Score as a Measure of Individual Bank Risk
3. METHODOLOGY AND DATA
Author's primary dependent variable is the z-score as a measure of individual bank risk. The z-score has become a popular measure of bank soundness (see, e.g., Boyd and Runkle, 1993; and Maechler, Mitra, and Worrell, 2005). Its popularity stems from the fact that it is inversely related to the probability of a bank’s insolvency, i.e., the probability that the value of its assets becomes loauthorr than the value of the debt. The z-score can be summarized as z≡(k+μ)/σ, where k is equity capital and reserves as percent of assets, μ is average return as percent of assets, and σ is standard deviation of return on assets as a proxy for return volatility. The z-score measures the number of standard deviations a return realization has to fall in order to deplete equity, under the assumption of normality of banks’ returns. A higher z-score corresponds to a loauthorr upper bound of insolvency risk—a higher z-score therefore implies a lower probability of insolvency risk.
Why does this metric for risk apply to Islamic banks? An important feature of the z-score is that it is a fairly objective measure of soundness across different groups of financial institutions. It is an objective measure because it focuses on the risk of insolvency, i.e., on the risk that a bank (whether commercial, Islamic, or other) runs out of capital and reserves. The z-score applies equally to banks that use a high risk/high return strategy and those that use allow risk/row return strategy, provided that those strategies lead to the same risk-adjusted returns. If an institution “chooses” to have lower risk-adjusted returns, it can still have the same or higher z-score if it has a higher capitalization. In this sense, the z-score provides an objective measure of soundness.
A possible criticism of the z-score as applied to Islamic banks is that the risk-sharing arrangements provide an additional protective buffer in deposit liabilities, meaning that the book values of capital and reserves may underestimate financial strength of these banks. A large portion of Islamic banks’ financial liabilities consists of investment accounts that can be viewed as a form of equity investment (generally based on the principle of Mudarabah). Investment accounts are offered in different forms, often linked to a pre-agreed period of maturity, which may be from one month upwards, and the funds in the accounts can generally be withdrawn if advance notice of one month is given. The profits and returns are distributed between the depositors and the bank, according to a pre-determined ratio, e.g., 80 percent to the depositors and 20 percent to the bank (Iqbal and Mirakhor, 2007). At the extreme, it could be argued that a bank with only restricted investment accounts would be close to a mutual fund in terms of its risk profile, with almost all risk passed to investors. Even with unrestricted investment accounts, much of the risk is in principle borne by investors.
A counterargument against this possible criticism is that even conventional banks usually have the ability to pass on risks to their customers, for example through their ability to adjust (and delay adjustments in) deposit and loan rates. Only after Islamic banks’ layers of protection have been exhausted and after the bank has started to incur losses, does a shock have an impact on capital and reserves. In other words, these additional layers of protection are ultimately reflected in the banks’ returns and capital, and thereby in their z-score. Moreover, the fact that most of the investment accounts can be withdrawn in a relatively short period of time, as well as the fact that the return distribution between the bank and the depositors/investors is pre-determined, diminishes the factual differences in risk profiles associated with the investment accounts, compared with floating-rate deposits and other conventional funding used by commercial banks. So, while the differences between Islamic and conventional banks should be born in mind, capital and reserves are still a reasonable proxy variable to assess the “bottom line” default risk.
As a preliminary step in the analysis, author performs basic statistical tests for the z-scores. Author compares z-scores in Islamic and commercial banks. Because bank size is an important factor in some of the existing papers on bank soundness, author also subdivide banks into large and small Islamic banks and large and small commercial banks (using total assets of US$ 1 billion as the cut-off point between small and large banks), and carry out pair wise comparisons of z-scores for these various subgroups.
3.1 Regression Analysis
The main part of our approach is to test, using regressions of z-scores as a function of a number of variables, whether Islamic banks are less or more stable than commercial banks. Author estimate a general class of panel models of the form
z B I T T I B T M C D , , , , 1 , 1 , 1 , , 1 , 1 , , =α +β +γ +Σδ +Σφ +Σϕ +ϖ +Σλ +Σπ +
where the dependent variable is the z-score i j t z , , for bank i in country j at time t; i, j,t−1 B is a vector of bank-specific variables; j,t−1 I contains time-varying industry-specific variables; s T and s j,t−1 T I are the type of banks and the interaction between the type and some of the industry-specific variables; j t M , , j C , and t D are vectors of macroeconomic variables, country and yearly dummy variables, respectively; finally, i, j,t ε is the residual.
To distinguish the impact of bank type on the z-score, author include a dummy variable that takes the value of 1 if the bank in question is an Islamic bank, and 0 otherwise (i.e., if it is a commercial bank). For example, if Islamic banks are relatively weaker than commercial banks, the dummy variable would have a negative sign in the regression explaining z-scores.
At the systemic (country) level, author wants to examine the Islamic banks’ impact on other banks and the hypothesis that the presence of Islamic banks lowers systemic stability. For this reason, author has calculated the market share of Islamic banks by assets for each year and country and interact it with Islamic and commercial bank dummies. For example, a negative sign for the interaction of the Islamic banks’ market share and the Islamic bank dummy would indicate that a higher share of Islamic banks reduces their soundness (reduces their z-scores).
In addition to the above key variables of interest, the regression includes a number of other control variables, both at the individual bank level and the country level.8 To control for bank-level differences in size, asset composition, and cost efficiency, author include the bank’s asset size in U.S. dollars billion, loans over assets, and the cost-income ratio. Also, to control for differences in the structure of the bank’s income, author calculate a measure of income diversity that follows Laeven and Levine (2005).9 This variable captures the degree to which banks diversify from traditional lending activities (those generating net interest income) to other activities. For Islamic banks, the net interest income is generally defined as the sum of the positive and negative income flows associated with the PLS arrangements (see, e.g., International Monetary Fund, 2004). To further capture differences of Islamic banks in their business orientation, author interact the income diversity variable with the Islamic bank dummy. Controlling for these variables is important because there are differences in these variables between Islamic banks and the other groups.
At the country level, author includes a number of variables that take on the same value for all banks in a given country. In particular, author adjusts for the impact of the macroeconomic cycle by including three macroeconomic variables (GDP growth rate, inflation rate, and exchange rate depreciation). To account for cross-country variation in financial stability caused by differences in market concentration, author includes the Herfindahl index, defined as the sum of squared market shares (in terms of total assets) of all banks in the country. The index can have values from 0 to 10,000 (for a system with only one bank).
Author also account for the impact of governance on stability by using the governance indicator that was compiled by Kaufmann, Kraay, and Mastruzzi (2005). Author average the 6 governance measures of voice and accountability, political stability, government effectiveness, regulatory quality, rule of law, and control of corruption across the available years 2004, 2002, 2000, 1998, and 1996 into a single index per country. The governance indicator captures crosscountry differences in institutional developments that might have an effect on banking risk.
All bank-specific and macroeconomic variables, the Herfindahl index, and the Islamic banks’ market share and its interaction with the Islamic and commercial bank dummies are lagged to capture the possible past effects of these variables on the banks’ individual risk. Author also tests for the robustness of the lagged effects by restricting the explanatory variables to contemporaneous effects.
Author start the regression analysis by the pooled ordinary least squares (OLS) technique. Given that our sample includes outliers, author use a robust estimation technique as an important estimation method. Hamilton (2002) provides a detailed description of the technique. In a nutshell, it assigns, through an iterative process, lower weights to observations with large residuals, making the estimation less sensitive to outliers. Standard errors are calculated using the pseudovalues approach (Street, Carroll, and Ruppert, 1988). To test the sensitivity of the results with respect to the estimation method, author also estimates fixed effects and median least squares regressions. The median least squares regressor minimizes the median square of residuals rather than the average and thus reduces the effect of outliers.
Author also assesses the robustness of the results with respect to the selected sample. To do that, author estimate the same regressions for different bank sizes. Specifically, author estimate the regressions separately for sub-samples of large banks (those with total assets of more than US$1 billion) and small banks (all others).
3.2 Data Analysis
To capture the importance of bank size on stability in Islamic and commercial banks, author present some of the results separately for sub-samples of large banks (assets over US$ 1 billion) and small banks (all others), using the same threshold for both Islamic and commercial banks. The threshold is arbitrary, but it has been used in previous research on small banks (e.g., Mercieca, Schaeck, and Wolfe, 2007), and, more importantly, the main results of our analysis are not sensitive with respect to moderate changes in the threshold. About 49 percent of the Islamic banks and about 62 percent of the commercial banks fall into the large bank category.
Several general issues relating to the BankScope data need to be mentioned. First, to be able to analyze Islamic banks’ impact on systemic stability, author has focused on countries where Islamic banks have a higher than negligible share of the financial system. El Qorchi (2005) notes that Islamic institutions operate in 75 countries, yet in most of those countries, Islamic banks have a very small market share. Author has included all the systems where Islamic banks according to the BankScope data accounted for more than 1 percent of the total assets in at least one year in the period under observation (1993-2005). The exclusion of the Islamic banks from the other countries does not appear material, since our sample still has a good worldwide coverage of Islamic banks. This is confirmed by the fact that Islamic banks covered in our starting sample have total assets of US$253 billion as of 2004, which is in line with the estimate of “about US$250 billion” worldwide assets of Islamic banks quoted for example by El Qorchi, 2005.
Second, our empirical analysis relies to a large extent on unconsolidated bank statements. Ideally, author would have opted for using only consolidated statements for all financial institutions. However, only about 1/3 of the relevant observations in BankScope are based on consolidated data; the rest are unconsolidated data. This scarcity of consolidated data limits their usefulness for econometric analysis. Author therefore use consolidated data when available, but when consolidated data are not available for a bank, author use unconsolidated data instead.
Third, BankScope, while being the most comprehensive commercially available database of banking sector data, is not exhaustive. Coverage varies from country to country; for most countries in author’s sample, the BankScope data cover 80-90 percent of the banking systems in terms of total assets. Moreover, author had to exclude 2 countries from our analysis because of data problems, bringing the number of countries on which the aggregate results are based from 20 to 18. For Lebanon and Kuwait, BankScope does not have unconsolidated observations for Islamic banks, so these countries are excluded from the regression analysis. The coverage of this paper, while less than 100 percent, is still higher than that for most banking studies (and in particular studies that focus on banks with particular features, such as large banks or banks that are listed on the stock market). Even after the exclusions, total assets of the Islamic banks included in the panel are about the same as the estimated total assets of Islamic banks in the world reported in earlier literature (see above). Author’s sample should therefore be large enough to provide reliable inferences.
Fourth, author largely relies on BankScope for data quality. There are a number of important issues relating to definitions of financial indicators for Islamic banks, for example what to include in capital, or how to measure (the equivalent of) interest income. The issue of financial soundness indicators in Islamic banks is discussed in more detail for instance in International Monetary Fund (2004). For the purposes of this paper, author have largely relied on BankScope’s definitions of the key variables, even though author have done basic crosschecking and also excluded outliers, some of which may be the result of deviations from common definitions.
Fifth, some commercial banks (including several major global players) have opened dedicated Islamic windows or Islamic branches conducting business according to Islamic banking principles. However, the available financial data do not allow author to distinguish the financial performance (and importance) of these windows or branches and analyze their separate impact on financial stability. Author therefore focuses only on the comparison of fully-fledged Islamic banks and commercial banks.
Sixth, data limitations prevent author from taking fully into account all aspects of Islamic financial contracts, for example, by controlling for type of Islamic instruments, distinguishing between PLS and other investments, distinguishing the different types of investment accounts, and return equalization funds. In addition to the bank-by-bank data, author also uses a number of macroeconomic and other system-wide indicators. Those are described in more detail in Appendix II.
