Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis

Journal of Banking & Finance xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Journal of Banking & Finance journal homepage: www.elsevier.com/locate/jbf

Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis Michael Doumpos a, Chrysovalantis Gaganis b, Fotios Pasiouras c,d,⇑ a

School of Production Engineering and Management, Technical University of Crete, Greece Department of Economics, University of Crete, Greece c Surrey Business School, University of Surrey, UK d Financial Engineering Laboratory, Technical University of Crete, Greece b

a r t i c l e

i n f o

Article history: Received 18 August 2014 Accepted 25 April 2015 Available online xxxx JEL classification: G21 G28 Keywords: Bank soundness Central bank independence Crisis Supervisory architecture

a b s t r a c t Over the last fifteen years, many countries introduced reforms into the supervisory architecture of their financial sector. However, there is no evidence on whether specific supervisory arrangements were more successful than others during the crisis. Empirical evidence on the topic is in general scarce and there are reasonable theoretical arguments for and against alternative approaches. Similarly, while the effect of central bank independence on price stability has attracted a lot of attention, our knowledge with regards to its effect on bank soundness remains limited. Using a large sample of commercial banks operating in various countries over the period 2000–2011, this paper investigates whether and how bank soundness is influenced by central bank independence, central bank involvement in prudential regulation, and supervisory unification. We find that central bank independence exercises a positive impact on bank soundness, which in the case of smaller banks is enhanced during the crisis. Supervisory unification and the central bank involvement appear to mitigate the adverse effects of the crisis. The power of the supervisory authorities and bank size also appear to be conditional factors. Ó 2015 Elsevier B.V. All rights reserved.

1. Introduction There are many studies examining the impact of regulations like capital requirements, restrictions on activities, deposit insurance and private monitoring on bank risk-taking and soundness (e.g. Laeven and Levine, 2009; Agoraki et al., 2011). However, research on the relationship between the architecture of the supervision system and bank risk-taking remains scarce. The main purpose of this paper is to add to this strand of the literature by investigating whether and how bank soundness is influenced by central bank independence, central bank involvement in prudential regulation, and supervisory unification. Such an empirical analysis is not only timely but also necessary for various reasons. First, over the last fifteen years or so, several countries around the globe introduced reforms into the supervisory structure of their financial sector.1 However, there is no evidence on whether specific supervisory arrangements were more successful than others during the crisis. Second, theory and limited ⇑ Corresponding author at: Surrey Business School, University of Surrey, UK. E-mail addresses: [email protected], [email protected] (F. Pasiouras). Masciandaro and Quintyn (2009a) highlight that between 1998 and 2009, 70 out of the 102 countries that were considered in their study have chosen to reform their financial supervisory structure. 1

empirical evidence provide mixed results, and the financial crisis has re-opened a debate on the advantages and disadvantages of alternative arrangements. For example, the ones that support the involvement of central banks in micro-prudential supervision put forward arguments like access to better information, more effective crisis resolution, economies of scale, the ability to retain better equipped staff, etc. However, others argue that there are drawbacks like conflicts of objectives, reputational risk, scope diseconomies, etc. Additionally, despite the general belief that central bank independence is beneficial not only for price stability but also for financial stability, the theoretical model of Berger and Kißmer (2013) contradicts this view. Third, the delegation of powers among the regulatory agencies and the central bank involvement in supervision has recently gained considerable attention at the policy making level. For example, on July 2010 US President Barack Obama signed the Dodd–Frank Act into law, with one of its aims being to streamline banking regulation, and reduce competition and overlaps between different regulators. In Europe, the ECB is taking over the supervision of the largest banks, whereas in the UK the prudential supervision has been assigned back to the Bank of England. At the same time, the literature acknowledges the difficulties in determining the optimal level of financial supervision unification through a traditional cost–benefit analysis (Masciandaro, 2009), illustrating the need for

http://dx.doi.org/10.1016/j.jbankfin.2015.04.017 0378-4266/Ó 2015 Elsevier B.V. All rights reserved.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

2

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

empirical studies that will explicitly link the degree of unification with bank soundness. Fourth, while the spotlight is usually on specific regulatory rules, frameworks (e.g. Basel III) and approaches (e.g. micro- versus macro-prudential), it should be emphasized that the supervisory agencies are the ones that must develop and implement all these regulatory initiatives. We use a sample of up to 1700 commercial banks operating in around 90 countries over the period 2000–2011, and we aim to extend the literature in various ways. First, earlier studies at the bank level examine profitability (Barth et al., 2003), efficiency (Gaganis and Pasiouras, 2013), and other bank attributes like non-performing loans or liquidity (Barth et al., 2002). In contrast to these studies we use the Z-score, an overall indicator of bank soundness, which is inversely related to the probability of bank insolvency (see e.g. Laeven and Levine, 2009). Second, half of the existing studies – including both studies on financial stability (i.e. Klomp and de Haan, 2009; Dincer and Eichengreen, 2012) – are at the country level. This results not only on a limited number of observations, but most importantly on loss of information about important bank characteristics. For example, the literature suggests that the influence of monetary policy and micro-prudential regulations on banking outcomes varies across different levels of bank size and market power (Kashyap and Stein, 2000; Agoraki et al., 2011; Zaheer et al., 2013; Brissimis et al., 2014). Within this context, the use of bank-level data in the present study allows us to examine whether the effect of regulatory structures and central bank independence on risk-taking differs between banks of different size. Third, all but one of the existing studies that focus on supervisory characteristics and aspects of bank risk and performance examine periods prior to the crisis (i.e. Dincer and Eichengreen, 2012).2 In contrast, we make use of newly constructed databases on regulatory structures and central bank independence around the world, to provide evidence over a recent time period. Apparently, this is not just an issue of offering more recent evidence or updating a dataset. Rather, it allows us to consider reforms in the supervision regime. More importantly, it allows us to examine the impact of the supervisory architecture on bank soundness around the financial crisis, and determine what works and what does not work during difficult times. For example, Herring and Carmassi (2008) mention that during normal times, an integrated supervisor located outside the central bank has the potential to achieve economics of scope, and to mitigate conflicts of interests and moral hazard problems; however, the real question is how this framework will perform during a crisis. The authors also mention that there are a number of instances in which political interference (i.e. lack of independence) in macro- and micro-prudential supervision has precipitated or exacerbated crises. The rest of the manuscript is as follows. Section 2 provides a background discussion. Section 3 describes the data and methodology. Section 4 presents the results, and Section 5 concludes the study.

2. Background discussion As discussed in the previous section, policy makers must take decisions in relation to central bank independence, central bank involvement in prudential regulation, and supervisory unification. In all the cases, there are reasonable arguments for and against the alternative approaches, but relatively little empirical analysis. So, 2 There are also recent studies that examine other issues like the assignment of banking supervision to central banks (Dalla Pellegrina et al., 2013), and the association between supervisory structure and GDP growth (Masciandaro et al., 2013).

in this section we discuss these issues in turn drawing where possible on the related literature. 2.1. Central bank independence Contrary to the large literature on the relationship between central bank independence (CBI) and inflation, the work on financial stability is limited (Berger and Kißmer, 2013). Barth et al. (2003) highlight the importance of the independence of regulatory agencies for the well-functioning of the banks, mentioning that it allows the agencies to supervise the financial condition of banks in a professional and consistent fashion. Furthermore, Quintyn and Taylor (2002) argue that bank regulatory and supervisory independence matters for financial stability for the same reasons that CBI matters for monetary stability – i.e. among other things, it can be seen as a device for mitigating the economic costs that are associated with a time-inconsistency problem. In a similar manner, Cihák (2010) mentions that greater independence from outside pressures should translate in central banks that are less politically constrained in acting to prevent financial distress. In contrast, a central bank that is subject to a lower degree of independence could be captured by political interests associated with weak financial institutions. Along the same lines, Hutchison and McDill (1999) argue that ‘‘A ‘‘dependent’’ central bank closely aligned with the government may be more inclined to provide monetary finance to problem institutions, thereby creating an additional channel for the moral hazard problem’’. (p. 160). The above discussion, implies that CBI could have a direct impact on the well-functioning of the banks, in cases where the central bank is involved in prudential supervision. However, CBI may also have an indirect influence on bank soundness through monetary policy and price stability, regardless of whether prudential supervision is assigned to the central bank or not.3 In particular, most central banks have a mandate to pursue price stability as the primary objective of their monetary policy, with various studies suggesting that CBI will result in lower inflation.4 At the same time, there appears to be a negative association between inflation and individual bank soundness (e.g. Baselga-Pascual et al., 2013; Uhde and Heimeshoff, 2009) or banking crises (e.g. Demirgüç-Kunt and Detragiache, 1998; Boyd et al., 2014). Additionally, Boyd et al. (2014) highlight that economies that fail to decrease their inflation rates during and after a banking crisis have a much higher probability of experiencing subsequent crises.5 Nevertheless, focusing on inflation may also result in problems in the banking sector. For example, inflationary pressures may call for interest rate increases that could translate into lower financial stability. Mishkin (1996) argues that ‘‘The theory behind credit rationing can be used to show that increases in interest rates can be one factor that help precipitate a financial crisis’’. (p. 19). Along the same lines, Cukierman (1992) mentions that when banks experience fast and substantial increases in interest rates, they cannot pass them to their assets as fast as they pass them to their liabilities. This increases interest rate mismatches and, thus, market risk.6 Others refer to the so called ‘‘paradox of credibility’’. For instance, Borio (2013) argues that ‘‘the establishment of regimes yielding low and stable inflation, underpinned by central 3 We would like to thank an anonymous referee for an interesting comment that motivated us to distinguish between the direct and indirect impact discussed in this section. 4 See Arnone et al. (2006) for a review of the literature. 5 Inflation can also be a conditional factor as for whether the probability of a banking crisis may be higher either under competition or under monopoly (see the theoretical model of Boyd et al., 2004). 6 The idea that low levels of inflation and interest increases can have adverse effects is not new. Fisher (1933), in the debt-deflation theory of great depressions, also argues that an interest rise can lead to, among others, reduction in net worth, decrease in profits, more liquidation, and bank failures.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

bank credibility, can make it less likely for signs of unsustainable economic expansion to show up first in rising inflation and more likely for them to emerge first as unsustainable increases in credit and asset prices’’ (p. R6).7 Thus, the indirect impact of CBI through monetary policies and inflation targeting, is ambiguous. The results of the literature that examines the impact of supervisory independence on bank outcomes are mixed. For example, studies focusing on bank performance, report that supervisory independence does not influence return on assets (Barth et al., 2003), that it decreases profit efficiency (Gaganis and Pasiouras, 2013), and that it enhances technical efficiency (Barth et al., 2013a). The ones focusing on bank risk and stability obtain similar results. Klomp and de Haan (2009) report that CBI has a significant and robust positive impact on financial stability at the country level, a finding that is mainly due to political independence. Dincer and Eichengreen (2012) also support the view that supervisory independence can be beneficial, since they find a negative association with nonperforming loans (% GDP). Nonetheless, in another country-level study, Barth et al. (2002) find the relationship between supervisory independence and non-performing loans to be significant only at the 10% level and in specific estimations. Finally, the theoretical model of Berger and Kißmer (2013) predicts that the higher the central bank independence, the more likely it is to withhold the implementation of preemptive monetary tightening to maintain financial stability. Thus, they challenge the idea of a positive relationship between CBI and financial stability. 2.2. Central bank involvement in supervision In a review of the evolution of the supervisory regimes on 88 countries for the period 1998–2008, Masciandaro and Quintyn (2009b) conclude that there was a supervision consolidation trend outside the central bank. There are various theoretical arguments against having the central bank as a prudential supervisor, like moral hazard risk, reputational risk, loss of independence, increase of bureaucratic powers, and scope diseconomies (Barth et al., 2002, 2003; Masciandaro and Quintyn, 2009b; Cihák, 2010; Beck and Gros, 2012). Possibly, the most well discussed issue is the conflict of interest between monetary policy and bank supervision. Due to the potential problems discussed in Section 2.1 the central bank could decide to relax its monetary policy, in order to avoid adverse effects on bank profitability and solvency (Goodhart and Schoenmaker, 1995). In turn, this could generate an inflationary bias and impair the credibility of the central bank. However, not everyone agrees that there should be a strict separation between monetary policy and bank supervision (Beck and Gros, 2012). Others mention that the potential conflicts depend on the structure of the banking and financial system (Goodhart and Schoenmaker, 1995) or that the issue of conflicting mandates can be possibly addressed by extending the policy horizon (Cihák, 2010). Additionally, reasonable arguments exist in support of the central banks supervision, like the access to accurate and timely information, the ability of independent central banks to enforce actions, and the comparative advantage of central banks in recruiting and retaining the best staff (Barth et al., 2002, 2003; Masciandaro and Quintyn, 2009b; Cihák, 2010; Beck and Gros, 2012). The few studies that examine the relationship between central bank involvement in supervision and bank outcomes consider performance, and a couple of additional attributes like non-performing loans and capitalization. Barth et al. (2002) report 7 Building on the arguments about the ‘‘paradox of credibility’’ and on studies on the risk-taking channel, Montes and Peixoto (2014) provide empirical evidence that a higher credibility of the inflation targeting regime increases the risk exposure of Brazilian banks.

3

that central bank supervision does not influence capital adequacy, profitability or the ratio of non-interest revenues to total revenues; however, it results in lower liquidity risk (nonetheless, this is not robust across their estimations), lower bank overhead costs and higher non-performing loans. Unlike Barth et al. (2002), Dincer and Eichengreen (2012) find that the capital to assets ratio is higher in countries where the central bank is the lead supervisor. Additionally, consistent with Barth et al. (2002) they find some evidence that in these countries the nonperforming loans to GPD ratio is higher. Barth et al. (2003) partially contradict the findings of Barth et al. (2002), reporting a negative and significant relationship (though only at the 10% level) between the central bank being a supervisory authority and bank profitability; however, this is not robust across their specifications. Gaganis and Pasiouras (2013) measure performance with profit efficiency, to conclude that efficiency decreases as the number of the financial sectors that are supervised by the central bank increases. Summarizing, we can conclude that: (i) policymakers are likely to face a trade-off between expected benefits and costs of central bank involvement in supervision, and (ii) existing research focuses mostly on bank performance and with mixed results, whereas there are no studies examining the association between central bank involvement and bank soundness. 2.3. Supervisory unification The second main element in the supervisory structure is the degree of unification of powers, with the adoption of a single regulator being a common trend across developed and developing countries over the last two decades (Demaestri and Guerrero, 2005; Herring and Carmassi, 2008; Masciandaro and Quintyn, 2009b). Unsurprisingly, there is again an active debate on the advantages and disadvantages of integrated supervision.8 Arguments in support of a single integrated supervisor include, among others: (i) economies of scale and scope, due to elimination of overlaps, more efficient transmission and interpretation of information, better allocation of physical infrastructure and human capital, development and improvement of supervisory methods, etc. (ii) higher efficiency in the resolution of conflicts that may arise due to different goals of supervision, (iii) easier cooperation with other regulators at an international context, (iv) higher flexibility in decision making, (v) enhanced accountability and transparency (i.e. a single regulator has by definition the sole responsibility), (vi) elimination of the opportunities for regulatory arbitrage, (vii) recruitment and retention of more suitably qualified personnel, due to enhanced career opportunities, and (viii) the ability to supervise financial conglomerates, with activities that would otherwise span across many specialized regulators. However, there are also various arguments against an integrated approach in supervision. These include: (i) potential cultural conflicts within a single supervisor, (ii) drawbacks associated with the monopoly powers over an integrated supervisor (e.g. potential to over-regulate, bureaucracy and decreased flexibility, enhanced potential for regulatory capture), (iii) diseconomies of scale and scope, (iv) moral hazard problems due to implicit contracts, and (v) problems that may arise due to heterogeneous objectives. Apparently, if these drawbacks overweigh the advantages of unification, the effectiveness of prudential supervision will decrease, possibly allowing banks to take more risk. As before, existing evidence is scarce. The few studies discussed below focus mainly on performance, and they provide conflicting results. Barth et al. (2002) find that a system with multiple 8 See among others, Barth et al. (2003), Demaestri and Guerrero (2005), Herring and Carmassi (2008) and Masciandaro and Quintyn (2009b) for more detailed discussions of the pros and cons that we highlight in the text.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

4

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

supervisors results in lower bank capital ratios and higher liquidity. However, this finding no longer holds when transition economies are included in the analysis. Additionally, having one or multiple supervisors has no impact on other bank characteristics like non-performing loans or profitability. In contrast, Barth et al. (2003) provide evidence that having a single supervisor enhances bank profitability. Gaganis and Pasiouras (2013) reach the opposite conclusion as they find a positive association between the number of supervisors (i.e. moving from a single supervisor to a specialized authority or each sector) and bank profit efficiency. Finally, at a macro-level, Masciandaro et al. (2013) find some evidence that unification has a negative effect on real GDP growth; however, this is not significant across all their specifications. 3. Methodology, variables and data 3.1. Variables 3.1.1. Dependent variable Our dependent variable is the soundness of banks, measured by the Z-score. It is calculated as Z-score = (mean ROA + mean EA)/r(ROA), where ROA is the return on assets, EA is the equity to assets ratio, and r(ROA) is the standard deviation of the rate of return on assets.9 The Z-score is widely used in the banking literature and shows the number of standard deviations below the mean by which profits must fall in order to eliminate equity. Thus, it can be seen as an accounting-based measure of distance to default, with higher figures indicating higher solvency. In the regressions that we present in Section 4, we use the natural logarithm of the Z-score (ln Z hereafter) to control for non-linear effects and outliers (e.g. Demirgüç-Kunt et al., 2008; Laeven and Levine, 2009).10 3.1.2. Supervisory structure and Central bank independence To capture the degree of central bank independence (CBIND), we use yearly updated figures of the Cukierman et al. (1992) index, taken by Bodea and Hicks (2012). This index considers information about the CEO of the central bank (e.g. tenure, appointment, dismissal), the policy formulation (e.g. who formulates the policy, final word in resolution of conflicts), the objectives of the central bank (i.e. importance of price stability), and the limitations of lending to government (terms of lending, interest rates, etc.). CBIND ranges from zero to one, with values closer to one indicating higher central bank independence. We use information from the World Bank database on the Organization of Financial Sector Supervision (Melecky and Podpiera, 2013) to construct a yearly supervision unification index (SUI) that is similar to the one in Masciandaro (2009). In particular, we assign a value of 4 when there is a single authority for all 3 9 The mean and the standard deviation were calculated over a 3 years rolling time window, consistent with many other studies (e.g. Schaeck et al., 2012; Beck et al., 2013). 10 We actually calculated the ln Z as the natural logarithm of (Z-score of bank i in year t + |min Z-score of all banks| + 1). Thus, a constant equal to the absolute value of the minimum Z-score of all banks plus one is added to every bank’s Z-score so that the natural logarithm is taken of a positive number, since the minimum Z-score can take negative values (for theoretically failed banks). Thus, for the bank with the lowest value of Z-score, the dependent variable will be ln (1) = 0. This is consistent with many studies on bank efficiency, which take the natural logarithm of profits (losses) while adding such a constant (e.g. Berger and Mester, 1997; Maudos et al., 2002), and along the lines of the modification of the Z-score in Demirgüç-Kunt et al. (2008), Mobarek and Kalonov (2014), Jeon et al. (2014), among others. This allows us to consider banks with a negative Z-score and avoid truncating it at zero. In total there were 19 observations with negative Z-score values in the sample, the minimum value being 3.678. As a robustness test, we re-estimated all the specifications in Tables 2 and 3, while excluding the observations with a negative Z-score. Thus, in these regressions we used the natural logarithm of the Z-score without adding the constant. The results remain the same and they are not reported to conserve space. They are available from the authors upon request.

sectors (total number of supervisors = 1), a value of 3 when there is a single authority for banking & securities (supervisors = 2), a value of 2 when there is a single authority for insurance and securities or insurance and banking (supervisors = 2), and a value of 1 one when there is a specialized authority for each sector (supervisors = 3).11 Thus, higher figures indicate greater unification. Similarly, we use information from Melecky and Podpiera (2013) to construct a yearly index of central bank’s involvement in financial supervision (CBFA) as in Masciandaro (2009). This index takes the value of 1 when the central bank is not assigned the main responsibility for banking supervision, the value of 2 when the central bank has the main or sole responsibility for banking supervision, the value of 3 when the central bank has responsibility in any two sectors, and the value of 4 when the central bank has responsibility in all three sectors.12 Thus, higher figures indicate greater central bank involvement in the supervision of the financial sector. 3.1.3. Other country level variables As noted earlier, we are interested in examining whether the impact of the supervisory architecture and CBI on bank soundness differs between normal periods and financial crises. Therefore, instead of defining the crisis on the basis of an arbitrary chosen time period common to all the countries in our sample (e.g. 2007), we use information from the IMF database on systemic banking crises (Laeven and Valencia, 2012) to construct a country-specific indicator. Thus, for each country in our sample we take into account the start and the end of a systemic banking crisis (if any) over the period 2000–2011, and we construct a dummy variable that takes the value of 1 during years of crisis and 0 otherwise (CRISIS). Additionally, to control for overall differences between the various economies, we use a dummy variable that distinguishes between developed (DEVED = 1) and developing countries (DEVED = 0). In further regressions we take into account additional country-level variables that have been used in the literature, like real GDP growth, inflation, institutional environment, freedom in the banking and financial services industry, concentration in the banking sector, and bank private credit to GDP. Given our interest in the role of the central bank, we also consider the ratio of central bank assets to GDP. Higher figures of this ratio reveal that the central bank has more resources relatively to the size of its economy, to conduct its policy making and supervise banks. 3.1.4. Bank level variables The main bank-level variable that we use is size, measured by the natural logarithm of total assets. There are at least three reasons for which we take size into account. First, it may have an impact on risk-taking (Hakenes and Schnabel, 2011). Second, the 11 Masciandaro (2009) suggests to assign a higher value to the single supervisor for the banking sector and securities markets because of: (i) the predominant importance of banking intermediation and securities markets over insurance in every national financial industry, and (ii) the observation that there is a higher degree of integration between banking and securities supervision than between banking and insurance supervision in the group of integrated supervisory agency countries. Consequently, as he mentions, this results to a degree of concentration of powers, that is, other things being equal, greater. Our results do not change when we re-estimate our models while re-scaling the SUI in the range of 1 to 3 (i.e. we treat the two intermediate categories as one). 12 As discussed in Masciandaro (2009) such an approach considers the fact that, whatever the supervision regime, the monetary authority has responsibility in pursuing macrofinancial stability. Therefore, he suggests the assignment of a greater value (2 instead of 1) if the central bank is the sole or the main authority responsible for banking supervision. We also estimated our models while rescaling the CBFA in a 1–3 scale, as follows: value of 1 when the central bank is not assigned the main responsibility for banking supervision, value of 2 when the central bank has the main or sole responsibility for banking supervision, value of 3 when the central bank has responsibility in two or three sectors. The results remain the same.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

literature suggests that the impact of monetary policy and micro-prudential regulations varies with bank size and market power (Kashyap and Stein, 2000; Agoraki et al., 2011). Third, bank size and the associated too big to fail issues have received considerable attention in the literature (Carbó-Valverde et al., 2013), and the success of policies associated with the supervision of large banks could be related to the characteristics of the supervisory authorities. For example, large banks will be more complex and powerful, making it more difficult to discipline them. As in other studies that examine bank soundness, we control for additional bank characteristics like asset quality (impaired loans to total loans), liquidity (liquid assets to deposits and short term funding), and expenses management (cost to income).13

3.2. Methodology Our dataset has a multi-level setting with individual banks being nested in countries over a number of years. Consequently, we employ a Hierarchical Linear Modeling (HLM) approach also known as multi-level modeling.14 This approach has been recently used in cross-country studies that examine firm performance, capital structure decisions, corporate risk-taking, and IPOs (see e.g. Kayo and Kimura, 2011). HLM is superior to OLS because it accounts for the fact that our data have different levels of aggregation and it provides error terms that control for the potential dependency due to nesting effects, which is not the case with OLS. In particular, by modeling simultaneously regressions at both the bank- and country-level, multilevel models consider that banks within a country are more similar to one another than banks from different countries. Furthermore, the HLM framework allows the separation of the variance in bank soundness explained by the bank-level versus country-level attributes. The model is fitted using an iterative maximum likelihood algorithm in which the fixed and random effects are estimated simultaneously until the model converges.15 In its combined form the model can be written as follows:

ln Z ijt ¼ a þ bX ijt1 þ cW jt1 þ uij þ ej þ eijt |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} fixed components

random components

where ln Zijt is the natural logarithm of the Z-score for bank i in country j in year t, Xijt1 is a vector of lagged bank-level control variables, and Wjt1 is a vector of lagged country-level variables.16 The random variables ui,j and ej allow the intercept (a + ui,j + ej) to be random and unique to every bank and country. The term eijt is the residual. So, the above model assumes that the intercept is random whereas the slope coefficients are fixed. 13

We do not use profitability and capitalization as control variables in any of the regressions, since they were taken into account during the calculation of the Z-score. 14 The terms hierarchical linear models, multilevel models, mixed-effects models denote essentially the same modelling approach. We use these terms interchangeably in our discussion. 15 An alternative would be to use the restricted maximum likelihood estimation (REML). This is a special case of the MLE that partitions the likelihood under normality into two parts, one being free of the fixed effects. Maximizing this part yields the REML estimators. In a sense, this approach incorporates the degrees of freedom used to estimate fixed effects into the estimation of the variance components. As a robustness test, we re-estimated our models with the REML. The results remain the same. 16 Demirgüç-Kunt et al. (2008) highlight that an individual bank’s soundness is unlikely to have an impact on country level measures of supervisory quality, arguing that reverse causality should not be a serious concern in such studies. In any case, to alleviate any reverse causality problems we lag all the explanatory variables (with the exception of the dummy variables for the crisis and the developed countries) by one year (see e.g. Anginer et al., 2014).

5

3.3. Data We start the construction of our sample by considering all the commercial banks in the Bankscope database, with total assets in excess of $500 million. We exclude: (i) banks operating in countries for which there is no information about at least one of the three main variables of interest (i.e. CBIND, SUI, CBFA) 17; (ii) banks for which we do not have information for the calculation of the Z-score and the control variables. Our final sample includes up to 1756 commercial banks from 94 countries, resulting in an unbalanced dataset of 10,248 bank-year observations.18 The definition of all the variables and the sources of information are available in Appendix A. Table 1 presents descriptive statistics for bank-specific and country-specific variables.19 Appendix B presents the mean value of the Z-score, and its components, by individual country. While in the regressions we use the natural logarithm of the Z-score, the Appendix presents the values in levels to be more informative. The average Z-score is 56.029, whereas the average ln Z is 3.616. When we split the sample by the level of overall development, the corresponding figures are 59.668 (Z-score) and 3.650 (ln Z) for developed countries, and 50.207 (Z-score) and 3.561 (ln Z) for developing countries. Finally, the average Z-score is lower in countries that experience a crisis compared to the ones that do not.20 Turning to the main independent variables, the mean CBIND is 0.552, the mean SUI is 1.946, and the mean CBFA is 1.813.21 4. Results 4.1. Base results Table 2 presents our base model that includes the main explanatory variables, the bank-level control variables, and the dummy variables for the crisis and the country development status. This specification allows the maximum use of our sample. Then, in Table 3 we present additional models, where we control for the country-specific factors discussed in Section 3. Note that the number of observations varies across the regressions, depending on the missing values for the country-level variables. In all the cases, the LR test that compares the estimated model with linear regression favors the multi-level specification. The coefficient estimates in Table 2 show that central bank independence has a positive and statistically significant effect on bank soundness. Thus, our results are consistent with the ones of country-level studies on financial instability and credit risk (i.e. Klomp and de Haan, 2009; Dincer and Eichengreen, 2012). In contrast, the impact of unification is insignificant. Column (3) of Table 2 shows that central bank involvement in supervision carries a positive impact, albeit significant at the 10% only. Additionally, 17 Information for these three indices is available up to 2010. Therefore, we initially assign the values of 2010 to dependent variables from both 2010 and 2011. However, this is not an issue in our case, as due to the use of lagged independent variables we end up having 2009 values for CBIND, SUI, and CBFA assigned to the Z-score for 2010, and 2010 values of the independent variables assigned to the Z-score for 2011. 18 These figures correspond to the actual number of observations after lagging our explanatory variables once. 19 To conserve space we do not report all the pairwise correlations in a Table. The highest correlation coefficients are: 0.868 (DEVED and INSTDEV), 0.686 (FINFR and INSTDEV), 0.592 (INFL and INSTDEV), and 0.569 (GDPGR and SUI). The correlations between the three main variables of interest are: 0.146 (CBIND and SUI), 0.115 (CBIND and CBFA), -0.311 (SUI and CBFA). All the remaining figures are available upon request. 20 The average Z-score and ln Z equal 41.957 and 3.307, respectively, in the case of countries that experienced a crisis. The corresponding figures for the countries that have not experienced a crisis are 59.410 (Z-score) and 3.690 (ln Z). 21 An inspection of our dataset shows that there are 32 countries in the sample that experience at least one change in their CBIND over the period of our analysis, and 34 countries that experience a change in SUI and CBFA.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

6

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

Table 1 Descriptive statistics. Variable

N

Average

St. dev.

Z-score ln Z ZPR ln ZPR ZLR ln ZLR CBIND SUI CBFA SIZE COST LIQ IMPL INFL GDPGR FINFR INSTDEV CONC CREDIT CBGDP SPOWER TENURE SUPIND

10,303 10,303 10,303 10,303 10,303 10,303 9110 10,248 10,248 10,303 10,303 10,303 10,303 9956 10,303 10,301 10,191 10,191 9771 9500 9918 6273 8969

56.029 3.616 7.357 3.116 48.673 3.326 0.552 1.946 1.813 22.615 59.001 21.989 4.151 3.536 2.986 62.340 0.695 47.117 72.445 6.584 11.943 11.013 2.084

122.712 0.917 15.299 0.324 109.608 1.027 0.197 1.264 0.671 1.854 17.954 21.203 5.525 3.817 3.562 20.801 0.796 19.923 40.171 5.624 2.433 3.737 0.818

Note: These figures correspond to the actual number of observations after lagging our explanatory variables once. All the variables are defined in Appendix A.

Table 2 Base model.

4.2. Normal vs crisis period (1)

(3)

(4)

0.032*** [0.008] 0.008*** [0.001] 0.002*** [0.001] 0.030*** [0.002] 0.458*** [0.023] 0.030 [0.082] 4.937*** [0.191]

0.060* [0.035] 0.031*** [0.008] 0.008*** [0.001] 0.002*** [0.001] 0.031*** [0.002] 0.457*** [0.023] 0.069 [0.078] 4.875*** [0.199]

0.399*** [0.139] 0.006 [0.024] 0.030 [0.036] 0.037*** [0.008] 0.009*** [0.001] 0.002*** [0.001] 0.037*** [0.002] 0.445*** [0.024] 0.183** [0.079] 4.794*** [0.228]

Random-effects parameters Country-level variance 0.051 [0.016] Bank-level variance 0.184 [0.011] Residual variance 0.530 [0.009]

0.080 [0.019] 0.179 [0.010] 0.528 [0.008]

0.078 [0.018] 0.179 [0.010] 0.528 [0.008]

0.050 [0.016] 0.185 [0.011] 0.531 [0.009]

Countries Banks Yearly observations LR test – chi2

94 1756 10,248 1714.40***

94 1756 10,248 1671.24***

69 1493 9055 1271.52***

Fixed-effects parameters CBIND (t1)

(2

0.405*** [0.139]

SUI (t1)

0.034 [0.024]

CBFA (t1) SIZE (t1) COST (t1) LIQ (t1) IMPL (t1) CRISIS DEVED Constant

this becomes insignificant in the last column where we include simultaneously all three variables. Thus, it appears that when we examine the impact of these characteristics over the entire period, only central bank independence has a statistically significant impact on bank soundness. In terms of control variables, we find that smaller size, lower costs, lower liquidity and fewer non-performing loans result in higher bank soundness. The results as for the effect of a country’s overall development are mixed; however, we find robust evidence that bank soundness is substantially lower during periods of crisis. Table 3 presents the coefficient estimates when we include the additional control variables in the regressions. We first add them one-by-one in columns 1–6 to control for macroeconomic conditions (column 1), freedom in the banking and financial services industry (column 2), institutional environment (column 3), concentration in the banking sector (column 4), financial development (column 5), and central bank size (column 6). The last column presents the results when we include these variables simultaneously in the equation.22 The use of these control variables does not change our main results. That is, central bank independence continuous to have a positive and statistically significant effect on bank soundness, whereas the supervisory structure does no matter.23 We also re-estimate the specification with all the variables (Table 3, column 7), while excluding from our sample, in turn: (i) all the countries with less than ten observations per country, (ii) Japan (1281 observations), and (iii) USA (2973 observations). In all three cases, the results remain the same.24

0.038*** [0.008] 0.009*** [0.001] 0.002*** [0.001] 0.037*** [0.002] 0.446*** [0.024] 0.170** [0.076] 4.855*** [0.212]

74 1510 9110 1329.87***

Notes: Maximum likelihood estimates from a multi-level model with fixed and random components. The LR test compares the Estimated Model with Linear regression. The dependent variable is the natural logarithm of the Z-score, calculated as shown in Appendix A. The independent variables are defined in Appendix A; standard errors in brackets. *** Statistically significant at the 1% level. ** Statistically significant at the 5% level. * Statistically significant at the 10% level.

The above specifications account for the influence of systemic banking crises on individual bank soundness; however, they do not allow the effects of the supervisory structure and central bank independence to differ during crisis and normal periods. For example, the insignificance of CBFA and SUI in Tables 2 and 3 does not necessarily mean that they do not matter. Rather, there can be particular situations where supervisory structure might matter most, with the period of the crisis being such a case. During difficult times, supervisory authorities choose policy responses that can improve bank soundness, maintain the status quo or make things worse. We try to capture the effects of supervisory structure and independence during times of crisis through the use of interaction terms in our regressions. Consequently, we interact each one of the main explanatory variables of interest with the banking crisis dummy, and we focus on the interaction terms which are moments in time where independence and supervisory structure can become critical.25 The results in Table 4 show that all three interactions are 22 The last column does not include INSTDEV which is highly correlated with DEVED (correlation coefficient = 0.853). 23 In terms of the control variables, we find that the bank-level variables and the dummy for the crisis retain their sign and significance. The dummy for the developed countries is significant in some of the estimations, but this significance disappears when we include simultaneously all the variables in the last specification (column 7). We obtain the same results when we re-estimate the specification in column 7 while replacing the dummy variable DEVED by INSTDEV. Consistent with our expectations we find that inflation and GDP growth carry a negative and positive sign, respectively. Additionally, concentration in the banking sector is negatively associated with bank soundness, whereas central bank size (% GDP) exercises a positive impact on the bank soundness. Finally, financial freedom and financial development do not appear to matter. 24 We do not report these results to conserve space. They are available from the authors upon request. 25 As discussed in Braumoeller (2004), among others, ‘‘When a statistical equation includes a multiplicative term in order to capture interaction effects, the statistical significance of the lower order coefficients is largely useless for the typical purposes of hypothesis testing’’. The reason for this is that b1 captures the impact of X1 on Y when X2 = 0 (and vice versa), not the impact of X1 on Y in general (see e.g. Braumoeller, 2004; Brambor et al., 2006).

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

7

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx Table 3 Regression results: controlling for country-level characteristics. Fixed-effects parameters CBIND (t1) SUI (t1) CBFA (t1) SIZE (t1) COST (t1) LIQ (t1) IMPL (t1) CRISIS DEVED INFL (t1) GDPGR (t1)

(1)

(2) ***

0.480 [0.135] 0.011 [0.023] 0.011 [0.035] 0.037*** [0.009] 0.009*** [0.001] 0.002*** [0.001] 0.035*** [0.002] 0.402*** [0.024] 0.129 [0.075] 0.025*** [0.004] 0.029*** [0.003]

FINFR (t1)

(3) ***

0.399 [0.141] 0.006 [0.024] 0.031 [0.036] 0.037*** [0.008] 0.009*** [0.001] 0.002*** [0.001] 0.037*** [0.002] 0.446*** [0.024] 0.181** [0.080]

(4) ***

0.413 [0.142] 0.014 [0.025] 0.028 [0.036] 0.036*** [0.008] 0.009*** [0.001] 0.002*** [0.001] 0.037*** [0.002] 0.432*** [0.024]

(5) ***

0.399 [0.138] 00.001 [0.024] 0.036 [0.037] 0.036*** [0.008] 0.009*** [0.001] 0.002*** [0.001] 0.038*** [0.002] 0.436*** [0.024] 0.197** [0.078]

(6) ***

0.421 [0.143] 0.016 [0.026] 0.041 [0.038] 0.039*** [0.009] 0.009*** [0.001] 0.002*** [0.001] 0.037*** [0.002] 0.448*** [0.027] 0.197** [0.098]

0.002** [0.001]

CREDIT (t1)

0.000 [0.001]

4.836*** [0.233]

4.762*** [0.234]

4.805*** [0.229]

4.852*** [0.233]

4.830*** [0.234]

0.018*** [0.004] 4.680*** [0.238]

0.042 [0.016] 0.188 [0.011] 0.526 [0.009] 69 1482 8898 1231.35***

0.050 [0.016] 0.185 [0.011] 0.531 [0.009] 69 1493 9053 1263.21***

0.054 [0.017] 0.185 [0.011] 0.531 [0.009] 69 1493 9055 1278.38***

0.048 [0.016] 0.185 [0.011] 0.531 [0.009] 69 1490 8953 1245.51***

0.055 [0.018] 0.184 [0.011] 0.535 [0.009] 69 1457 8713 1217.33***

0.059 [0.018] 0.183 [0.011] 0.532 [0.009] 68 1431 8653 1244.85***

CBGDP (t1)

Bank-level variance Residual variance Countries Banks Yearly observations LR test – chi2

0.551*** [0.145] 0.020 [0.026] 0.023 [0.038] 0.037*** [0.009] 0.009*** [0.001] 0.002*** [0.001] 0.037*** [0.002] 0.403*** [0.029] 0.091 [0.093] 0.023*** [0.004] 0.031*** [0.003] 0.002* [0.001]

0.102** [0.045]

CONC (t1)

Random-effects parameters Country-level variance

0.467 [0.147] 0.012 [0.026] 0.041 [0.039] 0.036*** [0.009] 0.009*** [0.001] 0.002*** [0.001] 0.039*** [0.002] 0.447*** [0.024] 0.184** [0.086]

0.000 [0.001]

INSTDEV (t1)

Constant

(7) ***

0.003** [0.001] 0.001 [0.001] 0.019*** [0.004] 4.709*** [0.256] 0.050 [0.017] 0.187 [0.011] 0.524 [0.009] 67 1407 8489 1195.27***

Notes: Maximum likelihood estimates from a multi-level model with fixed and random components. The LR test compares the Estimated Model with Linear regression. The dependent variable is the natural logarithm of the Z-score, calculated as shown in Appendix A. The independent variables are defined in Appendix A; standard errors in brackets. *** Statistically significant at the 1% level. ** Statistically significant at the 5% level. * Statistically significant at the 10% level.

positive and statistically significant.26,27 Thus, it appears that central bank independence, supervisory unification, and central bank involvement in supervision mitigate the negative effects of the crisis on bank soundness.

26 We do not include all the interactions in a single regression due to the extremely high correlations, those being 0.925 (CBIND*CRISIS, CBFA*CRISIS), 0.858 (CBIND*CRISIS, SUI*CRISIS), and 0.702 (SUI*CRISIS, CBFA*CRISIS). Some scholars argue that centering can mitigate multicollinearity issues. However, Brambor et al. (2006) illustrate that the centered and uncentered models are algebraically equivalent, and they go on to argue that ‘‘we can unequivocally state that centering does not change the statistical certainty of the estimated effects and, therefore, cannot really mitigate any multicollinearity issues that exist. Scholars should stop justifying the use of centered variables or the omission of constitutive terms in interaction models by claiming that this reduces multicollinearity’’ (p. 71). 27 In addition to the variables presented in Tables 4–7, all the regressions include the variables of the specification in column 7 of Table 3. The coefficients of the control variables and the random-effects parameters have been omitted from the Tables to conserve space. All the results are available from the authors upon request.

4.3. The role of supervisory power Barth et al. (2004) summarize the advantages and disadvantages from granting broad powers to supervisors. They mention various circumstances under which strong official supervision can prevent managers from engaging in excessive risk-taking behavior, contributing to bank development, performance and stability. They also highlight various studies pointing to the fact that supervisors may use their powers to benefit favored constituents, attract campaign donations, and extract bribes, as well as that there can be agency problems between taxpayers and supervisors. In such cases, powerful supervision might be related to corruption or impede bank operations. To take these issues into account we use the official supervisory power index (SPOWER) by the World Bank surveys on bank regulation (Barth et al., 2013b), that captures the extent to which supervisors can change the internal organizational structure of the bank and/or take specific disciplinary action against managers, directors, shareholders, and auditors.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

8

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

Table 4 Regression results: exploring the role of the crisis. Fixed-effects parameters CBIND (t1) SUI (t1) CBFA (t1) CRISIS CBIND (t1)  CRISIS

(1)

Table 5 Regression results: exploring the role of official supervisory power. (2)

***

0.517 [0.147] 0.022 [0.026] 0.019 [0.038] 0.582*** [0.087] 0.317*** [0.143]

(3) ***

0.538 [0.148] 0.029 [0.027] 0.026 [0.039] 0.498*** [0.045]

0.062 [0.022]

4.738*** [0.257] 67 1407 8489 1198.81***

CBIND (t1) SUI (t1) CBFA (t1) CRISIS SPOWER (t1)

CBFA (t1)  CRISIS

Countries Banks Yearly observations LR test – chi2

0.541 [0.144] 0.019 [0.026] 0.003 [0.040] 0.589*** [0.088]

***

SUI (t1)  CRISIS

Constant

Fixed-effects parameters ***

4.702*** [0.257] 67 1407 8489 1202.77***

(1)

(2) ***

0.545 [0.145] 0.014 [0.027] 0.036 [0.039] 0.420*** [0.029] 0.020*** [0.007]

0.608 [0.146] 0.178** [0.073] 0.216* [0.110] 0.411*** [0.030] 0.059** [0.024] 0.015** [0.006] 0.024** [0.010]

4.681*** [0.269] 66 1390 8241 1201.04***

5.465*** [0.349] 66 1390 8241 1211.81***

SUI (t1)  SPOWER(t1) 0.092** [0.041] 4.745*** [0.256] 67 1407 8489 1194.53***

Notes: Maximum likelihood estimates from a multi-level model with fixed and random components; the LR test compares the Estimated Model with Linear regression; the dependent variable is the natural logarithm of the Z-score, calculated as shown in Appendix A; in addition to the report coefficients, the model also includes the following independent variables, which are not reported to conserve space: SIZE (t1), COST (t1), LIQ (t1), IMPL (t1), DEVED, INFL (t1), GDPGR (t1), FINFR (t1), CONC (t1), CREDIT (t1), CBGDP (t1); The random effects parameters are also not reported due to space limitations; all the independent variables are defined in Appendix A; standard errors in brackets. *** Statistically significant at the 1% level. ** Statistically significant at the 5% level. * Statistically significant at the 10% level.

Column (1) of Table 5 shows the results when we simply add SPOWER. It appears that supervisory power exercises a positive impact on bank soundness. In column (2) we add interactions of SPOWER with the two indicators of the supervisory structure. In both cases, the interaction enters with a positive and statistically significant coefficient. Thus, a unified supervisor with the power to take various actions, and a powerful central bank that is involved in the supervision of various financial sectors, are in position to improve bank soundness. Finally, in column (3) we introduce a three way interaction that accounts simultaneously for the supervisory structure, the supervisory power, and the crisis period.28 We find that the interaction SUI*SPOWER*CRISIS is positive and statistically significant, indicating that during banking crises the effect of a powerful unified supervisor on bank soundness is larger than in normal times. 4.4. The role of bank size As mentioned earlier, the characteristics of the supervisory framework and central bank independence may have a different impact on banks of different size. To take this into account we construct an indicator of relative size that varies by country and year, and we split the sample in two groups. In particular, we calculate the annual average bank size (i.e. natural logarithm of total assets) for each country.29 Then, we compare the annual size of each 28 As discussed in Braumoeller (2004), when the equation includes a three way interaction, then: (i) b123 describes the impact of a joint increase of X1, X2, and X3 on Y, and (ii) all other coefficients reflect the singular or joint impact of the independent variables to which their subscripts correspond on Y when all other independent variables are equal to zero. In other words, the highest order interaction becomes the main point of interest, whereas the significance level of the main effects and lower-order interactions are no longer useful. 29 Apparently, only countries for which we have information for at least two banks per year are taken into account in this part of the analysis.

(3) ***

CBFA (t1)  SPOWER(t1) SPOWER(t1)  CRISIS SUI (t1)  SPOWER(t1)  CRISIS CBFA (t1)  SPOWER(t1)  CRISIS Constant Countries Banks Yearly observations LR test – chi2

0.564*** [0.150] 0.149* [0.076] 0.181 [0.113] 0.359** [0.153] 0.043* [0.025] 0.012* [0.007] 0.019* [0.010] 0.027** [0.012] 0.007** [0.003] 0.007 [0.005] 5.341*** [0.355] 66 1390 8241 1215.24***

Notes: Maximum likelihood estimates from a multi-level model with fixed and random components; the LR test compares the Estimated Model with Linear regression; the dependent variable is the natural logarithm of the Z-score, calculated as shown in Appendix A; in addition to the report coefficients, the model also includes the following independent variables, which are not reported to conserve space: SIZE (t1), COST (t1), LIQ (t1), IMPL (t1), DEVED, INFL (t1), GDPGR (t1), FINFR (t1), CONC (t1), CREDIT (t1), CBGDP (t1); the random effects parameters are also not reported due to space limitations; all the independent variables are defined in Appendix A; standard errors in brackets. *** Statistically significant at the 1% level. ** Statistically significant at the 5% level. * Statistically significant at the 10% level.

individual bank with this figure, and we split the sample in two by classifying the banks as above average and below average size. We then estimate separate regressions for each group. Tables 6 and 7 present the coefficients for the main variables of interests. It appears that some of our earlier findings hold for only one of the two groups of bank size. In particular, our results can be summarized as follows. Central bank independence continues to have a positive impact on bank soundness, regardless of bank size; however, its impact is enhanced during the crisis only in the case of smaller banks.30,31 Additionally, our earlier finding that supervisory unification appears to mitigate the negative impact of the crisis on bank soundness appears to be true only in the case of larger banks. This could imply that a single supervisor with concentrated powers, flexibility in decision making, and the ability to supervise financial conglomerates, is appropriate for the supervision of large banks during difficult times. Similarly, the central bank involvement in supervision mitigates the adverse effects of the crisis only in the case of smaller banks. Furthermore, the positive and statistically significant interactions in column (1) of Table 7 show that supervisory power can be an important conditional factor for the effect of both supervisory structure indicators on the soundness of larger banks. However,

30 The positive and statistically significant impact of the main effect of CBIND is also observed in unreported regressions, where we do not include interaction terms. 31 The difference between large and small banks could be related to monetary policy decisions by independent central banks. For example, as discussed in Zaheer et al. (2013), large banks are possibly less influenced than small banks by monetary shocks because of their ability to raise funds from capital markets and issue wholesale deposits.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

9

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx Table 6 Regression results by bank size. Fixed-effects parameters

CBIND (t1) SUI (t1) CBFA (t1) CRISIS CBIND (t1)  CRISIS

Above average size

Below average size

Above average size

Below average size

Above average size

Below average size

(1)

(2)

(3)

(4)

(5)

(6)

0.366** [0.165] 0.04 [0.031] 0.061 [0.045] 0.484*** [0.120] 0.125 [0.196]

0.604*** [0.181] 0.044 [0.034] 0.000 [0.049] 0.688*** [0.121] 0.532** [0.206]

0.373** [0.169] 0.029 [0.032] 0.068 [0.046] 0.558*** [0.065]

0.682*** [0.179] 0.045 [0.034] 0.009 [0.049] 0.437*** [0.062]

0.383** [0.161] 0.013 [0.030] 0.059 [0.047] 0.437*** [0.117]

0.668*** [0.177] 0.042 [0.033] 0.040 [0.051] 0.752*** [0.125]

0.089*** [0.031]

0.029 [0.032] 0.012 [0.055] 4.252*** [0.410] 66 746 4300 551.85***

0.178*** [0.059] 5.056 *** [0.437] 64 849 4189 584.83***

SUI (t1)  CRISIS CBFA (t1)  CRISIS Constant Countries Banks Yearly observations LR test – chi2

4.255*** [0.409] 66 746 4300 553.46***

5.009*** [0.436] 64 849 4189 584.97***

4.161*** [0.414] 66 746 4300 561.12***

4.916 *** [0.435] 64 849 4189 582.40***

Notes: Maximum likelihood estimates from a multi-level model with fixed and random components. The LR test compares the Estimated Model with Linear regression. Banks have been classified as above average size or below average size by comparing the annual size of each individual bank (i.e. natural logarithm of total assets) with the the annual average bank size for each country. The dependent variable is the natural logarithm of the Z-score, calculated as shown in Appendix A. In addition to the report coefficients, the model also includes the following independent variables, which are not reported to conserve space: SIZE (t1), COST (t1), LIQ (t1), IMPL (t1), DEVED, INFL (t1), GDPGR (t1), FINFR (t1), CONC (t1), CREDIT (t1), CBGDP (t1). All the independent variables are defined in Appendix A. The random effects parameters are also not reported due to space limitations. Standard errors in brackets. *** Statistically significant at the 1% level. ** Statistically significant at the 5% level. * Statistically significant at the 10% level.

for smaller banks, supervisory power matters only in the case of a unified supervisor. Finally, our earlier finding showing that powerful unified supervisors have a higher impact during the crisis appears to be the case only for larger banks.

In this section we attempt to shed some additional light on the effect of central bank independence on bank soundness, a finding that appears to be quite robust across the estimations that we presented so far.32 We start by examining whether and how central bank independence influences the two components of the Z-score. Following Lepetit et al. (2008) and Barry et al. (2011), we disaggregate the Z-score into the ZPR = mean ROA/r(ROA), and the ZLR = mean EA/r(ROA). ZPR is an inverse indicator of bank’s portfolio risk, also known as an indicator of risk-adjusted returns. ZLR is an inverse indicator of leverage risk or in other words, an indicator of risk-adjusted capitalization. For consistency with our earlier analysis of the Z-score, we use the natural logarithm of these two indicators in our regressions.33 The results in Table 8 (Columns 1 and 2) show that central bank independence has a positive impact on both risk-adjusted returns and risk-adjusted capitalization.

The literature on central bank independence, suggests that the actual or de facto independence will not necessarily be the same as the legal independence (see Arnone et al., 2006). Additionally, the differences between the legal and de facto independence may be more important in developing than in industrial countries (Cukierman, 2008), and across different levels of democracy (Bodea and Hicks, 2012). With regards to the latter, Bodea and Hicks (2012) argue that this is due to the fact that democracies and dictatorships differ significantly in terms of the law enforcement, the process of law changes, transparency, and freedom of speech. Given the difficulties in assessing and quantifying the actual independence, we attempt to partially address this problem, through some additional estimations.34 In the regressions presented in Tables 2–7 we accounted for the impact of the overall development as well as for the institutional environment, the latter reflecting the development of a country in terms of voice and accountability, political stability and absence of violence, government effectiveness, regulatory quality, rule of law, and control of corruption. Thus, the inclusion of these variables in our regressions allows us to control for the potential effect of the two country attributes discussed in Cukierman (2008) and Bodea and Hicks (2012) on bank soundness. However, we have not tested yet, whether the effect of CBIND on bank soundness differs between developed and developing countries, or across different levels of institutional

32 We are indebted to an anonymous referee for various comments that motivated us to undertake all the analysis that we present in Section 4.5. 33 As in the case of the ln Z, we calculate the ln ZPR as the natural logarithm of (ZPR-score of bank i in year t + |min ZPR-score of all banks| + 1). This allows us to take into account 947 bank observations with losses that resulted in negative values for ZPR. In the case of ZLR, we calculate the ln ZLR as the natural logarithm of (ZPR-score of bank i in year t + 1), allowing us to take into account three observations with a value of zero for ZLR. As a robustness test, we re-estimated the specifications shown in columns (1) and (2) in Tables 8, while excluding the observations with a negative (zero) ZPR (ZLR). Thus, in these regressions we used the natural logarithm of the ZPR or ZLR without adding the constant. The results remain the same and they are not reported to conserve space. They are available from the authors upon request.

34 For example, Cukierman et al. (1992) develop an index of actual independence based on a questionnaire answered by specialists on monetary policy. While this is by all means a very interesting exercise, there are various problems associated with the use of such an index in our study. First, it covers only 23 countries. Second, the responses refer to the 1980s (see Cukierman et al., 1992, p. 367). Third, the survey responses are, in general, subjective. In our context, asking policy makers to assess the independence of their own central banks is likely subject to a measurement bias. This could be even more problematic when dealing with answers across different countries where the responders have entirely different cultures, backgrounds, etc., and there is no common benchmark as for the answers to the questions (for example what is good versus mixed versus poor adherence when it comes to intermediate policy targets).

4.5. Further evidence on central bank independence

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

10

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

Table 7 Regression results by bank size: the role of supervisory power and the crisis. Fixed-effects parameters

CBIND (t1) SUI (t1) CBFA (t1) CRISIS SPOWER (t1) SUI (t1)  SPOWER (t1) CBFA (t1)  SPOWER (t1)

Above average size

Below average size

Above average size

Below average size

Above average size

Below average size

(1)

(2)

(3)

(4)

(5)

(6)

0.469*** [0.162] 0.173* [0.094] 0.265* [0.139] 0.426*** [0.041] 0.073** [0.030] 0.015* [0.008] 0.032** [0.012]

0.728*** [0.187] 0.275** [0.105] 0.052* [0.158] 0.381*** [0.041] 0.042** [0.033] 0.022** [0.009] 0.006 [0.014]

0.349** [0.173] 0.080 [0.107] 0.099* [0.047] 0.309 [0.527] 0.007 [0.020] 0.006 [0.009]

0.679*** [0.186] 0.275** [0.119] 0.013 [0.051] 0.157 [0.616] 0.025 [0.022] 0.022** [0.010]

0.468** [0.168] 0.017 [0.032] 0.409** [0.159] 0.943 [0.601] 0.064** [0.029]

0.709*** [0.191] 0.037 [0.036] 0.190 [0.181] 0.965 [0.667] 0.014 [0.033]

0.046** [0.014] 0.058 [0.055]

0.014 [0.016] 0.025 [0.061]

0.326 [0.247] 0.035 [0.024] 5.137*** [0.503] 65 737 4181 563.58***

0.316 [0.271] 0.015 [0.027] 5.501*** [0.545] 62 834 4060 592.30***

0.072* [0.039] 0.313* [0.188] 0.037** [0.016]

SPOWER (t1)  CRISIS SUI (t1)  CRISIS SUI (t1)  SPOWER (t1)  CRISIS

0.040 [0.045] 0.099 [0.217] 0.008 [0.018]

CBFA (t1)  CRISIS CBFA (t1)  SPOWER (t1)  CRISIS Constant Countries Banks Yearly observations LR test – chi2

5.270*** [0.510] 65 737 4181 564.10***

5.736*** [0.561] 62 834 4060 593.61***

4.358*** [0.465] 65 737 4181 573.84***

5.586*** [0.508] 62 834 4060 586.99***

Notes: Maximum likelihood estimates from a multi-level model with fixed and random components. The LR test compares the Estimated Model with Linear regression. Banks have been classified as above average size or below average size by comparing the annual size of each individual bank (i.e. natural logarithm of total assets) with the the annual average bank size for each country. The dependent variable is the natural logarithm of the Z-score, calculated as shown in Appendix A. In addition to the report coefficients, the model also includes the following independent variables, which are not reported to conserve space: SIZE (t1), COST (t1), LIQ (t1), IMPL (t1), DEVED, INFL (t1), GDPGR (t1), FINFR (t1), CONC (t1), CREDIT (t1), CBGDP (t1). The random effects parameters are also not reported due to space limitations. All the independent variables are defined in Appendix A. Standard errors in brackets. *** Statistically significant at the 1% level. ** Statistically significant at the 5% level. * Statistically significant at the 10% level.

development. Therefore, we now add in the regressions the interactions of CBIND with overall development (CBIND*DEVED), and CBIND with institutional development (CBIND*INSTDEV).35 Both interactions are insignificant, indicating that there are no significant differences between developed and developing countries, or across different levels of institutional development (Table 8, columns 3 and 4). As a further test, we remove from our sample 643 observations from eight countries with dictatorial regimes, and we re-estimate the specification with all the variables (i.e. the one of Table 3, column 7). Restricting the sample to democratic regimes does not change our results.36 In further results that we present in Section 4.6 we also replace CBIND by an alternative indicator that serves as a proxy of de facto supervisory independence. One could claim that the supervisory-independence channel that we document in the present study requires the central bank’s involvement in financial supervision. Nonetheless, the so far obtained results tend to suggest that bank soundness is not significantly influenced by the central bank involvement in supervision, with the exception of the smaller banks during the crisis. To explore this issue further, we use the interaction of CBIND with

35 As mentioned earlier, the correlation between DEVED and INSTDEV is 0.868, so we do not include these two variables simultaneously in the regression. 36 To classify the countries in democratic regimes (parliamentary democracy, semi-presidential democracy, presidential democracy), and dictatorial regimes (civilian dictatorship, military dictatorship, royal dictatorship) we use information from Cheibub et al. (2010). The results are available upon request.

CBFA. We find this interaction to be insignificant, indicating that there are no important differences across the different levels of central bank involvement in financial supervision.37 This result provides some support to our earlier argument that central bank independence may also have an indirect impact on bank soundness, through monetary policy and inflation targeting. Consistent with this view, our results document a negative relationship between inflation and bank soundness that is robust across all our estimations.38

4.6. Alternative indicators of supervisory independence In this section, we present the results of additional regressions in which we replace CBIND by two alternative indicators taken from Barth et al. (2013b). One interesting aspect of these two indicators is that they reveal what happens in the case of bank supervisory agencies in general, regardless of whether these agencies are 37 The results remain the same when we replace CBFA by a dummy variable that takes the value of 1 in the case of central bank’s involvement in financial supervision (i.e. we merge the categories previously coded as 2, 3, and 4) and 0 otherwise. 38 Clearly, our approach provides just a preliminary look at the issue of inflation targeting and bank soundness. However, conducting a monetary policy study as the one of Delis and Kouretas (2011) or Altunbas et al. (2014) is outside the scope of the present study. Nonetheless, within this context, it should be mentioned that while these studies examine the impact of interest rates, which is an operating instrument, our results show the relationship between the final objective of monetary policy (i.e. inflation) and bank soundness.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

11

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx Table 8 Regression results: Further evidence on central bank and supervisory independence. Fixed-effects parameters CBIND (t1) SUI (t1) CBFA (t1) DEVED

(1)

(2) **

0.087 [0.040] 0.004 [0.008] 0.002 [0.011] 0.024 [0.025]

(3) ***

0.673 [0.167] 0.022 [0.030] 0.040 [0.044] 0.040 [0.044]

(4) ***

***

0.640 [0.179] 0.023 [0.026] 0.030 [0.039] 0.267 [0.229]

0.605 [0.162] 0.026 [0.027] 0.023 [0.039]

INSTDEV (t1)

(5)

(6)

(7)

0.517 [0.414] 0.020 [0.026] 0.010 [0.150] 0.089 [0.095]

0.007 [0.033] 0.083* [0.046] 0.130 [0.129]

0.001 [0.027] 0.054 [0.040] 0.010 [0.095]

0.121 [0.127]

CBIND (t1)  DEVED

0.267 [0.318]

CBIND (t1)  INSTDEV (t1)

0.092 [0.181]

CBIND (t1)  CBFA (t1)

0.018 [0.200] 0.013*** [0.005]

TENURE (t1) SUPIND (t1) Constant Countries Banks Yearly observations LR test – chi2

3.336*** [0.080] 67 1407 8489 776.57***

4.481*** [0.293] 67 1407 8489 1329.66***

4.634*** [0.271] 67 1407 8489 1195.36***

4.681*** [0.267] 67 1407 8489 1195.58***

4.734*** [0.382] 67 1407 8489 1194.07***

4.443*** [0.294] 78 1194 5492 1037.78***

0.037** [0.017] 4.923*** [0.241] 86 1560 8143 1268.77***

Notes: Maximum likelihood estimates from a multi-level model with fixed and random components. The LR test compares the Estimated Model with Linear regression. Banks have been classified as above average size or below average size by comparing the annual size of each individual bank (i.e. natural logarithm of total assets) with the the annual average bank size for each country. The dependent variable in column (1) is the natural logarithm of the ZPR, calculated as shown in Appendix A. The dependent variable in column (2) is the natural logarithm of the ZLR, calculated as shown in Appendix A. The dependent variable in columns (3) to (7) is the natural logarithm of the Zscore, calculated as shown in Appendix A. In addition to the report coefficients, the model also includes the following independent variables, which are not reported to conserve space: SIZE (t1), COST (t1), LIQ (t1), IMPL (t1), INFL (t1), GDPGR (t1), FINFR (t1), CONC (t1), CREDIT (t1), CBGDP (t1). The random effects parameters are also not reported due to space limitations. All the independent variables are defined in Appendix A. Standard errors in brackets. *** Statistically significant at the 1% level. ** Statistically significant at the 5% level. * Statistically significant at the 10% level.

a central bank or another regulatory body. Thus, they also relate to the supervisory-independence channel discussed earlier. First, we replace CBIND by the actual average tenure (years) of professional bank supervisors (TENURE) and we re-estimate our model. As this corresponds to actual figures of tenure it can been seen as a rough proxy of de facto supervisory independence that may be more suitable for developing countries.39 As shown in column 6 of Table 8, we continue to find that supervisory independence, as measured by TENURE, has a positive influence on bank soundness. The central bank independence indicator (CBIND) that we have used so far examines various aspects with regards to the appointment and dismissal of the chief executive officer; however, one could argue that it gives particular emphasis on the monetary policy and the objective of price stability. As a result, in cases where the central bank is involved in prudential supervision, an implicit 39 Cukierman et al. (1992) suggest the use of the central bank’s governor’s turnover ratio as a measure of de facto independence. However, this indicator is not available in our case, and we use the average tenure of the professional bank supervisors. We believe that our indicator can serve as a proxy for de facto independence for similar reasons to the ones discussed in Cukierman et al. (1992). For example, a frequent turnover may reflect the removal of those bank supervisors that do not follow the guidelines of the Government or do not comply with its interventions. Following Cukierman et al. (1992), it is likely that turnovers of supervisory agents that occur simultaneously with or shortly after changes in the government would indicate lower independence that turnovers that last more than the electoral cycle. Quintyn and Taylor (2002) also argue that: (i) appropriate salary levels for the supervisors and clear career streams, and (ii) the terms of appointment – and even more critical-dismissal of its senior personnel, can help to ensure supervisory independence. Of course, as mentioned in Cukierman et al. (1992), a lower turnover does not necessarily imply a high level of independence. Additionally, a high turnover could capture other attributes beyond our control like experience, etc.

assumption that needs to be made is that the central bank’s supervisory independence is closely linked or even identical with its degree of independence in monetary policy.40 As a final exercise, we replace CBIND by an alternative overall indicator of the independence of the supervisory agency (SUPIND), taken by Barth et al. (2013b). SUPIND reflects: (i) the degree to which the supervisory authority is independent from political influence, in terms of responsibility and accountability, (ii) the degree to which the supervisory authority is legally protected from the banking industry,41 and (iii) the degree to which the supervisory authority is able to make decisions independently of political considerations, as reflected in the existence of a fixed term for the head of the supervisory agency (and other directors). A similar indicator has been used in Barth et al. (2003, 2004, 2013a) and Gaganis and Pasiouras (2013). As in the case of TENURE this indicator reveals what happens in the case of the bank supervisory agency, regardless of whether this is the central bank or a separate regulatory body, and therefore it is not

40 As mentioned by an anonymous referee, one could question this assumption. For example, Blinder (2010) argues that ‘‘. . ., people often forget that Federal Reserve ‘‘independence’’ is far stronger in monetary policy than in, e.g., bank supervision and regulation’’. (p. 125). As we discuss in the text, this is an assumption that needs to be made rather than a fact that will necessarily hold in all the cases. However, we believe that to some extent this will be the case in many countries. For example, the terms of the appointment and the dismissal of the chief executive officer of the central bank, which are part of the CBIND indicator could matter for both the formulation of the monetary policy and banking supervision, and as such we would expect them to be related to both the monetary and the supervisory independence of the central bank. 41 Quintyn and Taylor (2002) also highlight that bank supervisors should enjoy legal protection when executing their job. They argue that, when this is not the case, supervisors are afraid of imposing sanctions or enforcing them, since they could be personally sued by the supervised institution.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

12

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

subject to the aforementioned assumption. The results in Table 8 (column 7) show that SUPIND has a positive and statistically significant effect on bank soundness.

5. Conclusions In recent years, there has been a lively discussion about the potential need to reform the structure of the financial supervision, whereas even the independence of the Fed has been described to be ‘‘under some attack’’ (Blinder, 2010). This paper presents new evidence on the effect of central bank independence, central bank involvement in prudential regulation and supervisory unification on bank soundness around the years of the crisis. Using an international sample of banks, the results of this study could be of interest to both policy makers and supervisory authorities. First, they suggest that the independence of the central bank and other supervisory agencies influences the soundness of banking institutions. Second, the supervisory architecture appears to matter only during difficult times (i.e. period of the crisis) and it also depends on the power of the supervisors to take specific actions. Finally, our results imply that the impact of these characteristics may vary with bank size. For example, the impact of central bank independence is enhanced during the crisis; however this happens only in the case of smaller banks. Similarly, our finding that a unified supervisor mitigates the adverse effects of the crisis appears to be driven by the group of the larger banks, although we also find that a powerful unified supervisor matters for the soundness of smaller banks.

A next step in our analysis would be to further identify precise mechanisms through which supervisory structure and central bank independence help during times of crisis. For example, one could try to collect information on the specific actions taken by regulators, the speed of their reaction, etc. Unfortunately, such data are not currently available, but we hope that they will be incorporated in future research. Similarly, one could also try to explain why a unified supervisor matters for larger banks, whereas a central bank matters for smaller banks during the crisis. In an attempt to do so, one would need information on the main goals, the mandate of the supervisor, the relative importance placed on specific activities in banking supervision (e.g. regulatory compliance, risk profile and strategy). While the World Bank has started collecting such data, they became available only very recently, and they correspond to a period that is after the one that we examine. Thus, we hope that they will be part of future research. Acknowledgments We would like to thank an anonymous referee and Ike Mathur (Handling Editor) for valuable comments that helped us improve earlier versions of the manuscript. Special thanks are due to Christina Bodea and Raymond Hicks for providing the updated data on central bank independence. Thanks are also due to participants of the 4th National Conference of the Financial Engineering and Banking Society (Athens, 2013), the 4th International Conference of the Financial Engineering and Banking Society (Guildford, 2014), and the 13th Eurasia Business and Economics Society Conference (Istanbul, 2014) for their comments. Any remaining errors are our own.

Appendix A. Definition of variables ln Z

ln ZPR

ln ZLR CBIND

SUI

CBFA

SIZE COST LIQ IMPL

Indicator of bank soundness, calculated as natural logarithm of (Z-score of bank i in year t + |min Z-score of all banks| + 1), where Z-score of bank i in year t = (3 years average return on assets ratio + 3 years average equity to assets ratio)/3 years standard deviation of return on assets Indicator of bank risk adjusted returns, calculated as natural logarithm of (ZPR-score of bank i in year t + |min ZPR-score of all banks| + 1), where ZPR-score of bank i in year t = 3 years average return on assets ratio/3 years standard deviation of return on assets Indicator of bank risk adjusted capitalization, calculated as natural logarithm of (ZLR-score of bank i in year t + 1), where ZLR-score of bank i in year t = 3 years average of equity to assets/3 years standard deviation of return on assets Index of central bank independence. Updated figures of the Cukierman, Webb, and Neyapti index which is based on a weighted calculation of 16 indicators in 4 categories regarding the Chief Executive Officer, Policy Formation, Objectives, and Limitations on Lending to the Government. The indicators cover among others issues like: who appoints the Governor of the Bank, how long is the Governor’s appointment, who formulates monetary policy, does the central bank have the stated objective of price stability, and does the central bank set the terms of lending to the government. For each indicator, possible scores run in intervals from 0 to 1 with the intervals depending on the number of categories. Scores from the individual indicators are then aggregated into their broader categories. The weights are applied to each of the aggregate scores and then the totals are summed. The overall scores range from a possible 0 to a possible 1, with values closer to 1 indicating higher central bank independence. Source: Bodea and Hicks (2012) Index of supervision unification. It takes the following values: 4 = there is a single authority for all 3 sectors (total number of supervisors = 1), 3 = there is a single authority for banking & securities (supervisors = 2), 2 = there is a single authority for insurance and securities or insurance and banking (supervisors = 2), 1 = there is a specialized authority for each sector (supervisors = 3). Source: Author’s calculations based on information from the World Bank database on the Organization of Financial Sector Supervision constructed by Melecky and Podpiera (2013) Index of central bank’s involvement in financial supervision. It takes the following values: 1 = the central bank is not assigned the main responsibility for banking supervision, 2 = the central bank has the main or sole responsibility for banking supervision, 3 = the central bank has responsibility in any two sectors, 4 = the central bank has responsibility in all three sectors. Source: Author’s calculations as in Masciandaro (2009) based on information from the World Bank database on the Organization of Financial Sector Supervision constructed by Melecky and Podpiera (2013) Natural logarithm of total assets in US dollars (Source: Bankscope) Cost to income ratio (Source: Bankscope) Liquid assets to deposits and short term funding ratio (Source: Bankscope) Impaired loans to Gross loans ratio (Source: Bankscope)

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

13

Appendix A (continued) CRISIS DEVED INSTDEV

INFL GDPGR FINFR

CONC CREDIT CBGDP SPOWER

TENURE SUPIND

Country specific dummy variable that takes the value of 1 during years of systemic banking crisis and 0 otherwise (Source: IMF database on Systemic Banking Crises, constructed by Laeven and Valencia (2012)) Dummy variable that takes the value of 1 in the case of major advanced and advanced countries and 0 otherwise (Source: IMF) Overall indicator of institutional development, calculated as the average of six indicators accounting for: voice and accountability, political stability and absence of violence, government effectiveness, regulatory quality, rule of law, control of corruption. It takes values from 2.5 to 2.5, with higher scores corresponding to better outcomes. (Source: World Governance Indicators Database) Average consumer price Index (annual% change) (Source: WB GFD Database) Real annual GDP Growth (Source: GMID) Freedom in the banking and financial services industry. The Index scores an economy’s financial freedom by looking into the following five broad areas: (i) the extent of government regulation of financial services, (ii) the degree of state intervention in banks and other financial firms through direct and indirect ownership, (iii) the extent of financial and capital market development, (iv) government influence on the allocation of credit, and (v) openness to foreign competition. An overall score on a scale of 0–100 is given to an economy’s financial freedom through deductions from the ideal score of 100. (Source: Heritage Foundation) Concentration in the banking sector, calculated as assets of three largest commercial banks as a share of total commercial banking assets. (Source: WB GFD database) The ratio of private credit by deposit money banks to GDP (%) used as proxy for the development of the banking sector. (Source: WB GFD database) The ratio of central bank assets to GDP (%) used as an indicator of the relative size of a central bank. (Source: WB GFD database) Index of official supervisory power. It measures the degree to which the country’s bank supervisory agency has the authority to take specific actions. It takes into account the answers to the following questions: (1) Does the supervisory agency have the right to meet with external auditors about banks? (2) Are auditors required to communicate directly to the supervisory agency about elicit activities, fraud, or insider abuse? (3) Can supervisors take legal action against external auditors foe negligence? (4) Can the supervisory authority force a bank to change its internal organizational structure? (5) Are off-balance sheet items disclosed to supervisors? (6) Can the supervisory agency order the bank’s directors or management to constitute provisions to cover actual or potential losses? (7) Can the supervisory agency suspend the directors’ decision to distribute (a) dividends, (b) bonuses, and (c) management fees? (8) Can the supervisory agency supersede the rights of bank shareholders and declare a ban insolvent? (9) Can the supervisory agency suspend some or all ownership rights? (10) Can the supervisory agency (a) supersede shareholder rights, (b) remove and replace management, and (c) remove and replace directors? The SPOWER index takes values between 0 and 14, with larger numbers indicating greater power. Source: Barth et al. (2013b) The average tenure of banking supervisors (i.e. average number of years that staff members have been supervisors). Source: Barth et al. (2013b) Overall indicator of the independence of the bank supervisory authority. It reflects the degree to which the supervisory authority is independent from the government and legally protected from the banking industry. It takes into account the answers to the following three questions: (1) To whom is the supervisory agency legally responsible or accountable? (i) A legislative body, such as Parliament or Congress, (ii) The head of government (e.g. President, Prime Minister), (iii) The Finance Minister or other cabinet level official. This takes the value of one when the answer is Parliament or Congress, and the value of zero otherwise, (2) Are supervisors legally liable for their actions (e.g., if a supervisor takes actions against a bank can he/she be sued)? This takes the value of zero when the answer is yes, and the value of one otherwise, (3) Does the head of the supervisory agency (and other directors) have a fixed term? If yes, how long is the term? This takes the value of one when there is a fixed term of 4 years or greater, and the value of zero otherwise. The SUPIND index is calculated as the summation of the above three values. Therefore, it takes values between 0 and 3, with higher values indicating greater independence. Source: Barth et al. (2013b)

Notes: WB: World Bank; GFD: Global Financial Development; GMID: Global Market Information Database.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

14

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

Appendix B. Mean value of dependent variables, by country

Albania Algeria Argentina Australia Austria Azerbaijan Belarus Belgium Bolivia Bosnia & Herzegovina Botswana Brazil Bulgaria Canada Chile China Colombia Costa Rica Croatia Cyprus Czech Rep. Denmark Dominican Rep. Ecuador Egypt El Salvador Estonia Finland France FYROM Georgia Germany Greece Guatemala Guyana Honduras Hong Kong Hungary Iceland India Indonesia Ireland Israel Italy Jamaica Japan Jordan Kazakhstan Kenya Latvia

Z-score

ZPR

ZLR

Obs.

34.455 52.301 37.680 104.731 83.804 16.783 50.035 8.157 32.288 47.753 32.794 31.436 52.315 57.519 50.919 45.096 81.330 54.208 80.504 19.563 45.292 53.621 46.965 53.794 75.899 60.159 40.346 30.283 40.464 98.715 18.981 48.102 13.508 63.931 40.380 90.198 58.335 61.027 17.142 50.741 43.430 42.061 30.705 55.352 63.653 58.217 54.616 18.720 46.447 19.794

4.841 12.329 10.060 16.527 5.716 2.916 6.671 0.343 3.295 3.166 10.913 5.349 4.826 7.805 8.211 8.059 15.506 7.384 7.069 1.741 7.428 7.131 9.839 7.587 7.482 5.884 7.214 3.543 5.005 4.657 2.263 5.138 0.726 12.658 11.198 12.885 8.448 8.363 2.430 9.118 6.796 6.924 3.838 5.918 15.971 3.292 7.627 1.967 9.375 1.480

29.614 39.972 27.620 88.205 78.089 13.867 43.364 7.814 28.994 44.587 21.881 26.087 47.488 49.714 42.708 37.037 65.824 46.824 73.434 17.822 37.865 46.490 37.125 46.207 68.417 54.275 33.132 26.740 35.459 94.059 16.718 42.963 12.781 51.274 29.182 77.312 49.887 52.664 14.712 41.624 36.634 35.137 26.868 49.434 47.682 54.926 46.989 16.753 37.072 18.314

8 3 2 106 28 31 15 13 2 11 38 408 44 182 97 252 30 12 14 21 50 114 19 20 17 41 14 14 71 9 25 54 31 14 4 2 104 39 14 350 244 45 87 512 20 1281 85 31 35 41

Lebanon Lithuania Luxembourg Malaysia Malta Morocco Mauritius Mexico Moldova Mongolia Namibia Netherlands New Zealand Nicaragua Nigeria Norway Pakistan Panama Peru Philippines Poland Portugal Rep. of Korea Romania Russian Fed. Saudi Arabia Serbia Singapore Slovakia Slovenia South Africa Spain Sri Lanka Sweden Switzerland Syrian Arab Rep. Thailand Trinidad & Tobago Tunisia Turkey UAE UK Ukraine Uruguay USA Uzbekistan Venezuela Vietnam Zimbabwe

Z-score

ZPR

ZLR

Obs.

143.795 38.023 60.497 62.413 87.396 90.386 34.880 48.880 25.136 33.730 128.521 32.949 60.455 38.948 40.902 33.943 34.693 50.206 56.974 76.747 34.823 53.591 45.456 21.675 29.642 60.281 120.232 79.307 49.514 56.299 35.193 61.005 50.341 73.012 33.192 62.701 56.728 179.782 52.834 38.665 91.956 52.836 29.293 10.491 63.836 45.434 24.987 41.702 4.957

14.495 2.952 9.942 8.844 13.289 17.659 4.922 4.315 3.217 7.056 27.916 3.283 11.638 11.723 6.097 3.667 6.123 6.360 12.456 6.839 4.510 5.472 6.983 1.902 3.473 8.510 5.334 7.729 7.360 6.670 6.615 8.531 9.225 8.561 4.293 9.890 6.196 38.105 5.030 5.661 11.918 6.444 2.504 0.984 9.593 4.079 5.934 7.042 2.332

129.301 35.070 50.555 53.570 74.108 72.727 29.958 44.565 21.919 26.674 100.605 29.666 48.817 27.225 34.805 30.276 28.571 43.846 44.518 69.908 30.313 48.119 38.472 19.773 26.169 51.771 114.897 71.578 42.153 49.629 28.578 52.475 41.116 64.451 28.900 52.811 50.532 141.676 47.803 33.005 80.038 46.391 26.789 9.507 54.243 41.354 19.053 34.660 2.625

131 33 4 227 33 6 12 66 2 5 6 29 51 2 56 65 63 64 20 125 48 66 73 30 222 90 3 26 20 53 40 102 52 43 22 6 127 10 7 109 114 123 106 11 2973 10 93 19 1

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017

M. Doumpos et al. / Journal of Banking & Finance xxx (2015) xxx–xxx

References Agoraki, M.-E.K., Delis, M.D., Pasiouras, F., 2011. Regulations, competition and bank risk-taking in transition countries. Journal of Financial Stability 7, 38–48. Altunbas, Y., Cambacorta, L., Marques-Ibanez, D., 2014. Does monetary policy affect bank risk? International Journal of Central Banking 10, 95–135. Anginer, D., Demirgüç-Kunt, A., Zhu, M., 2014. How does deposit insurance affect bank risk? Evidence from the recent crisis. Journal of Banking and Finance 48, 312–321. Arnone, M., Laurens, B.J., Segalotto, J.-F., 2006. The measurement of central bank autonomy: survey of models, indicators and empirical evidence. IMF Working Paper, WP/06/227. Barry, T.A., Lepetit, L., Tarazi, A., 2011. Ownership structure and risk in publicly held and privately owned banks. Journal of Banking and Finance 35, 1327–1340. Barth, J.R., Dopico, L.G., Nolle, D.E., Wilcox, J.A., 2002. Bank safety and soundness and the structure of bank supervision: a cross-country analysis. International Review of Finance 3, 163–188. Barth, J.R., Nolle, D.E., Phumiwasana, T., Yago, G., 2003. A cross-country analysis of the bank supervisory framework and bank performance. Financial Markets, Institutions & Instruments 12, 67–120. Barth, J.R., Caprio Jr., G., Levine, R., 2004. Bank regulation and supervision: what works best? Journal of Financial Intermediation 13, 205–248. Barth, J.R., Lin, C., Ma, Y., Seade, J., Song, F.M., 2013a. Do bank regulation, supervision and monitoring enhance or impede bank efficiency? Journal of Banking and Finance 37, 2879–2892. Barth, J.R., Caprio Jr., G., Levine, R., 2013. Bank Regulation and Supervision in 180 Countries from 1999 to 2011. Mimeo, January. Baselga-Pascual, L., Trujillo-Ponce, A., Cardone-Riportella, C., 2013. Factors Influencing Bank Risk in Europe: Evidence from the Financial Crisis. Fundación de las Cajas de Ahorros (FUNCAS) (N° 722/2013). Beck, T., Gros, D., 2012. Monetary Policy and Banking Supervision: Coordination Instead of Separation. CEPS Policy Brief, No. 286, 12 December. Beck, T., De Jonghe, O., Schepens, G., 2013. Bank competition and stability: crosscountry heterogeneity. Journal of Financial Intermediation 22, 218–244. Berger, W., Kißmer, F., 2013. Central bank independence and financial stability: a tale of perfect harmony? European Journal of Political Economy 31, 109–118. Berger, A.N., Mester, L.J., 1997. Inside the black box: what explains differences in the efficiencies of financial institutions? Journal of Banking & Finance 21, 895–947. Blinder, A.S., 2010. How central should the central bank be? Journal of Economic Literature 48, 123–133. Bodea, C., Hicks, R., 2012. Price stability and central bank independence: discipline, credibility and democratic institutions. Mimeo, November 6, Available at: . Borio, C., 2013. On time, stocks and flows: understanding the global macroeconomic challenges. National Institute Economic Review 225, R3–R13. Boyd, J.H., De Nicoló, G., Smith, B.D., 2004. Crises in competitive versus monopolistic banking systems. Journal of Money, Credit, and Banking 36, 487–506. Boyd, J.H., Gomis-Porqueras, P., Kwak, S., Smith, B.D., 2014. A user’s guide to banking crises. Annals of Economics and Finance 15, 800–892. Brambor, T., Clark, W.R., Golder, M., 2006. Understanding interaction models: improving empirical analyses. Political Analysis 14, 63–82. Braumoeller, B.F., 2004. Hypothesis testing and multiplicative interaction terms. International Organization 58, 807–820. Brissimis, S., Delis, M., Iosifidi, M., 2014. Bank market power and monetary policy transmission. International Journal of Central Banking 10, 173–213. Carbó-Valverde, S., Kane, E.J., Rodriguez-Fernandez, F., 2013. Safety-net benefits conferred on difficult-to-fail-and-unwind banks in the US and the EU before and during the great recession. Journal of Banking and Finance 37, 1845–1859. Cheibub, J.A., Gandhi, J., Vreeland, J.R., 2010. Democracy and dictatorship revisited. Public Choice 143, 67–101. Cihák, M., 2010. Price stability, financial stability, and central bank independence. In: 38th Conference of the Oesterreichische Nationalbank. Available at: . Cukierman, A., 1992. Central Bank Strategy, Credibility, and Independence: Theory and Evidence. MIT Press, Cambridge, MA. Cukierman, A., 2008. Central bank independence and monetary policymaking institutions – past, present and future. European Journal of Political Economy 24, 722–736. Cukierman, A., Webb, S.B., Neyapti, B., 1992. Measuring the independence of central banks and its effect on policy outcomes. World Bank Economic Review 6, 353– 398. Dalla Pellegrina, L., Masciandaro, D., Pansini, R.V., 2013. The central banker as prudential supervisor: does independence matter? Journal of Financial Stability 9, 415–427.

15

Delis, M.D., Kouretas, G.P., 2011. Interest rates and bank risk-taking. Journal of Banking and Finance 35, 840–855. Demaestri, E., Guerrero, F., 2005. Financial supervision: integrated or specialized? The case of Latin America and the Caribbean. Financial Markets, Institutions & Instruments 14, 43–106. Demirgüç-Kunt, A., Detragiache, E., 1998. The determinants of banking crises in developing and developed countries. IMF Staff Papers 45, 81–109. Demirgüç-Kunt, A., Detragiache, E., Tressel, T., 2008. Banking on the principles: compliance with Basel core principles and bank soundness. Journal of Financial Intermediation 17, 511–542. Dincer, N.N., Eichengreen, B., 2012. The architecture and governance of financial supervision: sources and implications. International Finance 15, 309–325. Fisher, I., 1933. The debt-deflation theory of great depressions. Econometrica 1, 337–357. Gaganis, C., Pasiouras, F., 2013. Financial supervision regimes and bank efficiency: international evidence. Journal of Banking and Finance 37, 5463–5475. Goodhart, C., Schoenmaker, D., 1995. Should the functions of monetary policy and banking supervision be separated? Oxford Economic Papers 47, 539–560. Hakenes, H., Schnabel, I., 2011. Bank size and risk-taking under Basel II. Journal of Banking & Finance 35, 1436–1449. Herring, R.J., Carmassi, J., 2008. The structure of cross-sector financial supervision. Financial Markets, Institutions & Instruments 17, 51–76. Hutchison, M., McDill, K., 1999. Are all banking crises alike? The Japanese experience in international comparison. Journal of the Japanese and International Economies 13, 155–180. Jeon, B.N., Wu, J., Chen, M., Wang, R., 2014. Corruption and bank risk-taking: evidence from emerging economies. Mimeo, May 5. Available at: doi: 10.2139/ ssrn.2433068. Kashyap, A.K., Stein, J.C., 2000. What do a million observations on banks say about the transmission of monetary policy? American Economic Review 90, 407–428. Kayo, E.K., Kimura, H., 2011. Hierarchical determinants of capital structure. Journal of Banking and Finance 35, 358–371. Klomp, J., de Haan, J., 2009. Central bank independence and financial instability. Journal of Financial Stability 5, 321–338. Laeven, L., Levine, R., 2009. Bank governance, regulation and risk taking. Journal of Financial Economics 93, 259–275. Laeven, L., Valencia, F., 2012. Systemic banking crises database: an update. IMF Working Paper 12/163. Lepetit, L., Nys, E., Rous, P., Tarazi, A., 2008. Bank income structure and risk: an empirical analysis of European banks. Journal of Banking and Finance 32, 1452– 1467. Masciandaro, D., 2009. Politicians and financial supervision unification outside the central bank: why do they do it? Journal of Financial Stability 5, 124–146. Masciandaro, D., Quintyn, M., 2009a. Regulating the regulators: the changing face of financial supervision architectures before and after the crisis. European Company Law 6, 187–196. Masciandaro, D., Quintyn, M., 2009. Reforming financial supervision and the role of central banks: a review of global trends, causes and effects (1998–2008). Centre for Economic Policy Research, Policy Insight 30, February. Masciandaro, D., Pansini, R.V., Quintyn, M., 2013. The economic crisis: did supervision architecture and governance matter? Journal of Financial Stability 9, 578–596. Maudos, J., Pastor, J.M., Perez, F., Quesada, J., 2002. Cost and profit efficiency in European banks. Journal of International Financial Markets, Institutions and Money 12, 33–58. Melecky, M., Podpiera, A.M., 2013. Institutional structures of financial sector supervision, their drivers and historical benchmarks. Journal of Financial Stability 9, 428–444. Mishkin, F.S., 1996. Understanding financial crises: a developing country’s perspective. NBER Working Paper 5600. Mobarek, A., Kalonov, A., 2014. Comparative performance analysis between conventional and Islamic banks: empirical evidence from OIC countries. Applied Economics 46, 253–270. Montes, G.C., Peixoto, G.B.T., 2014. Risk-taking channel, bank lending channel and the ‘‘paradox of credibility’’ evidence from Brazil. Economic Modelling 39, 82– 94. Quintyn, M., Taylor, M.W. 2002. Regulatory and supervisory independence and financial stability. IMF Working Paper 02/46. Schaeck, K., Cihák, M., Maechler, A., Stolz, S., 2012. Who disciplines bank managers? Review of Finance 16, 197–243. Uhde, A., Heimeshoff, U., 2009. Consolidation in banking and financial stability in Europe: empirical evidence. Journal of Banking and Finance 33, 1299–1311. Zaheer, S., Ongena, S., van Wijnbergen, S.J.G., 2013. The transmission of monetary policy through conventional and Islamic banks. International Journal of Central Banking 9, 175–224.

Please cite this article in press as: Doumpos, M., et al. Central bank independence, financial supervision structure and bank soundness: An empirical analysis around the crisis. J. Bank Finance (2015), http://dx.doi.org/10.1016/j.jbankfin.2015.04.017