., 2017; Ashraf, 2018; Bourgain et al., 2012; Cubillas and Gonz´alez, 2014; Luo et al., 2016). Bank risk-taking in MENA countries is positively correlated with the degree of international financial openness (Bourgain et al., 2012). This study also discusses a new mechanism in which depositors influence bank risk-taking. Savers are attracted to higher-financial-transparency markets. Hence, they induce banks to conduct their business properly to attract these funds. The more open the financial systems become, the higher the transparency effect. However, the authors note that the positive effect of financial openness only appears at very high levels of openness. At low levels of financial openness, they observe excessive risk-taking behaviours. In another study, there is evidence that financial openness is the cause of declining bank stability in both advanced and developing economies (Cubillas and Gonz´alez, 2014). The authors also document the positive effect of regulation on both the developed and developing world. Similar results can be found in the work of Luo et al. (2016). Financial openness can directly reduce bank efficiency and indirectly lead to higher bank risk through the profit channel. Benbouzid et al. (2017a) supplement the evidence by showing that country-level financial stability is negatively associated with credit risk. On the contrary, bank risk-taking is found to be lower at high levels of trade openness (Ashraf et al., 2017). The authors explain the results using the diversification channel. They believe that opening the goods markets allows diversification possibilities and thus reduces total bank risk. Similar effects of trade openness on bank risk are found in the recent work of Ashraf (2018). However, the results of financial openness are inconclusive. Financial openness lowers the cost of credit and the total loans granted but increases the risk-taking behaviour. In general, on both theoretical and empirical grounds, the impact of economic openness on bank risk-taking is still open to debate. However, the theoretical model of Bourgain et al. (2012) brings a new perspective to look at the relationship. They find that excessive bank risk-taking behaviour disappears at a very high level of financial 8

openness. The authors suggest two effects of transparency in the banking system: the attractive effect and the profit-squeezing effect. When banks become more transparent, they will attract more depositors, thus creating the attractive effect. Besides, high financial openness induces more competition. Banks need to squeeze as much profit from the competitive market, and strict financial transparency can result in further bank losses, which refers to the profit-squeezing effect. The level of financial openness decides the dominated effect. At a lower degree of openness, the profit-squeezing effect will dominate the attractive effect, which leads to more risk-taking behaviour. The opposite occurs at a very high level of financial openness, where depositor attraction incentives are more significant than profit-squeezing motivation. In a recent study, Ahamed and Mallick (2019) also show that providing formal financial services, such as deposit services, is also an essential channel to maintain bank stability. Thus, this study aims to investigate this “level” empirically. From these results, we postulate the two main hypotheses of this study: H1: Bank risk-taking increases at low levels of financial openness, whereas it decreases at high levels of financial openness H2: Bank risk-taking increases at low levels of trade openness, whereas it decreases at high levels of trade openness 3. Empirical models 3.1. A PSTR model Testing the main hypotheses of the paper requires the estimation of a non-linear model. One traditional method that solves this type of non-linearity and heterogeneity is estimating a panel threshold regression (PTR), developed by Hansen (1999). The PTR assumes that analogous individuals should belong to one group. Thus, one can divide the individuals in the sample into several groups based on observables. For example, the effect of a variable x on the dependent variable y can be examined by a cut-off rule: yit = µi + β0 xit + β1 xit 1qit ≥c + uit 9

(1)

Equation (1) implies that the transition from the lower regime (qit < c) to the higher regime (qit ≥ c) requires a ”jump” at the threshold level qit . Changing regimes is governed by an indicator function, which results in a non-smooth transition2 . This paper employs the PSTR model developed by Gonzalez et al. (2017) to account for non-linearity and heterogeneity, which can also result in a smoother transition. This type of model allows regime-switching in a certain number of extreme regimes. The PSTR model requires several transition variables or threshold variables. When the value of these transition variables changes, the model would move from one extreme regime to another. Using a PSTR model in a panel context yields several advantages. First, the coefficients of the regressors can be different among individuals and across time. Hence, the model is capable of correcting heterogeneous bias and instability. Moreover, in contrast to the panel threshold model (PTR), the estimated coefficients at the extreme regimes of a specific individual at a particular time do not have to be equal. The coefficients of this type of model are allowed to vary across cross-sectional and time units. The coefficients of the regressors take the form of a continuous bounded function whose value depends on an observable variable (or the transition variable). The continuous bounded function is called the transition function. Its value varies between certain numbers of extreme “regimes.” Smooth transition autoregressive models employ the same logic in time series analysis (Gonzalo and Wolf, 2005). However, the PSTR is different from these types of models since it considers a panel specification and does not have lagged endogenous variable in the right-hand side of the equation 3 . To be more specific, a PSTR model with two extreme regimes and one transition function can be written as: 0

yit = µi + β0 xit + β1 xit g(qit ; γ, c) + Φ Zit + uit

(2)

where i = 1...N ; t = 1...T denote the total number of countries and time units in the panel; yit , the dependent variable; xit , a time-varying independent variable; Zit , an 2 3

One can refer to Wang (2015) for the implementation of a PTR model. For dynamic PSTR models, the work of Seo and Shin (2016) provides an excellent reference.

10

m-dimensional vector of time-invariant control variables; Φ, an m-dimensional vector of time-invariant coefficients; and µi , the individual fixed effect. The model assumes i.i.d. error term, denoted as uit . qit is the transition or threshold variable. Its value determines the value of the transition function g(qit ; γ, c). The form of this function is similar to that of Gonzalez et al. (2017): " g(qit ; γ, c) = 1 + exp

!#−1 m Y − γ (qit − cz ) ; γ > 0; c1 ≤ ... ≤ cm

(3)

z=1

By definition, the bounded continuous function g(qit ; γ, c) has its value fluctuate from 0 to 1. Hence, there are a certain number of extreme “regime 0” and extreme “regime 1”. The number of extreme regimes depends on the number of the location parameter c. 0

c = (c1 , ..., cm ) is an m-dimensional vector of location parameters. For example, m = 1 would create two extreme regimes, whereas m = 2 would have three extreme regimes. γ is the slope coefficient of the transition function, which controls the speed of the transition from one extreme regime to another. The function g(qit ; γ, c) will converge to the extreme regimes more quickly with a large value of γ, with all other things being equal. Taking the partial derivative of yit with respect to xit gives us the non-linear impact of xit on yit : ∂yit = β0 + β1 g(qit ; γ, c) = β¯ (If qit 6= xit ) ∂xit

(4)

The total impact of xit on yit depends on various parameters: the constant coefficients β0 , β1 , γ; the constant location parameter c; and the time and individualspecific transition variable qit . Since the values of g(qit ; γ, c) are bounded between 0 and 1 depending on the value of qit , the total impact will fluctuate between β0 (when g(qit ; γ, c) = 0) and β0 + β1 (when g(qit ; γ, c) = 1). Figure 1 illustrates the variation of the total impact and the transition speed with different values of γ, where β0 , β1 > 0. As qit increases, the transition function moves

11

from one extreme regime to another and the convergence speed depends on the value of γ. It can be seen that as qit is time- and individual-specific, the total impact for a different individual at a different time would not be identical. Thus, the empirical model can account for the non-linearity and heterogeneity. β¯ = β0 + β1 g(qit ; γ, c)

β0 + β1 γ = 25, c = 5

γ = 1.3, c = 5 β0

c=5

qit

Figure 1: The speed of transition with different values of γ

The first step in the estimation of the PSTR model requires removing individualspecific means to get rid of the fixed effects, then the transformed model can be estimated using non-linear least squares. However, before estimating the PSTR model, we need to ensure that the model satisfies two main specification tests (Gonzalez et al., 2017). It can be seen that the model (2) becomes the linear model when γ = 0 or β1 = 0. In such a case, there is no point in estimating a non-linear PSTR model. Thus, the first specification test is the linearity test against PSTR model. The test would have the following null hypothesis: either (i) H0 : γ = 0 or (ii) H0 : β1 = 0. That is to say, we can interpret H0 as: Equation (2) is linear ; and H1 as: Equation 2 is non-linear with at least one transition function. The test statistics can be calculated by employing a standard 12

LM test, which gives us the LM statistics and its F-version (Gonzalez et al., 2017). Under the null hypothesis, the LM statistics would be asymptotically distributed as χ2 (mk), whereas its F-version has an approximate F (mK, T N − N − mK) distribution. Another pseudo- LRT statistic can be calculated, which has a χ2 (mK) distribution. If the linearity test is rejected, the next stage requires determining the number of transition functions or the number of regimes in the model, before estimating the final PSTR model (Gonzalez et al., 2017). The procedure of the so-called non-remaining heterogeneity test is basically identical to the linearity test. Naturally, we can state the null hypothesis as H0 : Equation 2 has only one transition function.; and the alternative hypothesis as H1 : Equation 2 has at least two transition functions.. The test can be done by calculating the same LM statistics with modified degrees of freedom. If the null hypothesis is rejected, we need to continue the non remaining heterogeneity by increasing the number of transition functions. The test only ends when H0 cannot be rejected. 3.2. Variables In this study, we will use the bank Z-score as our main dependent variable. This measure of bank risk is widely used in banking literature (Houston et al., 2010; Berger et al., 2017; Ashraf et al., 2017; Ashraf, 2018). As we collect the Z-score variable from the Global Financial Development database, we use the World Bank’s definition: Z-score =

ROA + CAR σ(ROA)

(5)

where ROA is the annual return on assets, and CAR is the ratio of annual equity to total assets. σ(ROA) is the standard deviation of ROA. In the World Bank’s definition of Z-score, ROA, equity, and assets are country-level aggregate figures calculated from underlying bank-by-bank unconsolidated data. The Z-score measures the probability of default of a banking system. In other words, the Z-score compares the buffer of a country’s banking system (capitalisation and net income) with the fluctuations of those incomes. Z-score is a prominent candidate as a dependent variable and for measuring 13

bank risk since its value belongs to the domain of all real numbers (Lepetit and Strobel, 2015). The Z-score has high skewness (Ashraf, 2018). As a consequence, we take the natural logarithm of Z-score as our dependent variable. Two main independent variables in this study are the trade openness and financial openness. Trade openness is measured as the total volume of imports and exports over the annual GDP of a country. Kim et al. (2010) argue that this measure can provide unambiguous quantification of trade openness. This variable is also used as the primary independent variable in studies investigating the impact of trade openness on bank risk-taking (Ashraf et al., 2017; Ashraf, 2018). The second primary independent variable is the level of financial openness. In this study, we measure financial openness using the “de facto” index developed by Lane and Milesi-Ferretti (2007). This volume-based quantity captures the extent of a country’s capital control. There are also other measurements of financial openness, such as the “de jure” index of Chinn and Ito (2006), but they are not used in this study due to their characteristics. In the context of a PSTR model, the transition variable needs to be time-variant. However, the index of Chinn and Ito (2006) is time-invariant. In this sense, the “de facto” index is superior to “de jure” index as a transition variable in a PSTR model (Bui, 2018). Lane and Milesi-Ferretti (2007) provide two measures of financial openness, and we will use both variables in this study: F Ait + F Lit GDPit

(6)

P EQAit + F DIAit + P EQLit + F DILit GDPit

(7)

F Iit1 =

F Iit2 =

where F A and F L represent the stock of external assets and liabilities, respectively. These foreign assets and liabilities include foreign debt, foreign direct investment, and foreign indirect investment (portfolio investment). P EQA andP EQL denote the stock of assets and liabilities of portfolio investment; F DIA and F DIL denote the stock 14

of assets and liabilities of foreign direct investment, respectively. Hence, the second measure of financial openness F Iit2 in Equation (7) excludes the foreign debt from the calculation in comparison with the first measure F Iit1 in (6). Thus, F Iit2 is also considered a measure of capital account openness. We also introduce a set of control variables to capture both bank-specific factors and banking industry preference and macroeconomic indicators and institutional determinants. Bank level control variables included in the model are bank liquid assets, total bank assets, bank return on assets after tax, and bank capital to assets. We use the assets of the three largest commercial banks to total commercial banking assets as a measurement for banking concentration, a control variable for the banking industry of a country. Another control variable at the industry level is the presence of foreign banks in the domestic market4 . Furthermore, in the robustness tests, we add other control variables into the model to check the consistency of the findings. For macroeconomic indicators, we use the GDP growth rate. Then, to control the institutional factor, we use the control of corruption index. In a PSTR model, it is greatly important to choose a suitable transition variable. There is virtually no technical constraint in choosing a transition variable, and the choice depends totally on the economic issue. The two main hypotheses of this study relate to the investigation of the changes in risk-taking behaviour of banks as the economy moves to higher openness levels. Thus, in this study, the appropriate transition variable would be the level of economic openness. However, to control endogeneity problems, we use the lagged value of economic openness as the transition variable. Hence, there are two transition variables in this study, which are the first-order lag of the financial openness and trade openness index, respectively. 4

We proxy for the presence of foreign banks by using the percentage of the total banking assets that are held by foreign banks. A foreign bank is a bank where foreigners own 50 percent or more of its shares.

15

3.3. Data In this study, we consider a panel of 42 emerging and developing economies from 2000 to 2014. The year 2014 is the last year that Lane and Milesi-Ferretti (2007) update their data on financial openness. At first, we collected data for all emerging and developing markets. However, due to the problems of missing data and gaps in the panel, we choose 42 emerging markets. All the data are collected from the Global Financial Development database of the World Bank, except for the control of corruption and financial openness data. We collect the corruption index from the World Governance Indicator database, and calculate the financial openness index of Lane and Milesi-Ferretti (2007). The countries included in the study are: Albania, Argentina, Armenia, Bangladesh, Bulgaria, Bolivia, Brazil, Chile, China, Colombia, Costa Rica, Czech Republic, Ecuador, Estonia, Georgia, Ghana, Greece, Croatia, Indonesia, India, Kazakhstan, Kyrgyz Republic, Lithuania, Latvia, Morocco, Mexico, Malaysia, Nigeria, Oman, Pakistan, Peru, Philippines, Poland, Romania, Serbia, Slovenia, Thailand, Tajikistan, Turkey, Ukraine, Venezuela, and South Africa.

Table 1: Summary statistics

mean zscore 10.856 1 FI 114.725 F I2 46.569 Trade openness 77.699 Bank liquid liabilities 47.256 Bank capital to assets 10.685 Total banks assets 49.696 Bank return on asset 16.727 Bank concentration 60.178 GDP growth 4.522 Control of corruption -0.245 Foreign bank 39.358

sd 7.792 58.007 31.151 35.496 30.313 4.061 30.213 14.255 17.768 4.208 0.644 21.403

16

min -0.340 27.977 4.476 21.852 6.330 1.490 4.210 -59.890 23.320 -14.814 -1.497 3.000

max 47.730 392.408 180.613 220.407 180.930 30.500 148.700 98.850 100.000 33.736 1.592 86.000

4. Empirical results 4.1. Linear estimation results To begin with, we regress the empirical model (2) in its linear form, i.e., by assuming that γ = 0 or β1 = 0, to compare our preliminary results with other studies. In particular, the linear regression can be written as 0

Zscoreit = µi + β0 Opennessit + Φ Controlit + uit

(8)

We report various specifications of the linear model using individual fixed effect and time fixed effect in Table 2. The dependent variable is the Z-score with higher values indicating safer buffers from the volatility of the returns. Hence, a higher Z-score would mean more stability and less risk-taking. The primary independent variable Openness in the first two columns of Table 2 is trade openness, whereas in the other columns, we use both measures of financial openness. Both financial openness and trade openness are positive and significant in all linear specifications. The estimated coefficients of trade openness are only significant at 10% and 5% levels. These results imply that economic openness induces banks to behave more prudently and, hence, less risk-taking behaviour. The results are similar to the recent work of Ashraf et al. (2017); Ashraf (2018). Banking level control variables are also significant; more bank liquid liabilities and bank capital increase stability. Likewise, higher efficiency in using assets increases banks’ buffer against net income volatility. On the contrary, the coefficients of bank size are negatively significant. These results suggest that bigger banking systems take more risk than smaller ones. Our findings are in line with the “too big to fail” hypothesis. However, the bank industry and macroeconomic control variables are only significant in some specifications. The linear results indicate that both financial openness and trade openness decrease bank risk-taking behaviour, or increase bank stability. In the following specifications, we examine whether this impact on bank risk changes at different levels of economic 17

Table 2: Linear estimation with individual fixed effects and time fixed effects, dependent variable: Logarithm of Z-score

Trade openness

Fixed

Fixed& Time

0.254∗ (1.91)

0.338∗∗ (2.47)

F I1

Fixed

Fixed& Time

0.240∗∗∗ (2.86)

0.290∗∗∗ (3.21)

F I2

Fixed

Fixed& Time

0.247∗∗∗ (4.06)

0.239∗∗∗ (3.60)

Bank liquid liabilities

1.023∗∗∗ (3.45)

0.636∗∗ (2.00)

1.040∗∗∗ (3.53)

0.618∗ (1.96)

0.902∗∗∗ (3.09)

0.596∗ (1.90)

Bank capital to assets

0.047∗∗∗ (6.12)

0.048∗∗∗ (6.22)

0.048∗∗∗ (6.32)

0.048∗∗∗ (6.31)

0.044∗∗∗ (5.75)

0.045∗∗∗ (5.85)

Total banks assets

-1.097∗∗∗ (-4.80)

-1.090∗∗∗ (-4.76)

-1.235∗∗∗ (-5.22)

-1.295∗∗∗ (-5.28)

-1.138∗∗∗ (-5.18)

-1.089∗∗∗ (-4.88)

Bank return on asset

0.894∗∗∗ (6.43)

0.951∗∗∗ (6.82)

0.934∗∗∗ (6.88)

0.987∗∗∗ (7.20)

0.962∗∗∗ (7.15)

1.031∗∗∗ (7.53)

Bank concentration

-0.412∗∗∗ (-3.22)

-0.348∗∗∗ (-2.60)

-0.336∗∗∗ (-2.67)

-0.215 (-1.64)

-0.232∗ (-1.80)

-0.173 (-1.30)

GDP growth

0.153 (0.39)

0.852∗ (1.90)

0.297 (0.78)

0.908∗∗ (2.04)

0.211 (0.55)

0.833∗ (1.87)

Control of corruption

0.206∗∗ (2.00)

0.234∗∗ (2.25)

0.286∗∗∗ (2.75)

0.336∗∗∗ (3.18)

0.223∗∗ (2.21)

0.260∗∗ (2.53)

Foreign bank

0.475∗∗ (2.20) 0.317 0.235 455

0.350 (1.50) 0.355 0.256 455

0.356 (1.63) 0.325 0.243 455

0.099 (0.42) 0.361 0.264 455

0.057 (0.24) 0.338 0.258 455

-0.026 (-0.10) 0.366 0.269 455

R2 Adjusted R2 Observations ∗p < 0.1; ∗ ∗ p < 0.05; ∗ ∗ ∗p < 0.01

18

openness. Figure 2 shows some basic scatterplots for financial openness, trade openness, and Z-score. We also plot a linear fit and a quadratic fit. Visually, the quadratic fit (dashed line) and the linear fit are indistinguishable for the case of trade openness,

5 4 3

Z-score

0

1

2

3 2 0

1

Z-score

4

5

whereas the gap between these two curves is wider as financial openness increases.

3

3.5

4

4.5

Financial openness Observations Linear fit

5

5.5

Quadratic fit

2

3

4

Trade openness Observations Linear fit

5

Quadratic fit

Figure 2: Scatter plot: Financial openness, trade openness and Z-score

4.2. PSTR estimation results Before estimating the PSTR model, it is necessary to check whether the model satisfies the two main specification tests. The first step involves implementing the linearity test against the PSTR model. The next stage of PSTR model estimation requires that the null hypothesis of linearity is rejected. Otherwise, the estimation of a non-linear model is redundant. In this paper, we implement the linearity test using the following transition variables: trade openness and two measures of financial openness. The upper panel of Table 3 reports the test statistics for financial openness, whereas the lower panel of Table 3 reports the test statistics for trade openness. We report the test statistics for the LM test, its F-version, and a pseudo-LRT and the critical values at 1%, 5%, and 10% significant levels in the three last columns of 3 and Table 4.

According to these results, the test statistics for the two proxies of financial openness are larger than all the critical values, leading to the rejection of the null hypothesis of 19

6

Table 3: Linearity test against PSTR model

Transition variable: Financial openness Test LM LMF LRT

Test Test statistics 1 statistics 2 17.751 1.865 18.005

28.208 3.015 28.859

1%

Critical value at 5%

21.666 2.438 21.666

16.919 1.896 16.919

10% 14.684 1.642 14.684

Test statistic 1 is for the transition variable FI1 , while Test statistic 2 is for the transition variable FI2

Transition variable: Trade openness Test LM LMF LRT

Test statistics 4.426 0.513 4.439

Critical value at 5%

1% 20.090 2.538 20.090

15.507 1.952 15.507

10% 13.362 1.679 13.362

LM and LMF denote the standard LM test and its F-version. LRT denotes the pseudo-LRT test. The linearity test tests a linear model (H0 ) against a PSTR model with one transition function(H1 ). Source: Author’s calculation

20

linearity. However, the test statistics for the trade openness variable are smaller than the critical values. Thus, we cannot reject the null hypothesis of linearity for this case. Thus, the results only imply the non-linear effect of financial openness on bank risk-taking. Based on the results of the first test, we conduct the hypotheses of the second specification test to find the appropriate number of transition functions in the model. The no-remaining heterogeneity test begins two hypotheses : H0 : There is a single transition function in the model H1 : There are at least two transition functions in the model Table 4 reports the test statistics, with the results for the model with financial openness as the transition variable. All the test statistics are lower than the critical values. Hence we do not have enough evidence to reject H0 . As a consequence, we will estimate the PSTR models with only one transition function. 0

Zscoreit = µi + β0 F Iit + β1 F Iit g(T ransit ; γ, c) + Φ Controlit + uit

Table 4: Test of no-remaining heterogeneity

Transition variable: Financial openness Test LM LMF LRT

Critical value at 1% 5%

Test Test statistics 1 statistics 2 0.680 0.059 0.680

1.352 0.117 1.354

23.209 2.353 23.209

18.307 1.848 18.307

10% 15.987 1.610 15.987

Test statistic 1 is for the transition variable FI1 , while Test statistic 2 is for the transition variable FI2 LM and LMF denote the standard LM test and its F-version. LRT denotes the pseudo-LRT test. The test of no-remaining heterogeneity tests a model with one transition function (H0 ) against a model with two transition functions(H1 ). Source: Author’s calculation

21

(9)

Table 5 reports the estimated coefficients of Equation (9), with the two measures of financial openness as the transition variables. The estimation of the PSTR model with F I 1 as the transition variable is shown in the second column of Table 5, whereas the results with F I 2 are reported in the last column. Overall, the impacts of the bank-specific factors on bank risk-taking behaviour remain unchanged, which adds to the robustness of the results. As shown, the coefficients βˆ0 in both specifications are insignificant,whereas the coefficients βˆ1 are all significantly positive. We also observe a significant increase in the value of the adjusted R2 , compared with the previous linear models. However, similar to the logit or probit model, the coefficients should not be directly interpreted as there would be no meaningful conclusion. We can only interpret their signs. The first hypothesis of the study (H1), which states that bank risk-taking increases at low levels of financial openness, whereas it decreases at higher levels, is tested through the PSTR models with financial openness as the primary independent variable and its lagged value as the transition variable. Given that the transition function g(qit , γ, c) is an increasing function of qit and βˆ1 > 0, the total impact of financial openness on bank risk-taking will increase as the financial market becomes more open. At low levels of financial openness, the total impact would converge to the value of βˆ0 . However, the estimated coefficient is insignificant. In other words, financial openness would have an ambiguous effect on bank stability (insignificant) if the level of openness is sufficiently low. However, when an economy has a reasonable degree of openness (for qit that makes the value of g(qit , γ, c) converges to one), financial openness would enhance the banking system stability. Similarly, we can test the second hypothesis (H2) of the study using the PSTR model with trade openness as the primary independent variable and its first-order lag as the transition variable. However, as we do not find any evidence of non-linearity in this case, the impact of trade openness on risk-taking behaviour can be interpreted as in the case of linear models. In general, the results from various linear specifications show that trade openness would positively influence bank stability and discourage risk-taking 22

Table 5: Estimation of the PSTR model, dependent variable : Log of Z-score

qit : Financial openness 1 F I 1 (β0 )

qit : Financial openness 2

-0.106 (-1.57) 0.060∗∗∗

F I 1 (β1 )

(3.95) F I 2 (β0 )

0.002 (0.05)

F I 2 (β1 )

0.134∗∗∗ (4.84) 0.484∗∗

Bank liquid liabilities

(2.44) 0.054∗∗∗

Bank capital to assets

(7.36) -0.840∗∗∗

Total banks assets

(-4.84) 0.880∗∗∗

Bank return on assetes

(7.13) -0.343∗∗∗

Bank concentration

(-2.94) 0.357

GDP growth

(1.10) 0.279∗∗∗

Control of corruption

Foreign bank R2 Adjusted R2 Observations Location parameters cj Slope parameters γj ∗p < 0.1; ∗ ∗ p < 0.05; ∗ ∗ ∗p < 0.01 23

0.340∗ (1.70) 0.054∗∗∗ (7.42) -0.871∗∗∗ (-5.58) 0.857∗∗∗ (7.02) -0.287∗∗ (-2.43) 0.369 (1.13)

(2.97)

0.305∗∗∗ (3.34)

0.307∗ (1.96) 0.326 0.311 455 4.79 8

0.111 (0.65) 0.338 0.323 455 4.22 3

1 .8 .6 .4 0

.2

Value of the transition function

.8 .6 .4 .2 0

Value of the transition function

1

behaviour, which is in line with the studies of Ashraf et al. (2017); Ashraf (2018).

-2

0

2

4

Financial openness 1

6

8

-2

0

2

4

Financial openness 2

6

This figure shows the value of the transition function g(qit ; γ, c) in (3) with m = 1 smoothly changes from 0 to 1 as the level of financial openness increases. Source: Author’s calculation Figure 3: Estimated transition function g(qit ; γ, c)

Figure 3 shows the estimated value of the transition function g(qit , γ, c) as the value of qit increases. The convergent speed of the function with F I 1 as the transition variable is faster than the function with F I 2 as the transition variable, partly due to the higher value of the estimated slope parameter. The non-linear results of our study are in line with the recent studies (Cubillas and Gonz´alez, 2014; Bourgain et al., 2012), in favour of the market discipline perspective. For instance, in highly open banking systems and well-connected markets, banks would have more incentives to conduct their operations prudently to avoid the judgment of the markets or their depositors. Recently, the depositor-discipline channel is revisited in the study of Bourgain et al. (2012). In this context, depositors can correct the behaviour of banks using their savings. They can reallocate their savings to an overseas banking system in fear of the misconduct of the domestic banking system. In their theoretical model, they also show that a higher degree of financial openness increases competition and induces banks to shift to risky investments. However, this effect can be mitigated once the banking system is sufficiently open since banks must improve their transparency to attract more savings.

24

8

4.3. Transmission Channel In this section, we investigate several possible transmission channels in the literature that allow financial openness and trade openness to affect bank risk taking. We estimate a system of two simultaneous equations. T Cit = θ0 + θ1 T Ci,t−1 + θ2 Zscoreit + θ3 Opennessit + θ4 Controlit + uit Zscoreit = ψ0 + ψ1 Zscorei,t−1 + ψ2 T Cit + ψ3 Opennessit + ψ4 Controlit + vit

(10) (11)

where T C denotes the possible transmission channel. For example, we will use a proxy of bank’s competition as a dependent variable in Equation (10) if we want to examine whether the effect passes through the competition channel. The first equation in the system will control the impact of economic openness on competition. The 3SLS estimator then uses the estimated TˆC it as an explanatory variable in Equation (11). Thus, ψ2 will explain how financial openness (or trade openness) affects bank risk taking through the competition channel, and ψ3 will explain the effects through other channels. Apart from the competition channel, we explore other possible channels mentioned in Section 2. We look into the market discipline effect by using a proxy for the regulatory frameworks in developing countries, which is the regulatory quality index. This index is taken from the Worldwide Governance Indicators, measuring the extent that a government can formulate and implement sound policies and regulations that permit and promote private sector development (Kaufmann et al., 2011). The output channel is examined using GDP growth as a proxy for the domestic macroeconomic fluctuations. The presence of foreign banks in the domestic banking system can serve as a proxy for diversification channel. We use 3SLS to estimate Equations (10) and (11). We report the regression results of Equation (11) in Table 6. The coefficients of the regulatory quality index are significantly different from zero and positive for both trade openness and financial openness. These findings imply the presence of the market discipline effect. Opening the good markets and the financial markets requires improvements in domestic regulations, law and law enforcement. 25

Then, improving regulations will help enhance banks’ transparency and governance, which results in better bank stability. We do not find any evidence of the competition and output channel, since the coefficients of bank concentration and GDP growth are not statistically significant in all specifications. For the diversification hypothesis, only the coefficient of foreign bank for the trade openness case is positively significant. Table 6: Transmission channel of trade openness and financial openness, dependent variable: Logarithm of Z-score Trade openness

Trade openness

(1) 0.197∗∗ (2.42)

(2) 0.264∗∗ (2.40)

Financial openness

(3) 0.233∗∗ (2.22)

(4) 0.243∗∗∗ (2.76)

0.236∗∗∗ (3.22)

Financial openness 1

Bank concentration

-0.208 (-1.50)

(6)

0.277∗∗∗ (3.13)

(7)

0.245∗∗∗ (3.21)

GDP growth

-0.310 (-0.14) 0.257∗∗ (2.48)

Transparency

Bank capital to assets 0.042∗∗∗ (6.85)

0.260∗∗∗ (3.09)

0.301 (1.09) -0.577 (-0.31)

0.277 (1.18)

(8)

-0.162 (-1.08) 0.627∗∗ (2.33)

Foreign bank

Bank liquid liabilities

(5)

0.247∗∗ (2.10)

0.825∗∗∗ (2.65)

0.380 (1.46)

0.381 (1.53)

0.466∗ (1.79)

0.833∗∗∗ (2.70)

0.605∗ (1.92)

0.612∗∗ (2.20)

0.046∗∗∗ (5.94)

0.042∗∗∗ (6.30)

0.039∗∗∗ (5.81)

0.041∗∗∗ (6.27)

0.046∗∗∗ (6.02)

0.043∗∗∗ (5.94)

0.038∗∗∗ (5.31)

Total banks assets

-0.648∗∗∗ (-3.86)

-1.296∗∗∗ (-5.69)

-0.837∗∗∗ (-2.75)

-0.808∗∗∗ (-4.40)

-0.967∗∗∗ (-4.81)

-1.494∗∗∗ (-6.20)

-1.162∗∗∗ (-3.04)

-1.156∗∗∗ (-5.29)

Bank return on asset

1.042∗∗∗ (8.96)

0.851∗∗∗ (6.37)

0.882∗∗∗ (7.22)

1.067∗∗∗ (8.59)

1.093∗∗∗ (9.08)

0.870∗∗∗ (6.60)

0.901∗∗∗ (7.13)

1.115∗∗∗ (8.59)

Control of corruption

0.142∗∗ (1.97) 548

0.237∗∗ (2.24) 457

0.216∗∗∗ (2.90) 554

0.027 (0.26) 518

0.194∗∗ (2.45) 511

0.341∗∗∗ (3.17) 457

0.279∗∗∗ (3.41) 517

0.094 (0.80) 481

Observations

∗p < 0.1; ∗ ∗ p < 0.05; ∗ ∗ ∗p < 0.01

4.4. Robustness tests We perform several robustness tests to ensure the consistency of our findings. First, we check whether the previous empirical results are consistent with different specifications. Our empirical model may not include all the bank-specific factors or banking industry preference. Therefore, we reestimate the PSTR model by adding other control 26

variables available to us: the cost to income ratio, the ratio of foreign bank’s asset, the REER, the overhead cost ratio, and the ratio of regulatory capital. Not only does this approach help us to check the consistency of the estimated coefficients of our independent variables; it also helps to check the robustness of the location parameter c in the transition function g(qit , γ, c). Table 7 reports the robustness tests using F I 1 as the transition variable, whereas Table 8 shows the estimations of different specifications using F I 2 as the transition variable. Overall, the estimated coefficients of the two main independent variables and other control variables are robust in terms of their significance and signs. Furthermore, the estimated location parameters c in the transition function g(qit , γ, c) do not deviate greatly from the baseline model. Last but not least, we check for non-linearity using a quadratic regression model of the form: 0

Zscoreit = µi + β0 Opennessit + β1 Openness2it + Φ Controlit + uit

(12)

The estimations are reported in Table 9. The squared terms of financial openness are significantly positive, implying that the impact of financial openness on bank stability follows a U shape. However, the squared term of trade openness is not significant, which is consistent with the result of the linearity test in the lower panel of Table 3. 5. Conclusion This paper provides evidence on the impact of financial openness and trade openness on bank risk-taking behaviour. Using a PSTR approach for 42 emerging markets from 2000 to 2014, we show that the non-linear effect of financial openness on bank stability does exist. When the financial system is not sufficiently open, the impact of financial openness on bank stability is ambiguous. However, as the domestic financial market becomes more open, financial openness contributes positively to bank risk management. We also find evidence that the effects of financial openness and trade openness on bank 27

stability pass through the market discipline channel. However, the non-linear effect is not present in the case of trade openness. Trade openness increases bank stability linearly. In terms of policy implications, opening the goods and financial markets can help decrease bank risk-taking. However, there are only several mechanisms that ensure this effect because of the destabilising impact on the economy. Economic openness should pair up with other necessary conditions to be successful, such as discipline, transparency, and regulation. Better institutional quality and regulatory support can promote greater bank stability (Ahamed and Mallick, 2019; Mostak Ahamed and Mallick, 2017). For instance, the problem of financial fragility coming from financial openness can be mitigated by strong bank regulation and supervision (Klomp and de Haan, 2014). In Basel II, market discipline is an essential tool to control bank risk. Future study can look further into how effective this instrument is. In conclusion, to reap the benefits of economic openness, governments should put together other measures to strengthen the positions of their economies, including the real and the financial sectors, as the process of globalization continues.

28

Agca, S. and Celasun, O. (2012). Banking Sector Reforms and Corporate Borrowing Costs in Emerging Markets. Emerging Markets Finance and Trade, 48(sup4):71–95. Agoraki, M.-E. K., Delis, M. D., and Pasiouras, F. (2011). Regulations, competition and bank risk-taking in transition countries. Journal of Financial Stability, 7(1):38–48. Ahamed, M. M. and Mallick, S. (2017). Does regulatory forbearance matter for bank stability?

Evidence from creditors’ perspective.

Journal of Financial Stability,

28:163–180. Ahamed, M. M. and Mallick, S. K. (2019). Is financial inclusion good for bank stability? International evidence. Journal of Economic Behavior & Organization, 157:403–427. Ashraf, B. N. (2018). Do trade and financial openness matter for financial development? Bank-level evidence from emerging market economies. Research in International Business and Finance, 44:434–458. Ashraf, B. N., Arshad, S., and Yan, L. (2017). Trade Openness and Bank Risk-Taking Behavior: Evidence from Emerging Economies. Journal of Risk and Financial Management, 10(3):15. Ashraf, B. N., Zheng, C., and Arshad, S. (2016). Effects of national culture on bank risk-taking behavior. Research in International Business and Finance, 37:309–326. Benbouzid, N., Mallick, S. K., and Sousa, R. M. (2017a). Do country-level financial structures explain bank-level CDS spreads? Journal of International Financial Markets, Institutions and Money, 48:135–145. Benbouzid, N., Mallick, S. K., and Sousa, R. M. (2017b). An international forensic perspective of the determinants of bank CDS spreads. Journal of Financial Stability, 33:60–70. Berger, A. N., El Ghoul, S., Guedhami, O., and Roman, R. A. (2017). Internationalization and Bank Risk. Management Science, 63(7):2283–2301. 29

Bhaumik, S. K., Kutan, A. M., and Majumdar, S. (2018). How successful are banking sector reforms in emerging market economies? Evidence from impact of monetary policy on levels and structures of firm debt in India. The European Journal of Finance, 24(12):1047–1062. Bourgain, A., Pieretti, P., and Zanaj, S. (2012). Financial openness, disclosure and bank risk-taking in MENA countries. Emerging Markets Review, 13(3):283–300. Boyd, J. H. and Nicol´o, G. D. (2005). The Theory of Bank Risk Taking and Competition Revisited. The Journal of Finance, 60(3):1329–1343. Bui, D.-T. (2018). How Financial Freedom and Integration Change Public Debt Impact on Financial Development in the Asia-Pacific: A Panel Smooth Transition Regression Approach. Australian Economic Review, 51(4):486–501. Chinn, M. D. and Ito, H. (2006). What matters for financial development? Capital controls, institutions, and interactions. Journal of Development Economics, 81(1):163– 192. Cubillas, E. and Gonz´alez, F. (2014). Financial liberalization and bank risk-taking: International evidence. Journal of Financial Stability, 11:32–48. Fang, Y., Hasan, I., and Marton, K. (2014). Institutional development and bank stability: Evidence from transition countries. Journal of Banking & Finance, 39:160–176. Gonzalez, A., Ter¨asvirta, T., van Dijk, D., and Yang, Y. (2017). Panel Smooth Transition Regression Models. SSE/EFI Working Paper Series in Economics and Finance 604, Stockholm School of Economics. Gonzalo, J. and Wolf, M. (2005). Subsampling inference in threshold autoregressive models. Journal of Econometrics, 127(2):201–224. Hansen, B. E. (1999). Threshold effects in non-dynamic panels: Estimation, testing, and inference. Journal of Econometrics, 93(2):345–368. 30

Hellmann, T. F., Murdock, K. C., and Stiglitz, J. E. (2000). Liberalization, Moral Hazard in Banking, and Prudential Regulation: Are Capital Requirements Enough? American Economic Review, 90(1):147–165. Houston, J. F., Lin, C., Lin, P., and Ma, Y. (2010). Creditor rights, information sharing, and bank risk taking. Journal of Financial Economics, 96(3):485–512. IJtsma, P., Spierdijk, L., and Shaffer, S. (2017). The concentration–stability controversy in banking: New evidence from the EU-25. Journal of Financial Stability, 33:273–284. Kaufmann, D., Kraay, A., and Mastruzzi, M. (2011). The Worldwide Governance Indicators: Methodology and Analytical Issues. Hague Journal on the Rule of Law, 3(2):220–246. Kim, D.-H., Lin, S.-C., and Suen, Y.-B. (2010). Dynamic effects of trade openness on financial development. Economic Modelling, 27(1):254–261. Klomp, J. and de Haan, J. (2014). Bank Regulation, the Quality of Institutions, and Banking Risk in Emerging and Developing Countries: An Empirical Analysis. Emerging Markets Finance and Trade, 50(6):19–40. Lane, P. R. and Milesi-Ferretti, G. M. (2007). The external wealth of nations mark II: Revised and extended estimates of foreign assets and liabilities, 1970–2004. Journal of International Economics, 73(2):223–250. Lapteacru, I. (2017). Market power and risk of Central and Eastern European banks: Does more powerful mean safer? Economic Modelling, 63:46–59. Lepetit, L. and Strobel, F. (2015). Bank insolvency risk and Z-score measures: A refinement. Finance Research Letters, 13:214–224. Luo, Y., Tanna, S., and De Vita, G. (2016). Financial openness, risk and bank efficiency: Cross-country evidence. Journal of Financial Stability, 24:132–148.

31

Mirzaei, A., Moore, T., and Liu, G. (2013). Does market structure matter on banks’ profitability and stability? Emerging vs. advanced economies. Journal of Banking & Finance, 37(8):2920–2937. Mostak Ahamed, M. and Mallick, S. K. (2017). House of restructured assets: How do they affect bank risk in an emerging market?

Journal of International Financial

Markets, Institutions and Money, 47:1–14. Rajan, R. G. and Zingales, L. (2003). The great reversals: the politics of financial development in the twentieth century. Journal of Financial Economics, 69(1):5–50. Repullo, R. (2004). Capital requirements, market power, and risk-taking in banking. Journal of Financial Intermediation, 13(2):156–182. Seo, M. H. and Shin, Y. (2016). Dynamic panels with threshold effect and endogeneity. Journal of Econometrics, 195(2):169–186. Vives, X. (2006). Banking and Regulation in Emerging Markets: The Role of External Discipline. The World Bank Research Observer, 21(2):179–206. Wang, Q. (2015). Fixed-Effect Panel Threshold Model using Stata. The Stata Journal: Promoting communications on statistics and Stata, 15(1):121–134. Y¨ uksel, S. (2017). Determinants of the Credit Risk in Developing Countries After Economic Crisis: A Case of Turkish Banking Sector. In Global Financial Crisis and Its Ramifications on Capital Markets, Contributions to Economics, pages 401–415. Springer, Cham.

32

Table 7: Robustness tests using F I 1 as the transition variable; dependent variable : Log of Z-score

(1)

(2)

(3)

(4)

(5)

F I 1 (β0 )

-0.140∗∗ (-2.15)

0.135 (1.10)

0.0227 (0.17)

0.0210 (0.20)

0.0471 (0.49)

F I 1 (β1 )

0.0853∗∗∗ (4.27)

0.0630∗∗ (2.15)

0.0600∗∗∗ (2.63)

0.0667∗∗∗ (2.92)

0.0512∗∗∗ (2.91)

Bank liquid liabilities

0.194 (1.07)

0.611∗∗ (2.05)

0.985∗∗∗ (2.70)

0.639∗∗ (2.47)

0.595∗∗ (2.28)

Bank capital to assets

0.0456∗∗∗ (6.60)

0.0429∗∗∗ (5.05)

0.0381∗∗∗ (4.19)

0.0416∗∗∗ (5.88)

0.0400∗∗∗ (4.73)

Total banks assets

-0.557∗∗∗ (-3.30)

-1.028∗∗∗ (-4.31)

-1.287∗∗∗ (-4.46)

-0.990∗∗∗ (-4.59)

-0.938∗∗∗ (-4.37)

Bank return on assetes

0.141∗∗∗ (10.82)

0.158∗∗∗ (10.93)

0.186∗∗∗ (9.94)

0.136∗∗∗ (11.35)

0.145∗∗∗ (11.72)

Bank concentration

-0.312∗∗∗ (-2.92)

-0.371∗∗ (-2.50)

-0.0574 (-0.40)

-0.329∗∗∗ (-2.89)

-0.329∗∗∗ (-2.88)

GDP growth

0.000860 (0.00)

-0.606 (-1.38)

-0.669 (-1.43)

-0.250 (-0.71)

-0.270 (-0.78)

Control of corruption

0.182∗∗ (2.13)

0.280∗∗ (2.33)

0.127 (1.05)

0.247∗∗∗ (2.75)

0.235∗∗∗ (2.61)

Foreign bank

0.272∗ (1.88)

-0.115 (-0.50)

0.618∗∗ (2.32)

0.0825 (0.46)

0.0657 (0.36)

Cost to income

0.120 (0.70)

Foreign Asset

0.412 (1.55)

REER

-0.0187 (-0.08)

Overhead cost

0.439 (0.38)

Regulatory capital R2 Adjusted R2 Location parameters cj Slope parameters γj ∗p < 0.1; ∗ ∗ p < 0.05; ∗ ∗ ∗p < 0.01

0.449 0.435 4.52 4

0.460 0.442 4.67 334.5

0.476 0.455 4.75 10

0.438 0.424 4.67 5

0.187 (0.38) 0.448 0.434 4.70 8

Table 8: Robustness tests using F I 2 as the transition variable.

(1)

(2)

(3)

(4)

(5)

F I 2 (β0 )

-0.00621 (-0.19)

0.197∗∗∗ (2.70)

0.0581 (0.66)

0.0950 (1.63)

0.0870 (1.50)

F I 2 (β1 )

0.0982∗∗∗ (4.80)

0.0755∗∗∗ (2.94)

0.0776∗∗∗ (3.89)

0.0828∗∗∗ (3.89)

0.0828∗∗∗ (3.90)

Bank liquid liabilities

0.0851 (0.46)

0.296 (1.01)

0.573 (1.61)

0.345 (1.36)

0.286 (1.11)

Bank capital to assets

0.0444∗∗∗ (6.46)

0.0395∗∗∗ (4.70)

0.0338∗∗∗ (3.78)

0.0396∗∗∗ (5.60)

0.0400∗∗∗ (4.74)

Total banks assets

-0.595∗∗∗ (-3.81)

-0.880∗∗∗ (-4.03)

-1.161∗∗∗ (-4.48)

-0.860∗∗∗ (-4.35)

-0.800∗∗∗ (-4.09)

Bank return on assetes

0.139∗∗∗ (10.73)

0.159∗∗∗ (11.19)

0.193∗∗∗ (10.64)

0.134∗∗∗ (11.36)

0.144∗∗∗ (11.79)

Bank concentration

-0.310∗∗∗ (-2.88)

-0.279∗ (-1.83)

-0.0132 (-0.09)

-0.270∗∗ (-2.32)

-0.258∗∗ (-2.21)

GDP growth

0.0571 (0.19)

-0.724∗ (-1.71)

-0.853∗ (-1.88)

-0.236 (-0.69)

-0.304 (-0.89)

Control of corruption

0.187∗∗ (2.22)

0.257∗∗ (2.21)

0.0933 (0.81)

0.235∗∗∗ (2.67)

0.230∗∗∗ (2.62)

Foreign bank

0.0524 (0.33)

-0.554∗∗ (-2.22)

0.501∗ (1.81)

-0.219 (-1.12)

-0.221 (-1.12)

Cost to income

0.176 (1.02) 0.480∗ (1.84)

Foreign asset

REER

-0.125 (-0.61)

Overhead cost

0.270 (0.24)

Regulatory capital R2 Adjusted R2 Location parameters cj Slope parameters γj ∗p < 0.1; ∗ ∗ p < 0.05; ∗ ∗ ∗p < 0.01

0.455 0.441 4.10 4

0.476 0.459 3.99 344.5

0.496 0.476 3.92 10

0.450 0.436 4.06 5

0.000429 (0.00) 0.460 0.446 4.06 5

Table 9: Quadratic regression with individual fixed effects and time fixed effects, dependent variable: Logarithm of Z-score Fixed

Fixed& Time

Fixed

Fixed& Time

Fixed

Fixed& Time

Trade openness

0.040 (0.24)

0.349∗ (1.75)

Squared Trade

0.017 (0.92)

-0.017 (-0.71)

Financial openness 1

0.154 (1.32)

0.070 (0.47)

Squared Financial Openness 1

0.012 (1.05)

0.027∗ (1.87)

Financial openness 2

0.121 (1.55)

0.041 (0.43)

Squared Financial Openness 2

0.027∗∗ (2.54)

0.036∗∗∗ (2.81)

Bank liquid liabilities

0.966∗∗∗ (3.21)

0.625∗ (1.96)

1.058∗∗∗ (3.59)

0.610∗ (1.94)

0.846∗∗∗ (2.91)

0.531∗ (1.70)

Bank capital to assets

0.048∗∗∗ (6.21)

0.047∗∗∗ (6.08)

0.048∗∗∗ (6.32)

0.049∗∗∗ (6.38)

0.045∗∗∗ (5.85)

0.047∗∗∗ (6.11)

Total banks assets

-1.043∗∗∗ -1.071∗∗∗ -1.292∗∗∗ -1.361∗∗∗ -1.227∗∗∗ -1.145∗∗∗ (-4.54) (-4.66) (-5.33) (-5.50) (-5.55) (-5.15)

Bank return on asset

0.899∗∗∗ (6.45)

Bank concentration

-0.396∗∗∗ -0.304∗∗ (-3.11) (-2.28)

GDP growth

0.212 (0.53)

0.908∗∗ (1.97)

0.165 (0.41)

Control of corruption

0.221∗∗ (2.12)

0.228∗∗ (2.17)

Foreign bank

0.456∗∗ (2.09) 0.316 0.231 455

0.370 (1.57) 0.352 0.252 455

R2 Adjusted R2 Observations

0.978∗∗∗ (6.98)

∗p < 0.1; ∗ ∗ p < 0.05; ∗ ∗ ∗p < 0.01

35

0.946∗∗∗ (6.94)

1.007∗∗∗ (7.34)

0.979∗∗∗ (7.32)

1.040∗∗∗ (7.66)

-0.213∗ (-1.66)

-0.177 (-1.34)

0.762∗ (1.69)

-0.057 (-0.15)

0.662 (1.49)

0.294∗∗∗ (2.82)

0.346∗∗∗ (3.29)

0.260∗∗ (2.56)

0.305∗∗∗ (2.96)

0.362∗ (1.66) 0.327 0.243 455

0.122 (0.51) 0.367 0.269 455

-0.076 (-0.32) 0.349 0.268 455

-0.121 (-0.49) 0.378 0.282 455

-0.340∗∗∗ -0.227∗ (-2.70) (-1.73)

•

We control for both nonlinearity and heterogeneity in bank risk-taking.

•

We find a nonlinear impact of financial openness on bank risk-taking.

•

Beneficial effects on bank stability only appear at high levels of financial openness.

•

Trade openness only impacts bank risk-taking linearly.

•

Economic openness affects bank risk-taking through the market discipline channel.