Journal Issue
Share
Article

Does Compliance with Basel Core Principles Bring Any Measurable Benefits?

Author(s):
International Monetary Fund. Research Dept.
Published Date:
June 2006
Share
  • ShareShare
Show Summary Details

Following the series of crises in the 1990s, intensified attention to financial sector vulnerabilities has led to the adoption of a number of financial sector standards at the international level.1 The Basel Core Principles for Effective Banking Supervision (BCP), introduced in 1997 by the Basel Committee on Banking Supervision, are one of the most important standards, largely because of the dominant position banks have in many financial systems as well as the potentially serious macroeconomic consequences of banking instability.2

The International Monetary Fund (IMF) and the World Bank have been leading the BCP compliance assessments, mostly in the context of the Financial Sector Assessment Program (FSAP).3 The BCP assessments have been among the most rigorous, with detailed Core Principles Methodology, two assessors conducting each assessment, and a thorough review of the draft assessment to ensure consistency. Arrangements with cooperating supervisory agencies and central banks have ensured the participation of experienced experts in the assessments.

The introduction of international financial standards and the first assessment results have understandably generated interest in exploring the relationship between the observance of standards and the functioning of the financial sector. Recent work includes papers by Christofides, Mulder, and Tiffin (2003), who studied the impact of the observance of a variety of standards on spreads and ratings; Das, Quintyn, and Chenard (2004), who explored the link between financial sector soundness and regulatory governance; and Glennerster and Shin (2003), who focused on the effects of transparency on borrowing costs.

Despite the considerable attention the BCP has received in FSAPs and other IMF work, there is limited evidence about the relationship between compliance with the BCP and performance of the banking system. An initial attempt to explore this link was offered by Sundararajan, Marston, and Basu (2001). Their paper presented an empirical examination of the relationship between compliance with BCP (measured by a BCP noncompliance indicator constructed from the results of BCP assessments) and nonperforming loan (NPL) ratios and spreads between lending and risk-free rates. Their results suggested that BCP noncompliance had no direct effect either on the level of NPLs or on the level of lending spread, but that it could influence credit risk and soundness indirectly through its interaction with other macroeconomic and banking sector factors. The analysis provided by Sundararajan, Marston, and Basu (2001) should be considered preliminary because of the rather severe limitation of data then available. Separately, Barth, Caprio, and Levine (2002) examined the relationship between specific regulatory and supervisory practices and banking sector development, efficiency, and fragility and found little evidence of any impact of official supervisory power or bank activity restrictions on interest margins or NPLs.4

This paper reexamines the relationship between banking sector performance and the quality of regulation and supervision, as measured by compliance with the BCP. The basic question we address is whether following the BCP creates a regulatory and supervisory environment that helps improve banking sector performance. We use two of the common measures of banking sector performance: NPLs and net interest margin. The level of NPLs reflects the degree to which banks are able to perform one of their basic functions, that is, collect the money they lend. While there may be different reasons for an increase in NPLs, a high level of NPLs almost universally indicates serious problems in the banking sector. Net interest margin can be interpreted as a measure of the efficiency of banking sector performance, because it indicates the cost of banking intermediation that needs to be paid by banks’ customers.5

We use a new data set and different methodology than previous literature. Using panel data from 1998 to 2002, we estimated a model explaining the variation of the ratio of NPLs across 65 countries that went through the BCP assessment. Data from the 1998–2001 World Bank financial system structure database were used to estimate a model of net interest margin for the same set of countries. We include an index of BCP compliance in both models to explore whether BCP compliance has any measurable impact on banking sector performance after taking into account other determinants of NPLs and net interest margin.

Our results suggest that a higher degree of compliance with the BCP has a significant positive impact on asset quality of banks (as measured by the ratio of NPLs), even after taking into account the level of development of the economy and macroeconomic factors. We also find evidence that a higher degree of compliance with the BCP is associated with lower net interest margin. An effort to improve compliance with the BCP should therefore have a positive impact on banking sector performance across countries.6

I. Models and Data Description

BCP Compliance

We constructed a simple index of overall BCP compliance from assessments conducted mostly during FSAPs.7 We use detailed information about each assessment, including a four-grade rating for each core principle.8 For the 65 countries in our sample, we have assigned values to assessment grades—compliant (4), largely compliant (3), materially noncompliant (2), and noncompliant (1).9 The value of the index of overall compliance for a given country is equal to the sum of ratings for individual core principles.10 Therefore, the actual values of the index of overall BCP compliance will be between 30 and 120, with higher values indicating a higher degree of compliance.11

Our sample includes 13 advanced countries, 19 emerging market countries, and 33 developing countries. Figure 1 confirms that advanced countries achieved the highest level of BCP compliance, followed by emerging and developing countries. The variance of results, as measured by the difference between best and worst results in each group, also increases from advanced countries to emerging countries, and further to developing countries.

Figure 1.Compliance with the Basel Core Principles for Effective Banking Supervision (BCP) by Country Group1

Source: Author’s calculations.

1 Error bars denote minimum and maximum observations.

We also construct several subindices of BCP, using different groupings of the Core Principles and using the same procedure as described above for the overall BCP compliance. These include (1) objectives, autonomy, and powers of the supervisor; (2) licensing and structure; (3) prudential regulations; (4) methods of ongoing supervision; and (5) cross-border banking.12 The correlation matrix of these subindices presented in Table 1 suggests that the assessment results of the parts of BCP are rather closely correlated.

Table 1.Correlation Matrix of Main BCP Components
Objectives, Autonomy, PowersLicensing and StructurePrudential RegulationsMethods of SupervisionCross-Border BankingAll Core Principles
Objectives, autonomy powers (CP1)1.000.660.660.650.630.84
Licensing and structure (CP 2–5)1.000.680.670.570.80
Prudential regulations(CP 6–15)1.000.840.540.93
Methods of supervision (CP 16–20)1.000.640.90
Cross-border banking (CP 23–25)1.000.73
All core principles1.00
Source: Author’s calculations.
Source: Author’s calculations.

A higher degree of compliance appears to be associated with lower NPLs and narrower net interest margin. Figures 2 and 3 show that actual BCP compliance exhibits a considerable variation in our sample, from the perfect score of 120 to a rather low value of just over 50. The two measures of banking performance show substantial variation across the sample. There appears to be a negative relationship between BCP compliance and the two banking measures, with high compliance being associated with more favorable outcomes—that is, lower NPLs and narrower margins. This relationship appears to be tighter for countries with higher compliance, because the dispersion of observations increases with decreasing compliance—this holds for both NPLs and net interest margin.

Figure 2.Nonperforming Loans (NPLs) and Index of Overall BCP Compliance

Source: Author’s calculations.

Figure 3.Net Interest Margin and Index of Overall BCP Compliance

Source: Author’s calculations.

Model of NPLs

We model the share of NPLs in total loans as a function of macroeconomic variables in previous years (economic growth, changes in inflation, real interest rates, and exchange rates), a measure of compliance with BCP, and variables controlling for the level of development of the economy (or of the financial sector):

for panel data i = 1, …, 65 countries, t = 1, …, 5 years (1998–2002), where

  • bcp represents an index of compliance with the BCP (two different indices, bcp_all, a measure of the overall compliance, and bcp_pru a measure of compliance with the core principles related to prudential regulations, Core Principles 6–15).13 Only one observation of BCP compliance per country is available (from one assessment for each country performed mostly from 1999 to 2001) and this is assumed to remain constant over the five years;14
  • growth stands for the GDP growth; we include lagged growth (t−1) in the model;
  • cpi_ch denotes the change of consumer inflation in the previous two years;
  • real_ir_ch denotes a change of real lending rates over the previous two years;
  • exch_rate stands for the change the country’s exchange rate (we use a change in the nominal exchange rate relative to the U.S. dollar, exch, and a change in the real effective exchange rate, reer) and
  • development represents a variable capturing the level of development of the country’s economy or its financial system. We use the level of GDP per capita in purchasing power parity, ppp_gdp, as a measure of the level of development of the economy (and a proxy for the development of the country’s legal and financial system) as well as the ratio of M2 to GDP, m2/gdp, as a proxy for the degree of development of the financial system.

The temporal structure of the model reflects the expectation that different shocks will affect NPLs with different lags. Therefore, we include a contemporaneous change in the exchange rate, real GDP growth with one lag, and changes in inflation and real interest rates over two years. Appendix III provides information on data sources.15

We expect the parameters β1, β2; and β6 to be negative, because higher compliance with the BCP, higher economic growth, and higher level of economic (or financial) development can be expected to have a positive impact on asset quality in banks (that is, lower NPL ratios). Parameter β4 is expected to be positive, because an increase of real interest rates would be expected to worsen asset quality, making loan repayment more difficult. Parameters β3 and β5 could be either positive or negative, because the effect of accelerating inflation depends on whether its acceleration was anticipated or not, the flexibility of lending rates, and whether the acceleration signals general economic instability. For the exchange rate, the effect depends on the composition of outstanding credit (that is, the size of unhedged positions and the share of borrowers whose businesses benefit from a given change in the exchange rate); large exchange rate movements can signal general economic instability as well.

One of the econometric issues we face is the quality of NPL data. First, differences in definitions across countries can result in measurement errors. As long as we can assume that the measurement error of the dependent variable is uncorrected with the regressor(s), it can be absorbed in the disturbance of the regression and ignored. Second, the actual level of NPLs could be underreported in some countries, mostly in those with weak regulation and supervision. This would bias our estimates of the impact of BCP compliance on NPLs downward, that is, against us finding a significant relationship.16

Model of Net Interest Margin

The model of net interest margin explains the cross-country variation of the margin as a function of structural characteristics of the banking sector (overhead as an indicator of operating costs, and ratio of NPLs as an indicator of lending risks), a macroeconomic indicator (consumer price inflation) serving as a proxy of macro-economic stability, a measure of compliance with BCP, and measures of the level of development of the economy and of the financial system:

for panel data i = 1, …, 65 countries, t = 1, …, 4 years (1998–2001), where

  • bcp represents an index of compliance with BCP, as in equation (1) (again, we use both a measure of the overall compliance and a measure of compliance with the prudential core principles);
  • overhead stands for bank overhead costs as a share of total assets;
  • npl denotes a ratio of nonperforming loans;
  • cpi stands for the consumer price inflation; and
  • development represents a variable capturing the level of development of the country’s economy or its financial system. Here, we use (1) the level of GDP per capita in purchasing power parity, ppp_gdp;(2) the ratio of M2 to GDP, m2/gdp; and (3) the total financial system deposits as a share of GDP, all_dep/gdp.

We have also included a variable measuring the concentration of the banking sector and assets of the three largest banks as a share of banking system total assets, but it was not significant in any regression and we excluded it from the final model.17 Additional information about the data is provided in Appendix III.

We expect the parameters β2, β3, and β4 to be positive, with higher costs, higher NPLs, and a less stable macroeconomic environment all increasing the net interest margin that banks charge. The parameters β1 and β5, on the other hand, are expected to be negative, as better regulation and supervision and a higher level of development of the financial system should be associated with lower intermediation costs.

Econometric Issues

Several econometric issues had to be resolved, including (1) the potential endogenetity of bank regulation and supervision; (2) the choice of estimation method; and (3) potential sample selectivity bias.

The potential endogeneity of bank regulation and supervision is an issue, and we use instrumental variables to control for this problem. An effective instrumental variable needs to be correlated with the independent variable in question but uncorrected with the error term. We use two broad governance indices compiled by Kaufmann, Kraay, and Mastruzzi (2003)—an index of government effectiveness and an index of control of corruption. These indices are correlated with the index of BCP compliance, yet they are broad enough relative to the dependent variable to allow us to assume that they affect the dependent variable only through bank regulation and supervision. We also test whether these two indices are not rejected as valid instruments by the data (as described below).

The government effectiveness index is set up to measure the quality of public service provision, the quality of the bureaucracy, the competence of civil servants, the independence of the civil service from political pressures, and the credibility of the government’s commitment to policies. The index of control of corruption measures perceptions of corruption, with different measurement sources varying from the frequency of “additional payments to get things done,” to the effects of business corruption, to measuring “grand corruption” in the political arena.18

To estimate both models, we use a generalized method of moments (GMM) estimator that is robust to heteroscedasticity and autocorrelation on the pooled sample. This estimator allows us to test the validity of the instruments by imposing the orthogonality conditions that the instrumental variables are not correlated with the error term. The Hansen (1982) test of overidentifying restrictions can be interpreted as a test of whether the instruments are associated with the level of NPLs beyond their ability to explain compliance with the BCP (as a measure of bank regulation and supervision). If the null hypothesis is not rejected, then the data do not reject the validity of the instruments.19

We also estimate two panel data models with an adjustment for heteroscedasticity and the same instruments as above. These are (1) regression on country means over the sample period (a “between” estimator); and (2) a random effects model, which assumes that the individual country intercepts are drawn from a common distribution.20 For the estimates of the random effects model to be consistent, the individual intercepts cannot be correlated with independent variables. We test this correlation through the Hausman test statistic (H0 = random effects).

The FSAP and stand-alone assessments of the BCP are voluntary, so it is possible that countries that would score poorly did not participate in the program, which would create sample selectivity bias in our estimates. In reality, the decision to participate is related to a number of factors, including peer pressure, cooperation with the IMF in other areas (for example, through a program), and domestic political issues. The range of observed compliance with the BCP in our sample is relatively wide, so there does not appear to be a “cutoff compliance below which countries would opt out of the FSAP program or assessment.

Nevertheless, to explore the importance of the sample selectivity problem, we have followed the Heckman two-step procedure (for more details see, for instance, Davidson and MacKinnon, 1993). First, assuming that a “reservation compliance rating” exists, below which countries opt out of the assessment and above which they participate, we have enlarged our sample to include countries for which BCP assessment was not undertaken. Given data availability in the IMF International Financial Statistics database, our enlarged sample consists of 126 countries, including the 65 countries in our original sample. We have then estimated a probit model explaining the availability of BCP assessment for the enlarged sample, using macroeconomic variables (GDP growth and inflation), the level of development (income level per capital in purchasing power parity, PPP), and an indirect indication of the quality of banking supervision (the incidence of banking crises from 1995 to 2002).21 Second, based on the probit estimates, we have calculated the Heckman selectivity regressor (inverse Mills ratio) and incorporated it into our models of NPLs and the net interest margin to test whether selectivity is indeed a problem in our data set.22

II. Results

The estimation results for the nonperforming loans model, presented in Table 2, suggest that compliance with the BCP is indeed associated with a lower share of NPLs.23 Estimates of the parameter β1 are statistically significant and negative, implying that higher observance of the BCP is associated with lower NPLs. This holds for overall compliance and compliance with prudential core principles.24 The difference in magnitude of estimated coefficients is due to the different scale of the two indices.

Table 2.Nonperforming Loans Model: Results
cbcp_allbcp_prugrowth(−1)cpi_chreal_ir_chexchPPP-gdp1Heckman Selectivity RegressorR2NOIR-Test2
GMM/IV estimates (pooled sample)3
Baseline withBCP_all31.04***0.17***−0.42**0.060.070.04*−0.32***0.332745.97
(4.94)(0.06)(0.18)(0.04)(0.05)(0.02)(0.08)
Baseline with BCPJPMf26.28***−0.35***−0.41**0.070.070.04*−0.34***0.322745.77
(3.77)(0.14)(0.18)(0.05)(0.05)(0.02)(0.07)
Adding Heckman correction37.53***−0.17***−0.43**0.050.060.05**−0.39***−7.400.332745.68
(BCP_ALL)(8.08)(0.06)(0.17)(0.05)(0.05)(0.02)(0.09)(6.18)
Adding Heckman correction30.59***−0.36***−0.42**0.060.070.05**−0.40***−5.260.322745.59
(BCP_PRU)(6.69)(0.14)(0.18)(0.05)(0.05)(0.02)(0.09)(6.04)
Insignificant variables dropped30.10***−0.16***−0.280.05*−0.33***0.302906.53
(BCP_ALL)(4.93)(0.06)(0.19)(0.02)(0.08)
Insignificant variables dropped25.76***−0.36***−0.260.04*−0.34***0.302906.22
(BCP_PRU)(3.84)(0.14)(0.19)(0.02)(0.07)
BETWEEN estimator (OLS on means)4
Baseline with BCP_ALL60.56***−0.52***−0.440.27*0.29*0.010.020.43274
(15.11)(0.19)(0.39)(0.14)(0.16)(0.06)(0.23)
Baseline with RCP_PRV45.89***−1.15**0.440.26*0.28*0.01−0.010.43274
(10.62)(0.44)(0.39)(0.14)(0.16)(0.06)(0.23)
Adding Heckman correction56.80***−0.54***−0.460.30*0.32*−0.010.127.180.43274
(BCP_ALL)(16.92)(0.20)(0.39)(0.15)(0.17)(0.06)(0.27)(10.89)
Adding Heckman correction41.67***−1.19***−0.460.29*0.31*0.010.106.990.43274
(BCP_PRU)(12.54)(0.44)(0.39)(0.15)(0.17)(0.07)(0.27)(10.91)
Insignificant variables dropped58.92***−0.52***0.27*0.29*0.42289
(BCP_ALL)(7.22)(0.08)(0.14)(0.16)
Insignificant variables dropped44.77***−1.16***0.27*0.29*0.42289
(BCP_PRU)(5.11)(0.18)(0.14)(0.16)
Source: Author’s estimates.

Notes: Standard errors in parentheses. *** denotes significance at the 1 percent level, ** at 5 percent and * at 10 percent. Variable definitions: c stands for a constant, bcp_all measures overall BCP compliance, bcp_pru measures compliance with core principles related to prudential regulations, growth stands for GBP growth, cpi_ch denotes the change of consumer inflation in the previous two years, real_ir_ch denotes a change of real lending rates over the previous two years, exch denotes a change in the nominal exchange rate, and ppp_gdp stands for per capita GDP in PPP.

Reported coefficients and standard errors multiplied by 103.

Test of overidentifying restrictions (chi-square distribution, 2 degrees of freedom, 5 percent critical value is 6.0).

Generalized method of moments estimator, robust to heteroscedasticity and autocorrelation. Indices of government effectiveness and control of corruption were used as instruments. Estimated using the GMM routine in TSP 4.5,

Estimated using the PANEL routine in TSP 4.5. Indices of government effectiveness and control of corruption used as instruments.

Source: Author’s estimates.

Notes: Standard errors in parentheses. *** denotes significance at the 1 percent level, ** at 5 percent and * at 10 percent. Variable definitions: c stands for a constant, bcp_all measures overall BCP compliance, bcp_pru measures compliance with core principles related to prudential regulations, growth stands for GBP growth, cpi_ch denotes the change of consumer inflation in the previous two years, real_ir_ch denotes a change of real lending rates over the previous two years, exch denotes a change in the nominal exchange rate, and ppp_gdp stands for per capita GDP in PPP.

Reported coefficients and standard errors multiplied by 103.

Test of overidentifying restrictions (chi-square distribution, 2 degrees of freedom, 5 percent critical value is 6.0).

Generalized method of moments estimator, robust to heteroscedasticity and autocorrelation. Indices of government effectiveness and control of corruption were used as instruments. Estimated using the GMM routine in TSP 4.5,

Estimated using the PANEL routine in TSP 4.5. Indices of government effectiveness and control of corruption used as instruments.

Most other parameter estimates have the expected sign. Higher growth in previous years helps reduce NPLs, but an increase of real interest rates and an acceleration of inflation worsen bank asset quality, as measured by NPLs. The impact of nominal exchange rate depreciation appears to be negative, that is, a depreciation would have a negative impact on asset quality (higher NPLs). The exchange rate results suggest that, in our sample, the negative impact of a depreciation on asset quality because of unhedged positions is greater than its positive impact on borrowers benefiting from a weaker currency (exporters and producers of tradable goods).25

Countries with higher GDP per capita tend to have lower NPLs. GDP per capita expressed in PPP, used as a proxy for an overall financial and economic development, is a significant explanatory variable, and the parameter estimates have the expected negative sign. A somewhat more direct measure of financial sector development, m2/gdp, was not statistically significant.

Overall, the models provide a reasonably good explanation of cross-country variation in the share of NPLs. The models that attempt to explain variation across all available observations (pooled sample model) explain more than 30 percent of the variation, whereas the “between” estimator explains more than 40 percent of the variation of country means.

The estimation results for the net interest margin model are similar. The results of the GMM-estimated pooled sample model, presented in Table 3, suggest that a higher level of BCP compliance does help reduce intermediation costs (net interest margin). However, the BCP compliance coefficients in the model of country means are not statistically significant, although they have the expected sign.

Table 3.Net Interest Margin Model; Results
cbcp_all1bcp_pru1overheadnpl1cpi1ppp_gdp2Heckman Selectivity RegresserR2NOIR-Test3
GMM/IV estimates (pooled sample)4
Baseline with BCP_ALL0.04***−0.19*0.68***−0.290.76***−0.80***0.702412.98
(0.01)(0.11)(0.10)(0.18)(0.27)(0.22)
Baseline with BCP_PRU0.04***−0.64*0.70***−0.33*0.75***−0.70***0.702412.20
(0.01)(0.34)(0.11)(0.18)(0.27)(0.24)
Adding Heckman correction0.02−0.170.66***−0.270.62**−0.60**0.020.702413.58
(BCP_ALL)(0.02)(0.11)(0.10)(0.19)(0.26)(0.28)(0.02)
Adding Heckman correction0.02−0.63*0.68***−0.31*0.60**−0.460.030.702412.78
(BCP_PRU)(0.02)(0.34)(0.11)(0.18)(0.26)(0.29)(0.02)
BETWEEN estimator (OLS on means)5
Baseline withBCP_ALL0.07−0.450.65***−0.400.91***−0.570.75241
(0.05)(0.53)(0.11)(0.32)(0.24)(0.51)
Baseline with BCP_PRU0.06−1.050.65***−0.400.91***−0.550.75241
(0.03)(1.19)(0.11)(0.32)(0.24)(0.51)
Adding Heckman correction0.06−0.560.62***−0.420.74**−0.210.030.75241
(BCP_ALL)(0.05)(0.55)(0.12)(0.32)(0.31)(0.66)(0.04)
Adding Heckman correction0.04−1.310.61***−0.430.73**−0.200.030.75241
(BCP_PRU)(0.04)(1.22)(0.12)(0.32)(0.31)(0.65)(0.04)
Insignificant variables dropped0.07***−0.63**0.69***0.88***0.74250
(BCP_ALL)(0.02)(0.24)(0.10)(0.24)
Insignificant variables dropped0.05***−1.44***0.69***0.88***0.74250
(BCP_PRU)(0.02)(0.54)(0.10)(0.24)
Source: Author’s estimates.Notes: Standard errors in parentheses, *** denotes significance at the 1 percent level ** at 5 percent, and * at 10 percent. Variable definition: c stands for a constant, bcp_all measures overall BCP compliance, bcp_pru measures compliance with core principles related to prudential regulations, overhead stands for overhead costs as a share of total assets, npl denotes a ratio of nonperforming loans, cpi stands for consumer price inflation and ppp_gdp stands for per capita GDP in PPP.

Reported coefficients and standard errors multiplied by 103.

Reported coefficients and standard errors multiplied by 105.

Test of overidentifying restrictions (chi-square distribution, 2 degrees of freedom, 5 percent critical value is 6.0).

Generalized method of moments estimator, robust to heteroscedasticity and autocorrelation, Indices of government effectiveness and control of corruption were used as instruments. Estimated using the GMM routine in TSP 4.5.

Estimated using the PANEL routine in TSP 4.5. Indices of government effectiveness and control of corruption used as instruments.

Source: Author’s estimates.Notes: Standard errors in parentheses, *** denotes significance at the 1 percent level ** at 5 percent, and * at 10 percent. Variable definition: c stands for a constant, bcp_all measures overall BCP compliance, bcp_pru measures compliance with core principles related to prudential regulations, overhead stands for overhead costs as a share of total assets, npl denotes a ratio of nonperforming loans, cpi stands for consumer price inflation and ppp_gdp stands for per capita GDP in PPP.

Reported coefficients and standard errors multiplied by 103.

Reported coefficients and standard errors multiplied by 105.

Test of overidentifying restrictions (chi-square distribution, 2 degrees of freedom, 5 percent critical value is 6.0).

Generalized method of moments estimator, robust to heteroscedasticity and autocorrelation, Indices of government effectiveness and control of corruption were used as instruments. Estimated using the GMM routine in TSP 4.5.

Estimated using the PANEL routine in TSP 4.5. Indices of government effectiveness and control of corruption used as instruments.

Most other parameter estimates in table 3 have the expected sign and are statistically significant. Higher overhead costs clearly contribute to higher net interest margins, as does higher inflation. Also, a higher degree of development of the economy or the financial system as measured by GDP per capita is associated with lower net interest margins.26 However, the estimated coefficient for the impact of NPLs does not have the expected sign. This is difficult to explain, but it could be partly caused by the fact that it is the future probability of default that is being priced in the net interest margin, while the indicator of NPLs largely looks backward. The explanatory power of the models is rather good, as they explain 70–75 percent of the variation in net interest margins.

Because the Heckman selectivity regressor was not significant in either model, it appears that self-selection is not a major problem in our sample. We have explored the stability of the coefficients relative to sample selection by randomly dropping the observations for five countries and reestimating the pooled sample model by GMM. The results appear to be rather robust, as the magnitude of the estimates remained approximately the same and they remained statistically significant. Because our sample overlaps only partially with that of Sundararajan, Marston, and Basu (2001) and our methodology is considerably different, we were unable to compare the results more directly.27

In most cases, the null hypothesis of the test of overidentifying restrictions was not rejected, implying that the data do not reject the validity of the instruments. We have also confirmed, by a least-squares regression with instruments entered directly as right-hand-side variables, that the instruments do not have a direct explanatory power in most model specifications.

To explore whether the relationship between BCP compliance and the dependent variables is primarily driven by one type of country, we have estimated the models on two subsamples of approximately equal size: (1) advanced and emerging markets and (2) developing countries. The results were not very strong, possibly also because of a lower number of observations in the regressions. For NPLs, the results did suggest that the relationship may be stronger for advanced and emerging markets; for these, the BCP coefficient was statistically significant and larger than in the regression for the whole sample, whereas for developing countries, the coefficient was smaller and statistically insignificant. For net interest margin, neither subsample yielded a statistically significant BCP coefficient.28

III. Conclusion

This paper explores the relationship between banking sector performance and the quality of regulation and supervision as measured by compliance with the BCP. We use a new data set and different methodology than the initial attempt by Sundararajan, Marston, and Basu (2001). BCP assessment results for 65 countries are used, along with 1998–2002 panel data for NPLs and other explanatory variables. For the net interest margin, we use 1998–2001 data from the World Bank financial system structure database.

We find a direct positive effect of compliance with the BCP on banking sector performance, as measured by the share of NPLs and the net interest margin. Higher compliance with the BCP is associated with lower NPLs and lower net interest margin, suggesting that following the BCP creates a regulatory and supervisory environment that helps improve banking sector performance.

Our understanding of the interaction between banking regulation and supervision and banking sector performance and development is far from complete and there is substantial room for further research. Important data limitations continue to pose problems—most important, the lack of consistent data on regulation and supervision over time, which, if available, would make it possible to make full use of panel data techniques and explore the impact of changes in regulation and supervision on the banking sector.

APPENDIX I: Basel Core Principles for Effective Banking Supervision
C1Chapter 1. Objectives, Autonomy, Powers, and Resources (CP 1)
CP1Principle 1. Objectives, Autonomy, Powers, and Resources
SP11Principle 1(1). An effective system of banking supervision will have clear responsibilities and objectives for each agency involved in the supervision of banks.
SP12Principle 1(2). Each such agency should possess operational independence and adequate resources.
SP13Principle 1(3). A suitable legal framework for banking supervision is also necessary, including provisions relating to authorization of banking establishments and their ongoing supervision.
SP14Principle 1(4). A suitable legal framework for banking supervision is also necessary, including powers to address compliance with laws, as well as safety and soundness concerns.
SP15Principle 1(5). A suitable legal framework for banking supervision is also necessary, including legal protection for supervisors.
SP16Principle 1(6). Arrangements for sharing information between supervisors and protecting the confidentiality of such information should be in place.
C2Chapter 2. Licensing and Structure (CPs 2–5)
CP2Principle 2. Permissible Activities
CP3Principle 3. Licensing Criteria
CP4Principle 4. Ownership
CP5Principle 5. Investment Criteria
C3Chapter 3. Prudential Regulations and Requirements (CPs 6–15)
CP6Principle 6. Capital Adequacy
CP7Principle 7. Credit Policies
CP8Principle 8. Loan Evaluation and Loan-Loss Provisioning
CP9Principle 9. Large Exposure Limits
CP10Principle 10. Connected Lending
CP11Principle 11. Country Risk
CP12Principle 12. Market Risks
CP13Principle 13. Other Risks
CP14Principle 14. Internal Control and Audit
CP15Principle 15. Money Laundering
C4Chapter 4. Methods of Ongoing Supervision (CPs 16–20)
CP16Principle 16. On-Site and Off-Site Supervision
CP17Principle 17. Bank Management Contact
CP18Principle 18. Off-Site Supervision
CP19Principle 19. Validation of Supervisory Information
CP20Principle 20. Consolidated Supervision
C5Chapter 5. Information Requirements (CP 21)
CP21Principle 21. Accounting Standards
C6Chapter 6. Formal Powers of Supervisors (CP 22)
CP22Principle 22. Remedial Measures
C7Chapter 7. Cross-Border Banking (CP 23–25)
CP23Principle 23. Globally Consolidated Supervision
CP24Principle 24. Host Country Supervision
CP25Principle 25. Supervision over Foreign Banks’ Establishments
APPENDIX II: List of Economies in the Sample
1 Albania34 Kenya
2 Armenia35 Korea
3 Austria36 Kuwait
4 Bangladesh37 Kyrgyz Republic
5 Bolivia38 Latvia
6 Brazil39 Lithuania
7 Bulgaria40 Luxembourg
8 Cameroon41 Macedonia, FYR
9 Colombia42 Malta
10 Costa Rica43 Mauritius
11 Croatia44 Morocco
12 Czech Republic45 Mozambique
13 Dominican Republic46 Nigeria
14 Egypt, Arab Rep. of47 Oman
15 El Salvador48 Peru
16 Estonia49 Philippines
17 Finland50 Poland
18 France51 Russia
19 Gabon52 Slovak Republic
20 Germany53 Slovenia
21 Ghana54 South Africa
22 Guatemala55 Sri Lanka
23 Hong Kong SAR56 Sweden
24 Hungary57 Switzerland
25 Iceland58 Tanzania
26 India59 Thailand
27 Indonesia60 Tunisia
28 Ireland61 Turkey
29 Israel62 Uganda
30 Italy63 Ukraine
31 Jamaica64 United Kingdom
32 Japan65 Zambia
33 Kazakhstan
APPENDIX III: Additional Data Information

npl—gross nonperforming loans as a share of total gross loans; source: provisional Monetary and Financial Systems Department’s Financial Soundness Indicators database, original data from Financial Sector Assessment Programs, IMF’s Economic Data Sharing System, and central banks’ publications;

growth—real GDP growth; source: International Financial Statistics (IFS), IMF staff reports where IFS data were missing;

cpi_ch—a change of consumer price inflation in percentage points over the previous two years, that is, cpit-cpit-2; source: IFS, IMF staff reports where IFS data were missing;

real_ir_ch—a change of real lending rates in percentage points over the previous two years, that is, real_irt-real_irt2; ex post consumer price inflation used to estimate real lending rate; source: nominal lending rates obtained from IFS;

exch—an annual change in the nominal exchange rate relative to the U.S. dollar, source: IFS;

reer—an annual change in the real effective exchange rate; source: mostly IFS, IMF staff reports to replace missing data;

PPP-gdp—the level of GDP per capita in purchasing power parity (PPP) in U.S. dollar; source: William Davidson Institute Database; available only through 2001, so the 2001 data were used for 2002 GDP at PPP as well;

m2/gdp—the ratio of M2 (money + quasi money) to nominal GDP; source: IFS;

margin—net interest margin, an accounting value of a bank’s net interest revenue as a share of its interest-bearing (total earning) assets; source: World Bank Financial Structure Database; original data from Fitch’s Bankscope database;

overhead—an accounting value of a bank’s overhead costs as a share of its total assets; source: World Bank Financial Structure Database; original data from Fitch’s Bankscope database;

concentr—assets of the three largest banks as a share of assets of all commercial banks in the system; source: World Bank Financial Structure Database; original data from Fitch’s Bankscope database;

cpi—annual consumer price inflation; source: IFS;

all_dep/gdp—demand, time, and saving deposits in deposit money banks and other financial institutions as a share of GDP, calculated using the following deflation method: {(0.5)*[Ft/P_et + Ft−1/P_et]}/[GDPt/P_at] where F is demand and time and saving deposits, P_e is end-of period consumer price inflation, and P_a is average annual consumer price inflation; source: World Bank Financial Structure Database; original data from IFS.

REFERENCES

    BarthJames R.Gerard CaprioJr. and RossLevine2001The Regulation and Supervision of Banks Around the World: A New Database” (Washington: World Bank).Available via the Internet: www.worldbank.org/research/projects/bank_regulation.htm

    • Search Google Scholar
    • Export Citation

    BarthJames R.Gerard CaprioJr. and RossLevine2002“Bank Regulation and Supervision: What Works Best?”NBER Working Paper No. 9323 (CambridgeMassachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    BeckThorstenAshDemirgiic-Kunt and RossLevine2004Bank Concentration Competition and Crises: First Results” (unpublished; Washington: World Bank).

    • Search Google Scholar
    • Export Citation

    ChristofidesCharisChristianMulder and AndrewTiffin2003The Link Between Adherence to International Standards of Good Practice, Foreign Exchange Spreads, and Ratings,IMF Working Paper 03/74 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    DasUdaibirS.Marc Quintyn and KinaChenard2004“Does Regulatory Governance Matter for Financial System Stability? An Empirical Analysis,”IMF Working Paper 04/89 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    DavidsonRussell and James G.MacKinnon1993Estimation and Inference in Econometrics(Oxford: Oxford University Press).

    Demirgüç-KuntAsli and EnricaDetragiache2002“Does Deposit Insurance Increase Banking System Stability? An Empirical Investigation,”Journal of Monetary Economics Vol. 49 (October) pp. 13731406.

    • Search Google Scholar
    • Export Citation

    Demirguc-KuntAsliLucLaeven and RossLevine2003“Regulations, Market Structure, Institutions, and the Cost of Financial Intermediation,”NBER Working Paper No. 9890 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    GlennersterRachel and YongseokShin2003“Is Transparency Good for You, and Can the IMF Help?”IMF Working Paper 03/132 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    HansenLars Peter1982“Large Sample Properties of Generalized Method of Moments Estimators,”Econometrica Vol. 50 (July) pp. 102954.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund and World Bank2002“Implementation of the Basel Core Principles for Effective Banking Supervision, Experiences, Influences, and Perspectives.”Available via the Internet: http://www.imf.org/external/np/mae/bcore/2002/092302.pdf

    • Search Google Scholar
    • Export Citation

    International Monetary Fund and World Bank2003“Financial Sector Assessment Program—Review, Lessons, and Issues Going Forward.”Available via the Internet: http://www.imf.org/external/np/fsap/2003/review.pdf

    • Search Google Scholar
    • Export Citation

    KaufmannDanielAartKraay and MassimoMastruzzi2003“Government Matters III: Governance Indicators for 1996-2002,”Policy Research Working Paper No. 3106 (Washington: World Bank).

    • Search Google Scholar
    • Export Citation

    SundararajanV.DavidMarston and RituBasu2001“Financial System Standards and Financial Stability: The Case of the Basel Core Principles,”IMF Working Paper 01/62 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
*Richard Podpiera is an Economist, IMF Monetary and Financial Systems Department. The author is grateful to Adolfo Barajas, Patricia Brenner, Martin Čihák, Udaibir S. Das, Gilda Fernandez, Plamen lossifov, Matthew Jones, Mark O’Brien, Jiří Podpiera, Marc Quintyn, Mark Stone, Jan Willem van der Vossen, participants of a Monetary and Financial Systems Department seminar, and two anonymous referees for useful comments. Kalin Tintchev provided research assistance.
1A compendium highlighting the 12 key standards for sound financial systems can be found at the Financial Stability Forum website (www.fsforum.org).
2A review of the Basel Core Principles has commenced recently.
3IMF and World Bank (2003) provides the most recent review of the FSAP program.
4In a recent related paper, Demirguc-Kunt, Laeven, and Levine (2003) examined the impact of bank regulations, market structure, and national institutions on bank interest margins and overhead costs, and concluded that tighter regulations on bank entry and bank activities boost the cost of financial intermediation, along with inflation. They also found, however, that bank regulations become insignificant when controlling for indicators of economic freedom or property rights protection.
5Large net interest margins often indicate inefficient banking operations, high risks in lending, and monopoly power of banks; thus, lower margins would be preferable. However, in some cases, overcompetition could temporarily depress margins so low that financial stability may be threatened.
6We effectively test a joint hypothesis that (i) the quality of banking supervision and regulation matters for the performance of the banking system, and (ii) the BCP and the assessments measure the relevant features of quality of banking supervision and regulation. The theory does not offer a clear prediction of the impact of more intensive regulation and supervision on bank performance; for a more detailed discussion, see Barth, Caprio, and Levine (2002).
7A few assessments were conducted on a stand-alone basis.
8Appendix I provides the list of the Basel Core Principles. See IMF and World Bank (2002) and Sundararajan, Marston, and Basu (2001) for useful background on BCP assessments.
9The list of countries in our sample is provided in Appendix II.
10In several assessments, some Core Principles were either “not assessed” or “not applicable” these were assigned an average value of BCP compliance of other principles in a given country, so that these countries were not penalized.
11We treat the six subcategones of Core Principle 1 (CP 1) as separate principles. This does not have any significant impact on the index—the correlation coefficient of our index (with six subcategories treated as separate principles) and an index with only one entry for CP 1 (equal to the average of the six subcategories) is 0.991.
12Appendix I provides the list of the BCP included in each of these categories.
13The BCP include CP 15 (Money Laundering) under Prudential Regulations and Requirements, although it reflects market integrity rather than prudential requirements. We have calculated both indices, with and without CP 15, and found that there is virtually no difference between them (correlation coefficient of 0.993). We use the official definition and include CP 15 in the prudential subindex.
14While this is clearly a limiting assumption, it appears reasonable because the level of BCP compliance—particularly the practical application of the core principles—is unlikely to change quickly and it is likely that there will be a substantial time lag before any impact of changed regulation and supervision becomes observable in banking system performance.
15Data limitations prevented us from including a measure of government ownership and foreign ownership in the financial system. This information is available in the database provided by Barth, Caprio, and Levine (2001), but only for one time and for fewer than 50 countries in our sample. Including these two variables and using a limited sample (assuming the ownership data are constant across 1998–2002) did not change the results to any substantial extent and only the government ownership variable was statistically significant (and positive).
16For instance, assume that the true NPL ratio is npl* and we can only observe an underreported ratio npl, with (i) a measurement error ξ independent of all other variables and the overall disturbance of the regression; and (ii) underreporting proportional to our measure of the quality of banking regulation and supervision (BCP). Then npl* = npl + φ (120 − bcp) + ξ for φ > 0, that is, the lower the quality, the higher the underreporting. Denote the coefficient reflecting the impact of BCP compliance on npl* as β1 and assume it is negative. If we use npl as the dependent variable in our regression instead of npl*, we actually estimate (β1 + φ), a smaller coefficient in absolute value than if we could use npl* because β1 < 0 and φ > 0. The measurement error ξ will be absorbed in the disturbance of the regression.
17The same data limitations concerning government ownership and foreign ownership described for the NPL model above apply here.
19The test statistic has chi-square distribution with two degrees of freedom (for two instruments, that is, two overidentifying restrictions).
20We also run the basic pooled regression (equivalent to the model estimated by GMM). The fact that we have only one observation for BCP compliance does not allow us to estimate a fixed-effects model in which the parameter at the BCP compliance variable would be identified. This also implies that the coefficient at the BCP index could reflect the differences in means across countries not picked up by other explanatory variables. We have tried to address this problem by including a variable reflecting early participation in an FSAP, which could capture unobserved characteristics of the countries in our sample. The variable was statistically significant only in some model specifications and did not significantly change the estimates of the BCP coefficients.
21For the definition of banking crises, we have used the data from Beck, Demirgüç-Kunt, and Levine(2004), who updated the data prepared by Demirgüç-Kunt and Detragiache (2002). The sources of other data for the enlarged sample are the same as for the original sample; see Appendix III for more details.
22The Heckman two-step procedure provides both a correction and a test of the selectivity bias problem.
23The Hausman test rejected the validity of the random effects model in almost all specifications, so we do not report the estimates here.
24Similar results were obtained using some other subcomponents of the BCP described in Table 1: CPs 1–5 (objectives, autonomy, power, resources, licensing, and structure), CPs 16–20 (methods of ongoing supervision), and CPs 23–25 (cross-border banking).
25The real exchange rate was not significant in virtually any specification. We therefore included the nominal exchange rate measure in the final model.
26Higher m2/gdp and the ratio of total financial system deposits to GDP were also associated with a lower net interest margin (when substituted for GDP per capita). We also included a measure of concentration of the banking sector into the model, but it was not statistically significant.
27Our sample includes 24 of the 35 countries listed by Sundararajan, Marston, and Basu (2001), partly because of the exclusion of several very early assessments from our sample. These were done before the assessment methodology was fully developed. Another factor complicating the replication of their sample is the fact that, according to the reported results, only 24–29 countries were actually used in the estimation.
28Details are available from the author upon request.

Other Resources Citing This Publication