A. The Sample

The dataset used in this study comprises bank-level data and country-level data; it is compiled from a number of sources: (a) the IMF and World Bank Basel Core Financial Sector Assessment Program (FSAP) database, which includes detailed assessment of a country’s compliance with the Basel Core Principles for Effective Bank Supervision (BCP) during 1999–2010; (b) the Barth and others (2004, 2006, and 2008) surveys on bank regulation, supervision, and monitoring; (c) the World Bank Economic Indicators database (WDI); and (d) the Bankscope database provided by Bureau van Dijk and Fitch Ratings.

Bank-level information comprises balance sheet and income statement data for all publicly quoted commercial banks and bank holding companies. We focus on publicly quoted banks, on the assumption that these institutions are subject to more stringent regulatory controls and need to comply with international regulations, such as capital regulation. This focus should also enhance cross-country comparability, not least because publicly quoted banks follow international accounting standards to report end-of-year accounting variables (Laeven and Levine 2008). When constructing the dataset, we exclude banks with missing information on relevant accounting variables (total assets, loans, other earning assets, deposits, equity capital, interest and non-interest income, and interest and non-interest expenses). To prevent the possibility of outliers driving the results, we exclude the top and bottom 1 percent levels, in terms of bank size (total assets), for all years.⁸ Finally, we apply data cloud methods to identify and remove outliers in terms of input/output mix⁹ (see Section 2.B for more detail on input/output specification).

We then match the bank-level information with country-level information to investigate the link between regulatory compliance and bank performance, accounting for cross-country differences in macroeconomic and institutional factors. Our final cross-sectional sample includes 863 banks across 63 countries over the period 2001–10 (Table A.1).¹⁰ Our sample is unbalanced and it includes countries with vastly different banking systems and economic conditions, with some countries only represented by a few listed banks, while others have a much higher sample share. Specifically, U.S. banks account for approximately 35 percent (304 banks) of the sample. To ensure that our findings are not overly influenced by U.S. banks, we examine results with and without them.

For the purpose of the analysis, we categorize the 63 countries in our sample both in terms of both economic development and geographic region, combining information from the International Monetary Fund (IMF) and the European Bank for Reconstruction and Development (EBRD). Countries are classified into four categories of economic development: (i) Major Advanced (countries in the G7 group); (ii) Advanced, (iii) Emerging and Developing; and (iv) Transitional. In addition, countries are also classified into 10 geographical regions (Central and Eastern Europe (EEU); Latin America and the Caribbean (LAM); Middle East and North Africa (MEA); Newly Independent States of Former Soviet Union (FSU); North America (NAM); Other Pacific Asia (PAS); Pacific OECD (PAO); South Asia (SAS); Sub-Saharan Africa (AFR); Western Europe (WEU)).

Because the country-level regulatory data (data source (b)) are collected in three survey exercises (1999, 2002, and 2005/2006), following Barth and others (2013) we match the data for the regulatory variables as follows: the 1999 survey data are used for period 2000–01; the 2002 survey data are used for period 2002–04, and 2005/2006 survey data are used for period 2005–10. As each bank is in the sample is assessed at one point in time during the sample period, the data is cross-sectional.

B. Empirical Set-Up and Definition of Variables

Frontier methodologies for the analysis of firm performance have generated a large literature since the seminal work of Leibnestein (1966) introduced the concept of x-inefficiency as the gap between ideal efficiency and actual efficiency. Frontier approaches measure firm performance relative to ‘best practice’ in the industry. Such measures summarize performance in a single statistic that controls for differences among firms using a sophisticated multidimensional framework (Banker & Cummins, 2010). The evaluation of efficiency is based on the assumption that the production frontier of the fully efficient organization is known. In practice, data is used to estimate this idealized frontier. Over the last half century, estimations of this best practice frontier developed along two empirical paths, a parametric and a non-parametric one. In this study, we follow a non-parametric approach which uses linear programming methods to assign each observation to its own set of ‘coefficients’ from which inefficient behavior can be assessed. More specifically, we employ the most well known of these ‘data-oriented’ methods, Data Envelopment Analysis (DEA), first introduced by (Charnes, Cooper, & Rhodes, 1978) and later extended by (Banker, Charnes, & Cooper, 1984). Our methodological approach represents an extension of the traditional DEA model.

Formally, DEA is a methodology directed to frontiers rather than central tendencies. It floats a piecewise linear surface to rest on top of the observations rather than fitting a regression plane through the ‘middle’ of the data using statistical methods (Cooper, Seiford, & Zhu, 2011). DEA produces exact in-sample estimates of efficiency; that is a measure of the performance of an institution relative to the other institutions which are producing the same good or service. This method is non-stochastic; it assumes that all deviations from the frontier are the result of inefficiency. This represent a drawback of the approach, as statistical inference about estimates comparisons are precluded without further simulation techniques such as bootstrapping. To overcome this problem, we follow a double bootstrapping procedure, as proposed by Simar and Wilson (2007).

Another of the key issues that arises in the use of frontier methods for cross-country comparisons of bank efficiency is the existence of significant heterogeneity. Several studies have proposed alternative methodologies to overcome this problem. In this paper, we begin with the estimation of a common global frontier by means of Data Envelopment Analysis (DEA). In the next step, to account for cross-country heterogeneity we adopt a form of the two-stage approach with a double bootstrapping procedure (Simar and Wilson 2007). In this two-stage approach, the first stage measures efficiency by a DEA estimator and the second stage uses truncated regression to infer how various (bank-specific and country-specific) factors influence the (bias-corrected) estimated efficiency. The choice of this methodological approach is driven by the characteristics of our sample, which includes a large number of countries (and a relatively small number of banks per country) and therefore presents considerable challenges to accounting for differences in operating, regulatory, and macroeconomic conditions experienced by banks.¹¹ Our study is the first cross-country study to apply the double bootstrapping two-stage procedure in a consistent manner.¹²

In more detail, our approach can be broken up into three steps. In the first step, we use a nonparametric input-oriented DEA model to measure bank efficiency.¹³ However, ignoring the complex and heterogeneous nature of the sample and simply benchmarking performance on the basis of a global common frontier would yield biased efficiency estimates. As a consequence, in the next stage, we apply the double bootstrapping procedure proposed by Simar and Wilson (2007) to explicitly account for the complex serial correlation in a two-stage DEA efficiency estimation. In the estimation of DEA technical efficiency scores, this procedure recognizes that certain bank-specific and environmental variables influence the estimate of the true unobserved efficiency score and is thus consistent with the second-stage truncated regression analysis (Glass and others, 2010)

In a second step, the results from a truncated regression of the initial DEA efficiency estimates on a set of environmental variables are used in a nonparametric bootstrap to generate biased corrected efficiency estimates. This step adjusts for the influence on the DEA efficiency estimations of the correlation between observable bank/country level factors and the inputs/outputs in a bank production process. Finally, in a third step, the bias-corrected DEA efficiency estimates from step two are used as the dependent variable in a further truncated regression on the same set of environmental variables, and a parametric bootstrap is used to provide more efficient estimates of the statistical relationships between the environmental variables and bank efficiency.

C. Estimating Bank Efficiency

We proceed to evaluate bank operating efficiency as follows. Let us define a production set as

where $x\in {\mathbb{R}}_{+}^p$ denotes a vector of p inputs and $y\in {\mathbb{R}}_{+}^q$ denotes a vector of q outputs. The technology or production frontier is defined as

which is then used to measure the i^th banks’ input technical efficiency, defined as

the proportion by which input quantities can be reduced without reducing output quantities (Coelli and others, 2005).¹⁴

When a large cross-country sample is used to build a best-practice frontier, inefficiency for bank i in country j is measured in terms of distance from this global common frontier. This implies that any cross-country differences in the initial DEA efficiency scores are entirely attributed to bank-level managerial decisions regarding the scale and mix of inputs. If this assumption is not correct, it will result in biased efficiency estimates. In the next two steps, unlike most cross-country bank efficiency studies, we apply a double bootstrapping procedure (Simar and Wilson 2007) to account explicitly for the complex serial correlation in a two-stage DEA efficiency study. This procedure will adjust for the bias in the first stage DEA estimates of bank efficiency. We will then use these bias-corrected efficiency scores to improve statistical efficiency in the second-stage truncated regression estimates.

Mathematically this bias can be described as follows:

where δ_i is the true (unobservable) efficiency score for i^th bank, ${\hat{\delta }}_{\iota }$ is the nonparametric DEA estimate of δ_i, BIAS$\left({\hat{\delta }}_{\iota }\right)$ is the bias of the nonparametric estimate which is strictly negative in finite samples, and μ_i is the error in the nonparametric estimate, which will disappear asymptotically.¹⁵ As can be seen from (4) the true unobserved efficiency scores are generally downward biased, and nonparametric efficiency estimates that ignore this bias will provide a more favourable estimate of efficiency. In the context of our study, the bank would appear to be performing better, in terms of the efficient allocation of its resources, than is actually the case. Our estimation will implicitly account for this bias. Full details of the process used are described in Algorithm 2 in Simar and Wilson (2007).

D. Definition of Inputs and Outputs

The inputs and outputs used to estimate efficiency are defined based on an extension of the intermediation approach (Sealey and Lindley 1977), which does not penalize nontraditional banking activity and takes into account a bank’s ability to manage risk. We estimate a model that has three inputs and three outputs.

The inputs are: (i) customer deposits and short-term funding; (ii) total costs (defined as the sum of interest expenses and noninterest expenses), and (iii) equity capital to adequately account for the impact of risk (Berger 2007). Hughes and Mester (2010) argue that the inappropriate treatment of equity capital can bias bank efficiency estimates because banks can use either capital or deposits to fund loans, and this choice has a direct effect on funding costs.¹⁶ As equity capital has a minimum level due to capital adequacy regulation, it should be treated as a quasi-fixed input; a variable whose control is not at the complete discretion of the management.¹⁷

The three outputs are: (i) loans; (ii) other earning assets; and (iii) noninterest income as a proxy for off-balance-sheet activities.¹⁸ We include the latter output to ensure that we do not penalize banks that have a substantial share of nontraditional activities (Barth and others, 2013)

E. Measuring Bank Compliance

The principal variable of interest, BCP compliance, is derived from the IMF and World Bank Basel Core Financial Sector Assessment Program (FSAP) database.¹⁹ Our study extends the work of Demirgüç-Kunt and Detragiache (2011) by using assessment data covering 1999–2010,²⁰ which includes a U.S. banking sector assessment. The Basel Core FSAP is an exhaustive global exercise, capturing the compliance features of banking industries in both developed and developing economies. The 25 BCP core principles are considered by regulators and by international organizations to be the best practice to date of compliance with banking regulation and supervision. These principles were issued in 1997 by the Basel Committee of Banking Supervision, and have been adopted by most countries in the world. Since 1999, the IMF and the World Bank have conducted regular assessments to gauge countries’ compliance with these principles, mainly within their joint FSAP. The 25 BCP core principles are organized into seven chapters, as follows:

- Chapter 1 (BCPch1): Preconditions for effective banking supervision. Six subprinciples constitute the prerequisite to perform supervisory activities including objectives, responsibilities, adequate resources, independence, legal infrastructure, and the existence of arrangements of cooperation between supervisors for sharing information and protecting its confidentiality.

- Chapter 2 (BCPch2): Licensing and structure. Four principles set the powers of supervisors in terms of their authority to grant banks licenses and to review major acquisitions and investments.

- Chapter 3 (BCPch3): Prudential Regulations and Requirements. Ten principles are in place to ensure that supervisors set prudent and appropriate minimum capital adequacy requirements for all banks. These requirements reflect banks’ risks, credit policies, loan provisioning, concentration, large exposures, risk mitigation policies, risk monitoring, audit, and code of conduct. These requirements impose a cost for banks, which have to put in place risk measurement, management, and governance systems to ensure compliance.

- Chapter 4 (BCPch4): Methods of Ongoing Supervision. Five principles impose regular supervisory visits to banks and contacts with bank management at group and subsidiary levels. In practice, compliance teams in banks must be kept ready to address supervisory matters when they arise.

- Chapter 5 (BCPch5): Information Requirements. One principle requires banks to maintain adequate records to enable supervisors to obtain a true and fair view on the financial conditions.

This principle implies that banks must keep comprehensive internal data on exposures (on- and off-balance sheets, clients, risks) and share it with supervisors.

- Chapter 6 (BCPch6): Formal Powers of Supervisors. One principle aims to ensure that adequate corrective supervisory measures are in place for distressed banks.

- Chapter 7 (BCPch7): Cross-Border Banking. Three principles to encourage supervisors to practice global consolidated supervision over internationally active banks. These principles may not work in practice because of the lack of exchange of confidential information between supervisors, and the difficulty of home-host issues, particularly in case of distress.

Following Demirgüç-Kunt and Detragiache (2011), the level of bank compliance is assessed using an aggregate BCP compliance score and a disaggregated approach, to distinguishing among various dimensions of regulation and supervision. The variable “BCP compliance” specifies a measure of compliance for each country in our sample at one point in time. Bank-level and country-level information are matched with the year of assessment to produce a cross-sectional sample. More specifically, to assess the compliance rate with each of the 25 principles, a four-point scale is used: (i) noncompliant; (ii) materially noncompliant; (iii) largely compliant, and (iv) compliant. Numerical values are assigned to each of the grades (from 0 for noncompliant to 3 for compliant). An overall index of compliance is computed based on the sum of the seven regulatory dimensions (BCP score). Seven indexes of compliance are then calculated based on the individual dimensions of regulation. All indices are normalized to take values in the interval [0, 1]. This normalization also has the intuitively appealing property of a percentage interpretation on initial analysis.

F. Environmental Variables

Bank efficiency is normally expressed as a function of both internal and external contextual variables. Internal factors are usually related to bank management and are defined from a bank’s financial statements and thus specific to its individual character. External factors describe the regulatory, macroeconomic, and financial development conditions that are likely to affect a bank’s performance. The contextual variables used in this study are chosen to best fit the primary purpose of the analysis and include both bank-specific and country-specific variables.

The bank-specific variables include log of total assets (logta), loans to assets ratio (lta), book value equity to assets (eqta) and return on equity (roe). Bank size, lending behaviour, capitalization, and risk profile are considered key determinants of bank performance.

It has long been established in applied bank efficiency studies that bank size (logta) can significantly affect bank performance. Banks enjoy economies of scale as they grow larger. One of the reasons put forward in the literature is that larger banks can better diversify risk (particularly credit and liquidity risk), which should reduce the relative costs of risk management. This in turn should allow banks to conserve equity capital, reserves, and liquid assets. Further, the spread of overhead costs (especially those associated with information technology) can also bring about larger efficiencies of scale in banking production (Hughes and Mester 2013). Although large banks can experience scale diseconomies and there might be costs associated with diversification, we expect an overall positive relationship between size and efficiency.

A bank’s production process is underpinned by its ability to improve information asymmetries between borrowers and lenders. This implies that a measure of relative lending behaviour such as loans to assets (lta) can be an important determinant of bank performance. Furthermore, as banks make choices about their capital structure and the amount of risk to assume, capitalization decisions have a direct impact on performance. We model this potential impact by including equity to assets (eqta). We expect higher levels of capital to be related to a reduction in a bank overall risk, and posit a positive relationship between the equity to assets ratio and bank performance. Finally, we also control for performance differences resulting from a manager’s ability to optimize the risk-return tradeoff, by including a bank’s return on equity (roe). Although no consistent picture emerges in the literature as to the relationship between risk and bank efficiency, a bank risk-taking profile is an important determinant of performance.

Moving to the external country-specific characteristics, these are a vector of the macroeconomic conditions and financial conditions in the banking industry of each country in the sample. The business and economic cycle fluctuations are modelled using annual growth in GDP (gdpg) and annual rate of inflation (inf) measured as the percentage change in the consumer price index. Favorable economic conditions will stimulate an improvement in the supply and demand for banking services, and will consequently have a positive effect on bank efficiency (Lozano-Vivas and Pasiouras 2010). Furthermore, high inflation can affect bank performance in a number of ways: it might encourage banks to compete through excessive branch networks, thus affecting cost (Kasman and Yildirim 2006), or it might have a beneficial effect on bank margins (Demirgüç-Kunt and others 2004).

The level of financial intermediation is controlled by including the ratio of private sector banks’ claims to GDP (PrCrGDP) as in Barth and others (2004). A higher ratio indicates increased loan activity, which is likely to improve bank efficiency. Higher efficiency resulting from high intermediation activity may be the effect of a bank’s risk preferences; recall that our model takes this into consideration by including equity capital as both a quasi-fixed input and in ratio form (eqta) as an environmental variable.

Lastly, we control for bank sector concentration using the ratio of the assets held by the three largest banks as a proportion of all bank assets of the country (conc). Higher concentration is thought to have a negative impact on bank efficiency because market power allows managers to relax their efforts to improve performance (Berger and Hannan 1998).²¹

The primary purpose of our study is to investigate whether BCP compliance plays a role in bank performance; therefore, it is vital to appropriately model the regulatory conditions within which each bank operates. We include six key features of banking regulation, which were first defined in Barth and others (2004, 2006). RESTR is a measure of the level of restriction placed on a bank’s activity. Barth and others (2004) argue that allowing banks to be involved in a range of activities may encourage the rise of larger, more complex entities that are more difficult to regulate. Reduced competition and efficiency may result, because banks may systemically fail to manage their diverse set of financial activities beyond the traditional model, lowering profitability (Barth and others, 2003).

COMP measures the level of regulation in place that would reduce competition (this includes entry requirements, limitations of foreign bank entry/ownership, and the fraction of new applications for banking licenses that are denied). As mentioned above, limited competition is likely to induce appropriating management behavior that may have a detrimental effect on bank efficiency. CAPRQ measures capital risk management restrictions. Pasiouras and others (2009) argue that capital requirements can affect bank efficiency and productivity in three ways. First, binding regulatory capital requirements reduce aggregate lending and affect loan quality, which in turn will affect efficiency. Second, stricter capital requirements influence a bank’s asset portfolio mix, resulting in different portfolio returns; this will affect input-oriented bank efficiency, because these returns will require the management of different resources. Finally, as mentioned earlier, differing capital standards will influence a bank’s decisions on the mix of deposits and equity, which entail different costs.

PRMON is a variable measuring the degree of private sector monitoring. This is a proxy for the third pillar of Basel II and can be related to the private monitoring hypothesis, which argues that supervisory power can incorporate business corruption and/or political motivation which, if increased, would affect bank lending integrity and compromise efficient credit allocation.²² Many economists support a greater reliance on private sector monitoring to promote better-functioning banks, although the quality of private monitoring largely depends on the quality of information disclosure. Although we expect the effect to be country-specific, we expect a positive link between the degree of private market monitoring and bank efficiency.

DEPSEC is a measure of the amount of security in place for depositors, in terms of deposit insurance schemes. Research suggests that increased levels of deposit insurance will exacerbate moral hazard issues and reduce the incentives for private monitoring. In terms of the effect on bank efficiency, higher levels of security for depositors would reduce banks’ incentives to efficiently allocate resources to the most productive opportunities, thereby resulting in a negative effect on efficiency.

CORPGOV is a measure of the level of effective corporate governance; it is derived from the External Governance Index. Better external governance is expected to enhance the private monitoring and disciplining of banks and thus boost banking efficiency.

Finally we control for differences in economic development of the countries in our sample. Countries are classified into four categories (Major Advanced, Advanced, Emerging and Developing, and Transitional) by development status. There may also be large regional differences, so countries are also defined into 10 regions (Central and Eastern Europe (EEU), Latin America and the Caribbean (LAM), Middle East and North Africa (MEA), Newly Independent States of the former Soviet Union (FSU), North America (NAM), Other Pacific Asia (PAS), Pacific OECD (PAO), South Asia (SAS), Sub-Saharan Africa (AFR), Western Europe (WEU)) and which are used to capture regional differences.

G. Methodology

The relationship between bank efficiency and regulatory compliance is evaluated using the following baseline specification:

where i indexes bank i, j indexes country j, EFF_i,j²³ is the efficiency score of bank i in country j, estimated by means of Data Envelopment Analysis and bias-corrected, as discussed in Section C. BCPIndex_j is the overall compliance index for country j, as discussed in Section E. The remaining environmental variables are included to capture observable cross-country and bank-characteristic differences. These have been discussed in Section F; Table A.3 presents the details of how variables were constructed.

More specifically, R_j is a vector of bank regulatory condition indicators (described in Barth, Caprio, and Levine (2006)) for country j; MF^j is a vector of macroeconomic and financial condition variables for country i; B_i,j is a vector of bank-specific characteristics for each bank i in country j, and Ij is a vector of dummy variables controlling for regional or economic development differences. The error terms and are assumed to be random noise elements of the dependent variable.

We estimate the model using a truncated regression in the double bootstrap procedure, as detailed in Section C.

In a second step, we decompose our main variable of interest, BCP, into the seven component chapters using the following disaggregated model:²⁴

where BCPch_j is an index of compliance, calculated based on the individual dimensions of regulation: BCPch₁ is Preconditions for Effective Banking Supervision; BCPch₂ is Licensing and Structure; BCPch₃ is Prudential Regulations and Requirements; BCPch₄ is Methods of OnGoing Supervision; BCPch₅ is Information Requirements; BCPch₆ is Formal Powers of Supervisors, and BCPch₇ is Cross-Border Banking. The remaining variables are defined as in equation (1).

H. Summary Statistics

Table 1 presents summary statistics of the full sample. Panel 1 describes the mean and standard deviation of each variable, while panel 2 provides an exposition of the median values categorized by economic development. Table 1b presents the correlation matrix of the BCP chapters. A few salient features emerge.

[Table 1: Descriptive Statistics]

Bank-level variables in panel 1 illustrate a host of differences between the 63 nations in our cross-country survey, indicative of variations in banking industry sophistication. The median values suggest that the sample is positively skewed, with a small number of large banks. Furthermore, there is a high degree of full-sample heterogeneity, with values varying widely about their means according to standard deviation figures.

From panel 1 the full-sample mean of the overall BCP compliance index (BCPscore) is 0.22, a much lower value than in the Demirgüç-Kunt and Detragiache (2011) study. This difference is likely owing to the inclusion of the U.S. banking sector, which dominates the sample and performed poorly in their 2010 BCP compliance assessment.

From panel 2 a number of interesting sample features emerge. First, overall compliance with the BCP appears to be higher in emerging and developing countries (45 percent), suggesting that these countries adhere more closely to BCP because their banking industries are nascent. A closer look at the regulatory control variables of emerging market and developing countries also suggests that these banking industries have many more restrictions placed on their activities (RESTR = 0.67), lack competition (COMP=0.67), and have no security in place for depositors (DEPSEC=0). All these characteristics would suggest a greater sensitivity to BCP compliance of banks in developing countries.

A. Efficiency Estimations

We begin our empirical investigation with the estimation of (bias-corrected) efficiency scores. Table 2 presents summary statistics for the bias-corrected DEA estimates. These include the coefficient of variation (CV, the standard deviation scaled by the mean).²⁵ This is a scale-free measure of dispersion that represents a comparative measure of efficiency volatility, with lower values indicative of more stable bank performance.

The results are presented in three panels, one for each of the groupings mentioned above. In panels 1 and 2 estimates are disaggregated by level of economic development, while in panel 3 the disaggregation is by geographical region.

Firstly, the overall mean bias-corrected input technical efficiency is 0.419 (the equivalent mean for the Non-U.S. bank sample and the emerging market and developing countries bank sample are 0.584 and 0.803 respectively). This mean value implies that a typical bank could improve its input efficiency by 58 percent; or, if the average bank were producing on the frontier rather than at its current location, only 42 percent of inputs being used would be required to produce the output set.

This global average is lower than in recent studies (Lozano-Vivas and Pasiouras 2010; Pasiouras 2008) that used DEA methods and similar samples. This difference is perhaps attributed to our explicit treatment of sample heterogeneity in the efficiency estimates.

As discussed above, nonparametric efficiency estimates using an unbalanced sample are inherently biased and will provide a more favorable picture of bank efficiency if this bias is ignored. A comparison of the overall mean raw efficiency²⁶ (0.537) and its bias-corrected counterpart (0.419) suggests that, on average, this estimated bias is 0.12. The comparative estimate of average bias in the non-U.S. and emerging market and developing country bank samples is 0.07 and 0.03 respectively.²⁷ The bias-corrected efficiency estimates thus reveal that the performance of banks is generally more inefficient than the raw, uncorrected DEA estimates suggest.

[Table 2: Bank efficiency estimates]

Second, there are some trends evident when moving from advanced to less advanced economies. In panel 1, mean efficiency scores exhibit little difference across major advanced and advanced country banks, while banks located in emerging market, developing, and transition countries are, on average, less efficient. Panel 2 results for the non-U.S. sample provide a much clearer picture of this trend. The most efficient banks are located in the major advanced countries (mean=0.642) while the least efficient banks are located in the emerging market and developing countries (mean=0.534).

Third, there appears to be an increase in the dispersion of bank efficiency estimates as we move from the most developed to the least developed countries. Panels 1 and 2 show some differences in the coefficient of variation across developmental levels, with major advanced country banks exhibiting the most stability in efficiency estimates (lower CV figure) while emerging market and developing country banks experience the most volatility in efficiency estimates (higher CV figure). This trend is most pronounced in the non-U.S. sample, where the coefficient of variation of the emerging and developing country banks (0.208) is nearly 30 percent higher than the corresponding figure for major advanced country banks (0.142).

The latter findings suggest that bank efficiency in emerging markets and developing countries is much more volatile. This increased volatility would suggest the necessity for tighter compliance with a set of effective banking supervision principles, and indicates the need for a more detailed analysis of these banks. Results from this analysis are summarized in panel 3: they suggest that a typical emerging market and developing country bank has a bias-corrected efficiency of 0.803²⁸ when benchmarked against best-practice peers of this subsample. This suggests that, on average, a bank producing on this emerging market and developing country bank frontier, instead of at its current location, would only need 80 percent of its inputs to produce the same amount of outputs. Overall there is little discernible difference across regions for banks located in emerging market and developing countries.²⁹

B. Truncated Regression Results

The main aim of this study is to provide consistent estimation of the relationship between bank efficiency and BCP compliance, given the heterogeneous nature of the sample. Using the approach described in Section II (G), we adopt two model specifications to provide a commentary on whether overall BCP compliance or compliance with any of its component chapters influence bank efficiency. Following Simar and Wilson (2007), we use a truncated regression model to investigate how producer-specific and country-level variables influence bank efficiency, with parameters being estimated by maximum likelihood. The authors’ Algorithm 2 is used to obtain the bootstrapped confidence intervals for these estimates. Specifically, the confidence intervals are constructed via the second part of the Simar and Wilson (2007) Algorithm 2 double bootstrapping procedure, using 30,000 replications. Tables 3a and 3b present the parameter estimates of the 12³⁰ regressions produced when the three groupings described above are used.

C. Regulatory Compliance in Emerging Markets

Because of the significant differences between emerging markets and advanced economies, in this part of the analysis we focus on a sample of emerging market and developing economies. Differences in the level of institutional development, law and order, contract enforcement, and corruption, for example, may affect both the level of compliance and the efficiency of banking institutions. We therefore examine whether the impact of regulatory compliance on bank efficiency is conditioned by the level of development. A higher level of compliance would provide a secure environment for stable industry growth, and thus improve banking efficiency; however, regulatory compliance with international standards is costly and could negatively affect bank resource allocation.

To this end, we re-estimate the (bias-corrected) efficiency scores on our subsample of banks from emerging market and developing economies, and then proceed with the two-stage analysis. The results are reported in Table 3b. Although the results on the aggregate BCP index do not support the hypothesis of an association between regulatory compliance and bank efficiency, when we explore the relationship between efficiency and compliance with the different chapters or group of principles, we find some evidence of a negative relation. In particular, for banks in emerging countries, compliance with Chapter 4 (Methods of Ongoing Supervision) has a negative and significant impact on the bias-corrected efficiency measures. Specifically, this chapter relates to the effectiveness of the existing supervisory framework and ability of supervisors to carry out their duties. Against a background of (potentially) increased supervisory scrutiny required to meet international standards, banks are likely to face more substantive compliance costs, such as investments in accounting systems, risk management systems, equipment, and training. This in turn can distort their business objectives, lowering investment and decreasing lending, and resulting in lower efficiency.

On the other hand, the extent to which supervisors apply international global standards is positively associated with bank input efficiency (Chapter 7). This latter result may indicate that in emerging markets, adherence to international standards of best practice has a positive effect on bank performance.

D. Sensitivity Analysis

For the purpose of robustness we re-estimate the models in Tables 4a and 4b using the Papke and Wooldridge (1996) fractional logit regression approach, as described in McDonald (2009). He argues that DEA efficiency is the outcome of a fractional logit process (taking values between zero and one) rather than a latent variable from a truncated process, as described by Simar and Wilson (2007). Using the raw uncorrected DEA estimates, Table 3 reports the parameters that were estimated using quasi-maximum likelihood methods.

Overall, the results seem to corroborate the key findings reported in Table 2. Specifically, we continue to find no evidence of any beneficial relationship between bank efficiency and compliance with the BCPs.

[Tables 4a and 4b: Fractional logit regressions]