Bank capital buffers around the world: Cyclical patterns and the effect of market power

We examine the effect of competition and business cycles on bank capital buffers around the world. We use a dataset of 3,461 banks from 25 developed and 54 developing countries over the 2001-2013 period. Banks tend on average to exhibit pro-cyclical behavior. But capital buffers seem to be more pro-cyclical in developing countries. Our results show that more competition leads to higher buffers in developed countries but to lower buffers in developing ones. This evidence suggests that the “competition-stability” thesis adheres in developed economies, whereas “competition-fragility” makes more sense in developing countries. This asymmetric result may have important policy implications, particularly with regard to new, globally-negotiated capital adequacy standards.


Introduction
The term 'capital buffers' refers to any of a bank's capital holdings that are in excess of the regulatory minimum.
In this study, we measure capital buffers by finding the difference between observed regulatory capital-to-asset ratios and minimum regulatory ratios 1 . Positive capital buffers are persistent both across countries and over time, and they well exceed regulatory capital standards (Barth et al., 2006). According to the initial buffer calculations we use in this paper, average capital buffers are 9.20% for developed economies and 11.22% for developing economies (Table 1) 2 . This is notable as, according to corporate finance theory, issuing equity is costly, in part due to information asymmetries and agency problems faced by firms (Myers, 1984;Jensen and Meckling, 1976). As long as the costs of obtaining capital remain sufficiently high with respect to the cost of deposits, we would expect leverage to increase consistently with regulation. However, banks on the whole maintain substantial regulatory capital buffer holdings.
There are bank-and country-level factors that come to bear on a bank's decision to have capital buffers. Some examples of bank-level variables include the bank's business model, its relative size, and its profitability and risk levels. Country-level factors, meanwhile, include systemic levels of funding liquidity, banking regulations, the degree of competition within the banking industry, and the economy-wide business cycle. In this paper we control for these bank-and country-level variables when analyzing capital buffer determinants, and focus our analysis on the role played by market power and the business cycle.
Market power may affect buffers through multiple channels. Large banks with market power may become more risk-averse when they see their charter value threatened by competition. Rates charged by banks with market power can cause adverse selection effects across borrowers, increasing risk and prompting banks to increase buffers. Credit rationing by large banks can determine risk levels in their pool of borrowers, implying different buffer profiles. Given this theoretical line of reasoning, we are interested in further exploring the role of market competition in stimulating buffer creation. There are documented global differences (across regions and income groups) in levels of financial development and bank competition (Čihák et al., 2012); an interesting but rarely touched-upon question concerns the extent to which regional differences in competition translate to differential patterns in capital buffers (after taking into account country-level heterogeneity in the business cycle).
The business cycle may also affect buffers. When banks fail to accumulate capital buffers during economic booms, they can find themselves with insufficient levels of capital during economic downturns, meaning the cost of raising equity capital may force them to deleverage assets in order to meet regulatory minimum capital requirements (Repullo and Suarez, 2013; Borio and Zhu, 2012) 3 . 1 See Table B, Appendix B for a detailed description of variables and sources. 2 We also separate the also into four groups: OECD non-emerging, OHI (other high income, non-OECD), emerging, and other developing, following the classification in Claessens and Van Horen, 2015. Initial regression results indicate that the variables have a similar effect on the OECD (non-emerging) and OHI (non-OECD) groups, and the emerging and developing groups. "OECD" only includes the core OECD countries and "OHI" includes all countries classified as high-income by the World Bank in 2000 that do not are not OECD members. 3 We refer to pro-cyclical behavior, throughout the paper, as behavior that tends to reinforce the current cyclical phase. We refer to procyclical co-movement or fluctuations as the correlation of a particular variable with the cycle. Thus, a negative (positive) coefficient of the Before addressing these questions, though, it may make sense to sketch the recent history of macro-prudential regulation. After the date for implementation of the Second Accord was agreed by 2008, numerous countries were expected to gradually adapt Basel II. But the eruption of the global financial crisis postponed the adoption of the Second Accord in most of the world. Basel III calls for an overhaul of macro-prudential regulation, particularly for issues related to financial pro-cyclicality. Because of both the timing of the implementation and drastic regulatory changes after the crisis, the current state of banking regulation in developing markets can be characterized as non-convergentcountries have not fully implemented either Basel II or Basel III. Nonconvergent regulation is thus another possible source of country-level heterogeneity.
To prevent the destabilizing cyclical impacts of capital buffer fluctuations, Basel III requires banks to increase capital buffers during economic booms, through a "mandatory capital conservation buffer" of 2.5%, as well as a "discretionary counter-cyclical buffer" of up to 2.5% more in times of credit booms. This counter-cyclical capital buffer tool is a central element of Basel III. However, since these proposals were calibrated mostly against BIS membership information, crucial questions remain unanswered: Are buffers pro-cyclical globally?
What is the role of bank competition in explaining buffers? How do these factors interact across regions? Figure 1 shows average capital buffers in both developed and developing countries from 2001 to 2013. It also shows average real GDP growth rates for the two groups. During the period of economic expansion prior to the global financial crisis, average capital buffers in both regions seem to have followed the same pro-cyclical patterns. Capital buffers depleted progressively as the global economy expanded. However, after the crisis erupted, the buffer levels in the two regions started to diverge. In developed economies, the recovery was accompanied by increasing buffer levels, first because of rapid de-leveraging, and later because of increased regulatory pressure. Developing economies had a different experience. The post-crisis era was not accompanied by strong buffer accumulation. With economic deceleration beginning in 2010, these countries have seen both a rather rapid depletion of bank buffers and an increase in leverage. It seems that buffers became more countercyclical after the crisis, but the implications of this shift are various. Given the divergent experiences of developed and developing nations, future regulatory policy approaches in both regions should be closely informed by the existence of differences between countries and cyclical patterns, as well as the evolution of leverage and competition levels. Other factors to take into account include idiosyncratic characteristics specific to different banking systems. <Insert here Figure 1> At present there does not exist a wide literature on the cyclical patterns of buffers and effects of competition on a global scale. It is true that the heterogeneity of financial systems across the world has been well documented. business cycle in the buffer regressions is an indicator of the counter-cyclical (pro-cyclical) fluctuation of the capital buffer over the business cycle. This implies pro-cyclical (counter-cyclical) behavior.
Significant differences persist in levels of financial development, and bank competition in particular, across regions and income groups (Čihák at al., 2012). In light of this heterogeneity, and the potential channels through which competition can affect bank capital decisions, a worthwhile research question might be asked regarding the extent to which these regional differences in competition translate into differential patterns in capital buffers. Country and regional evidence is mixed. In this paper, we use Boone indicators of market competition, searching for data on the effect of banking competition on buffer formation. We employ, for our baseline specifications, a database containing 10,632 bank-year observations from 54 developing and 25 developed countries during the period 2001 to 2013. Prompted by previous literature, we also explore regional differences in the cyclical patterns of buffers, taking into account both developed and developing economies.
The empirical evidence presented in this paper is used to systematically explore the marginal effect of market power and the business cycle on buffer levels. Although we find that, taking the world as a whole, the average bank exhibits pro-cyclical behavior, implying that implementation of Basel III's counter-cyclical buffer tools has been extensive and effective, we also find cross-national differences connected to each country's respective levels of development and of bank competition.
This paper contributes to the existing literature by examining an expanded range of banks and countries, including multiple dimensions, and incorporating various methodological improvements and robustness checks.
There are, to the best of our knowledge, only two other previous studies that examine the determinants of global capital buffers: Fonseca and Gonzalez (2010) and Chen et al. (2014). Our empirical exercise is close in nature to that of Fonseca and Gonzalez (2010), as we include both bank-and country-level determinants in our analysis. In particular, our results indicate that competitive conditions in the banking system may play an acutely important role in the determination of capital buffer levels. This effect differs, however, for developed and developing economies.
The paper is ordered as follows. In Section 2, we present a review of the related literature with regard to both theory and evidence. In Section 3, we present the econometric approach used across the empirical work in the field. The data and empirical results are presented in Section 4, after which we conclude with some policyrelated comments.

Related Literature and Empirical Evidence
A plethora of recent articles scrutinize the cyclical behavior, at both a national and a regional level, of bank capital buffers. However, the evidence in these studies is inconclusive as to whether buffers are pro-cyclical for either segments of the banking sector or the sector as a whole. For advanced economies, some studies find general evidence of pro-cyclical behavior (Ayuso et al., 2004;Lindquist, 2004 savings banks. This paper also detects a strong competition effect: the elasticity of a bank's buffer capital with respect to an increase in competitors' buffers is equal to one; higher competition is thus found to incentivize an increase in buffers. Stolz and Wedow (2011) examine German commercial and saving banks in the period 1993-2003, also finding a negative correlation between buffers and the economic cycle. In addition, they observe that savings banks correlate more strongly with the cycle. Bikker and Metzemakers (2004) use a large worldwide dataset to search for global patterns in bank capital behavior. They find a significant, negative correlation between capital ratios and the cycle, but these effects are small in magnitude.
Cyclical patterns are far from homogenous, however, even within advanced economies. Jokipii and Milne (2008) find evidence of pro-cyclical capital buffer behavior for commercial and savings banks and large banks in Europe. The capital buffers of cooperative and smaller banks are found to change in line with the cycle. They also find evidence of counter-cyclical behavior for banks located in recently admitted countries to the EU. In Among more comprehensive, cross-national studies, Saadaoui (2014) analyzes the relationship between capital buffers, the business cycle, and the default risk of 740 commercial banks in 50 emerging countries in the period 1997 to 2008. Bank capital buffers and loan default risk are found to be negatively correlated with the business cycle, supporting the implementation of counter-cyclical buffers in these countries. Yet, notably, higher market power is found to be a critical mediating factor, as it attenuates this pro-cyclical behavior. Thus, for emerging economies, higher market poweras measured by bank Lerner indexesis found to reduce the pro-cyclical behavior of banks. Returning to the only two papers that examine the global determinants of bank capital buffers for both advanced and developing regions: Fonseca and Gonzalez (2010) look at global determinants of capital buffers for a sample of 1,337 banks from 70 countries, in the period 1995 to 2002. Their model uses a partial adjustment framework and pays special attention to the effect of market power and the role of regulation and institutional frameworks. With respect to bank-specific variables, the study includes factors such as the cost of funds, profitability and size. Buffers are not found to be systematically related to the business cycle, but they are found to be positively related to market power and the cost of deposits. However, these effects are sensitive to variations in regulation between countries, especially accounting disclosure rules and the stringency of regulations on the scope of banking activities. We find the paper's use of country-level controls and bankspecific factors to be critically important, as it illustrates differentiated patterns in the levels of capital buffer holdings across and within developed and developing countries. One failing of the paper derives from the weak robustness of the results 4  The relationship between competition and bank risk attitudes demands complex and multifaceted analysis (Allen and Gale, 2004). The relationship between competition and financial stability is likewise a controversial question, and one which remains unresolved despite a large body of work having been produced on the issue.
According to the so-called "competition-fragility" view, competition increases risk-taking, undermining the "charter value" of banks; it would follow that banks with market power have less incentive to engage in risky activities (Keeley, 1990). Another version of this view states that larger banks diversify better and are more efficient due to economies of scale that allow them to spread out their operations geographically (Boyd and Prescott, 1986;Méon and Weill, 2005). This lack of consensus has manifested itself in recent contributions to the literature. The "charter value hypothesis" has also been rationalized as large banks enjoy increased profit buffers (Allen and Gale, 2004; Matutes and Vives, 2000; Cordella and Yeyati, 2003). Larger banks engaging in "credit rationing" enhance soundness as they pick fewer but higher quality investments (Boot and Thakor, 2000).
Under the "competition-stability" view, the "too big to fail" doctrine posits that implicit public guarantees led to increased risk-taking by the largest banks (Mishkin, 1999). Boyd and De Nicoló (2005) argue that rates charged by banks with market power induce increased risk-taking by entrepreneurs. Increased scope and scale economies can be counterbalanced by increased managerial inefficiency ("X-inefficiency") due to increased complexity (Beck et al., 2006). While some empirical work supports the "competition-fragility" view (Beck et al., 2006), other studies provide evidence supporting a "competition-stability" hypothesis (De Nicoló et  This debate bears on the relationship between banking competition and capital ratios. Schaeck and Cihák (2012) use data from ten European countries to test the hypothesis that competition creates incentives for greater capital retention. Their paper arose partly out of Allen et al. (2011), who analyze the credit side of banks, and characterize capital holdings as a means of signaling commitment to monitoring in the presence of market discipline. According to their model, when banks compete for good lending opportunities, borrowers may demand that banks use their own capital for loans, thus incentivizing proper monitoring. As some borrowers do not fully internalize the cost to banks of raising capital, the resulting capital level could be above the regulatory requirement. Indeed, Schaeck and Cihák (2012) find that market competition increases bank capital holdings.
Moreover, market power may be associated with lower capital holdings if large banks enjoy lower adjustment and capital costs, and larger margins and profitability, with increasing bank revenues reducing the role of capital as a buffer to absorb future losses (Elizalde and Repullo, 2007) 5 . As noted above, Lindquist (2004) also finds that in Norway bank competition increases buffer holdings. This suggests that banks probably use excess capital to signal solvency and stability. It may also be the case that banks with market power are more vulnerable to systemic risk during large downturns, as the experience of the recent global financial crisis has shown (Demsetz and Strahan, 1997).
To summarize, our assessment of the status of knowledge regarding the determinants of capital buffers lead us to assume that regional and country effects are at play with regard to both the cycle and market competition.
This calls for an empirical approach able to systemically capture this complex interplay. With this in mind, we use a partial adjustment framework to explore: 1. The cyclical patterns of capital buffers, both globally and in developed and developing economies.

The influence of levels of competition in national markets in determining the level of capital buffers.
We presume that, given well-documented differences across regions in both dimensions, observed capital buffers should also reflect differential patterns.

A Partial Adjustment Framework
Given the evidence that capital buffer levels are persistent (see Figure 2), we use a partial adjustment framework to model banks' decisions regarding the optimum target for capital buffers. In line with previous literature (Ayuso et. al., 2004;Jokipii and Milne, 2008;Shim, 2012;Carvallo et al., 2015), the partial adjustment model can be represented as: where , and , * denote the actual and target capital buffer of bank at time , respectively. Coefficient measures the speed of adjustment of the actual capital buffer towards its target level. Factors are bankspecific and permanently affect banks' capital structure decisions. , is independently distributed with zero mean.
<Insert here Figure 2> Adding the lag of , to both sides of Eq. (6), we obtain: The optimum level of capital buffer ( ). In addition, in the empirical analysis we also control for the level of competition, the regulatory environment, and the country's financial depth as potential determinants of the optimal capital buffer level. Hence the empirical model is specified as follows: Where , is a vector with variables that we assume influence bank 's optimal buffer at time , with 1 = 1 − first differences. However, Arellano and Bover (1995) and Blundell and Bond (1998) suggest differencing the instruments instead of the regressors in order to make them exogenous to the fixed effects. This leads to a shift from the difference GMM to the system GMM estimator, which is a joint estimation of the equation in levels and in first differences. We therefore use two-step system GMM estimators, with the Windmeijer (2005) corrected standard error.

Data
This study combines data from an unbalanced annual bank panel with a set of country regulatory and macroeconomic variables for 3,461 banks from 79 countries for the period 2001-2013. First, we use Bankscope 6 data for all bank-specific variables. Second, we consider a set of macroeconomic and regulatory variables publicly available on the World Bank database as the "The Regulation of Banks around the World": surveys I, II, III and IV. After merging these data sets, the initial sample of banks in which we are able to calculate the dependent variable (i.e., capital buffer) is 7,090 7 . However, after estimating our baseline specifications (Table   2, below), we obtain subsamples ranging from 10,632 (3,461 banks) to 6,538 observations (2,193 banks), depending on specifications. Nonetheless, the resulting sample represents a wide set of countries and a relevant amount of observations and banks from developed and developing countries (see Table C This method helps preserve the sample size in panels with gaps. A detailed description of the variables considered in the regressions is displayed in Table B, Appendix B 8 . Table 1 reports descriptive statistics that aggregate countries into a developed and a developing group. Following Claessens and Van Horen (2013, 2015), we classified "developed" countries as OECD-non-emerging plus Other High Income (OHI) (labeled as OECD), while the rest of the countries are classified as "developing" (labeled as DEV + EME). Using this income aggregation, our sample shows buffers, profitability and risk to be larger in developing economies, whereas developed economies present greater financial depth. The countries 6 BankScope is a database for bank-sheet level data based on financial statement information provided by Bureau van Dijk. BankScope has been extensively used in the banking literature and provides harmonized data that permit cross-country comparisons. 7 However, in the specific case of the variable SizeCoa dummy that takes 1 if the bank size belongs to the top quartile in their national size distributionwe use all the information of the initial annual unbalanced bank panel (24,087 banks). We were able to calculate buffers for 7,090 banks from 143 countries during the period between 2001 and 2013 (40,096 observations). 8  Because of the large number of US banks in our sample (15% of the total), and given the complexity of this market, we ran our regressions first using all the banks in the sample, and then excluding US banks (last five columns in Tables 2-6). In Table 2 In Table 2, the coefficient on the lagged buffer (L.BUF) is highly significant in all specifications, revealing the presence of significant adjustment costs for banks to change their capital buffers to their target amounts.  (2007), which states that larger banks with larger margins, lower adjustment costs, and higher profitability could exhibit lower capital holdings.
A key variable in our analysis is the Boone indicator (Boone 2001(Boone , 2008, which is calculated as the elasticity of profits to marginal costs. Its rationale is that higher profits can be achieved by more efficient banks. Appendix A has more details about this indicator. The more negative the Boone indicator, the higher the degree of competition, because the effect of reallocation is stronger. Throughout all of our results this market power indicator is significant and positive, indicating that in banking markets where more competitive conditions prevail, banks tend to have lower capital buffers ( Table 2). This result is reminiscent of theories based on the "charter value" of banks and the detrimental effect of competition with regard to bank risk attitudes (Keeley, 1990).

<Insert here Table 2>
However, as Table 3 shows, these general results are sensitive to the bank's location in either developed or developing regions, with a sign reversal in the former. More competition is now associated with higher capital buffers in developed countries and lower buffers in developing ones. The evidence suggests that the "competition-stability" view is correct in developed economies, whereas the "competition-fragility" view is Yeyati, 2003). Large banks in these countries can enjoy increased profit buffers or more easily engage in "credit rationing", thus enhancing stability as they choose fewer but higher quality investments (Boot and Thakor, 2000).

<Insert here Table 3>
The average global bank capital buffer is negatively correlated with the growth of economic activity (GGDP) in other words, it behaves pro-cyclically. This strongly supports Basel's III proposed counter-cyclical buffer tools and is also consistent with findings from several country-level and other more general studies ( 9 . The results for the sample excluding the US are generally the same as those from the global sample (Table 2).
To uncover cross-national differences in the effect of these factors on buffer levels, we first analyzed the variation in income or economic development across countries. We thus grouped countries according to income and classified them as either "developed" or "developing". We introduced a dummy equal to one for the developing country region and interacted it with the business cycle variable, the growth rate of real GDP. The average coefficient for developing countries is significantly negative, which indicates that on average the buffer levels for developing banks are pro-cyclical (Table 4). Notably, the average effect for banks in developed countries is counter-cyclical (positive sign), although mostly insignificant. These results indicate that procyclicality is generally more of a problem in developing countries. This is particularly worrying as capital buffers have been rapidly decreasing in developing countries during the recent economic slowdown (Figure 1).

<Insert here Table 4>
For robustness, we replicate the specifications in Table 3, but use alternative measures of market power. First we use country averages of banks' Lerner indexes (Table 5). Next we use the asset market share of the five largest banks, a measure of market concentration (Table 6) 10 . The effect of the average Lerner indexes on buffers is found to be statistically insignificant for seven out of the ten specifications. Moreover, the concentration measure is found to be statistically significant only for the regressions that exclude US banks, presenting a positive sign.
One of the factors that complicate this picture is the complexity inherent in the relationship between bank  11 . In light of these developments, more 9 A surprising result in our regressions is the sign of Official Supervisory Power (OSP). It is meant to capture "whether the supervisory authorities have the authority to take specific actions to prevent and correct problems". This variable generally has a negative sign, which seems counterintuitive, although its significance is not robust across specifications. It could indicate some form of substitution between private and public risk-control mechanisms. Fonseca and Gonzales (2010) get a similar result via its interaction with the cost of capital. 10 To the extent that Bankscope does not have 100% coverage of banking systems, and is limited to commercial banks, all these measures share potential selection biases equally. 11 Bikker and Haaf (2002) study the interrelation between market structure and competitiveness, using cross-country data. They provide support for the conventional view that concentration reduces competitiveness. Claessens and Laeven (2004), however, find that competition recent studies have focused on non-structural measures of competition, i.e., measures that do not rely on the link between structure and conduct to infer market power. Both the Boone and Lerner indicators fall within this category. However, one of the theoretical weaknesses of the country-wide Lerner indices is that they are averages of the market power of individual banks. Even if individual Lerner indices decrease with competition, the average degree of market power may increase, decrease or remain stable due to relocation effects from inefficient firms to efficient firms (Boone, 2008). Thus, within the context of this global study, we consider our Boone indicators to be superior measures of the effective level of competition in banking systems.
<Insert here Table 5> <Insert here Table 6> It might also be argued that, because of our particular sample's time window, which includes the global financial crisis, our results may capture reverse causality from the banking system to the economic cycle. We examine these "tranquil countries" (tranquil, that is, with respect to the global financial crisis). For both cases, we found that our results regarding the cycle held. We also looked at the cyclical behavior of different components of capital buffers, by running the previous set of regressions on three key buffer components: (1) the actual regulatory capital ratio itself; (2) the tier 1 capital ratio; and (3) the ratio of loan-to-total assets, which is a key determinant of risk-weighted assets. We found global banks behave pro-cyclically with respect to the regulatory capital ratio. They do so by increasing (decreasing) tier 1 capital ratios during periods of decreased (increased) economic activity and by reducing (increasing) lending ratios during periods of decreased (increased) economic activity. However, for most specifications, the magnitude of the adjustment effect seems to be larger for the loan ratio than for the tier 1 adjustment effect 12 .

Conclusions
In this paper, we look for the determinants of bank capital buffers at a global scale, using a large dataset of This would seem to underline a global need for extensive and effective implementation of Basel III's proposed counter-cyclical buffer tools. The average coefficient for developing countries is, however, consistently larger is not negatively related to market concentration. 12 Results available upon request.
than the average effect worldwide (Table 6). Of note is the fact that the average effect for banks in developed countries is mostly insignificant, with a positive sign. This result indicates that pro-cyclicality seems to be a more generalized problem in developing countries.
Another Since the global financial crisis there has been substantial deleveraging and capitalization levels have strengthened. However, higher pro-cyclicality in developing countries implies that capitalization levels are actually more compromised than previously thought. Transaction costs and market power are known to be greater in the financial systems of developing economies. Furthermore, in these countries, given the absence of correctly-functioning liquid capital markets and market discipline, market power remains critical for sustaining higher capital buffer levels. These differences should be considered while deploying regional efforts to revamp the regulatory landscape by creating more systemic and cycle-sensitive capital regulation. As an increasing number of countries prepare for full implementation of Basel III, the optimal sequencing and mix of instruments and priorities should therefore consider simultaneously the competitive and prudential aspects of particular countries. Competition-enhancing regulation in developing countries should go hand in hand with market reform aimed at reducing transaction costs and information asymmetries. This would improve financial market depth, liquidity and discipline. A comprehensive approach must also consider the stringency of capital regulations, along with policies aimed at improving competition and capital market reform. .48*** Two sample mean comparison t-test for unequal variances * p < 0.10, ** p < 0.05, *** p < 0.01. Sd denotes the standard deviation. We also separate the data into four groups: OECD non-emerging, OHI (other high income, non-OECD), emerging, and other developing, following a recent classification in Claessens and Van Horen, (2015). Initial regression results indicate that the effects of variables are similar in OECD (non-emerging) and OHI (non-OECD), and in emerging and developing. "OECD" only includes the core OECD countries and "OHI" includes all those countries classified as highincome by the World Bank in 2000 but that do not belong to OECD. SizeCo is a dummy variable that takes 1 if bank's size belongs to the top quartile in its national size distribution; ROAA is the return on assets; ROAE is the return on equity; LLRGL is the ratio of non-performing loans to total bank assets; BOONE is the competition index; Lerner is an index of market power; 5-conc represents the assets of the country's five largest banks as a share of total commercial banks; CF is the cost of funding; OCS is an index that measures overall capital stringency; OSP is an index that measures regulatory authority power; and BACC is a dummy for bank accounting. Notice these statistics are different from those in Table 1, since they are un-weighted averages of country variable averages.  Observations  10632  10556  6880  6538  6538  7275  7199  5316  4974  4974  Panels  3461  3440  2316  2193  2193  2930  2909  1885  1762  1762  j  84  85  84  85  87  84  85  84  85  500 Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01. The Hansen is a test of the over-identifying restrictions for the GMM estimator. AR1 and AR2 are tests for the first-order and secondorder serial correlation. N denotes number of observations, Ng number of banks and J number of instruments. Year denotes if year dummy variables were introduced in the regressions. BUF(-1) is the lag of bank's capital buffer; SizeCo is a dummy variable that takes 1 if bank's size belongs to the top quartile; ROAA is the return on assets; ROAE is the return on equity; LLRGL is the ratio of non-performing loans to total bank assets; BOONE is the competition index; CF is the cost of funding; OCS is an index that measures overall capital stringency; OSP is an index that measures regulatory authority power; BACC is a dummy for bank accounting; MCM is the ratio of money and quasi-money to GDP; and Cooperative and Savings are dummy variables that identify the type of bank.   916 Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01. The Hansen is a test of the over-identifying restrictions for the GMM estimator. AR1 and AR2 are tests for the first-order and secondorder serial correlation. N denotes number of observations, Ng number of banks and J number of instruments. Year denote if year dummy variables were introduced in the regressions. BUF(-1) is the lag of bank's capital buffer; SizeCo is a dummy variable that takes 1 if bank's size belongs to the top quartile; ROAA is the return on assets; ROAE is the return on equity; LLRGL is the ratio of non-performing loans to total bank assets; Lerner is the market power index; CF is the cost of funding; OCS is an index that measures overall capital stringency; OSP is an index that measures regulatory authority power; BACC is a dummy for bank accounting; MCM is the ratio of money and quasi-money to GDP; and Cooperative and Savings are dummy variables that identify the type of bank. in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01. The Hansen is a test of the over-identifying restrictions for the GMM estimator. AR1 and AR2 are tests for the first-order and secondorder serial correlation. N denotes number of observations, Ng number of banks and J number of instruments. Year denote if year dummy variables were introduced in the regressions. BUF(-1) is the lag of bank's capital buffer; SizeCo is a dummy variable that takes 1 if bank's size belongs to the top quartile; ROAA is the return on assets; ROAE is the return on equity; LLRGL is the ratio of non-performing loans to total bank assets; 5-conc is the market share of total assets held by the 5 largest banks; CF is the cost of funding; OCS is an index that measures overall capital stringency; OSP is an index that measures regulatory authority power; BACC is a dummy for bank accounting; MCM is the ratio of money and quasi-money to GDP; and Cooperative and Savings are dummy variables that identify the type of bank.  More competition raises the market share of a more efficient bank relative to a less efficient one. A larger negative value of  is an indication of more competitive conditions. However, positive values of  are also possible, implying either that the market has an extreme level of collusion or that banks are competing on quality.

The Lerner Index
The Lerner index is a bank-level alternative measure of market power, used for robustness purposes in our empirical exercise. It is positively correlated with market power. It is averaged at the country level. It can be defined as the difference between marginal price and marginal cost divided by marginal price, as follows: where it P is the average price of output, and it MC is the marginal cost of producing this output.

Variable
Description Source

Bank Accounting (BACC)
Dummy variable that takes a value of 1 when the income statement includes accrued or unpaid interest or principal on nonperforming loans and when banks are required to produce consolidated financial statements and zero otherwise.

Growth rate of Gross Domestic Product (GGDP)
Growth rate of Gross Domestic Product. Our measure of the business cycle.