Mobile Edition
Print
Get the Mag
Weekly UpdatesRATING = [[gamma].sub.0] + [[gamma].sub.1]AUDITOR + [[gamma].sub.i][X.sub.i] + [[epsilon].sub.i] (6)
Expanding equation (6) to include variables for bank characteristics and performance measures provides the following testable regression equation:
RATING = [[gamma].sub.0] + [[gamma].sub.1]AUDITOR + [[gamma].sub.2]ROE + [[gamma].sub.3]lnASSETS + [[gamma].sub.4]RBA/TA + [[gamma].sub.5]BRANCHES + [[epsilon].sub.i] (7)
Variables
The dependent variable, RATING, is the 1996 peer or national Sheshunoff rating. This is a percentile ranking (99=best, 0=worst), within the bank's peer group or on a national basis, based on a weighted composite score which represents four of the five CAMEL factors (C, A, E, and L).
AUDITOR, a proxy for audit quality, is a categorical classification based on auditor size (Big 5 versus non-Big 5). As discussed previously, DeAngelo (1981) argues that the size attribute of large auditors allows them to be independent and skillful, which ultimately lets them provide high quality audits. Big 5 auditors are the largest five auditors measured by total fee revenue or assets audited. If regulators perceive the auditing work of Big 5 auditors to be of higher quality, they will place greater weight on the financial disclosures. Given the incentives of bank managers to choose a high quality auditor and the incentives of regulators to rely on the auditor's work (discussed previously), we expect the sign on the AUDITOR variable to be positive. AUDITOR is the variable of interest.
The remaining variables are controls. ROE (Return On Equity) controls for bank performance, which is likely to be an important determinant for all CAMEL factors, but certainly for the E rating since current earnings are added to equity at the end of the operating period. It could be that Big 5 and Non-Big 5 clients differ because of differences in the underlying economics of the bank. If these differences are important for regulator's evaluations, a difference in regulatory evaluations may result even if auditor class is not relevant. Adjusting for these underlying economic factors in the regression is important in order to measure the incremental effect of auditor class on regulatory evaluation. We expect regulators to take a favorable view of higher levels of bank performance. Thus, ROE should be positively related to the Sheshunoff rankings.
Previous research (Atiase, 1985 and El-Gazzar, 1998) shows that size is a proxy for the strength of public monitoring (also referred to as the disclosure environment). Larger firms are more closely monitored because of larger investor and analyst following. The monitoring provided by these external agents may be correlated with auditor monitoring and should be controlled. Also, capital assets are specifically used in calculating the CAMEL rating. To determine whether auditor class has incremental explanatory power over assets, we include the natural log of total bank assets (lnASSETS) as a size proxy. The natural log is used to correct for heteroscedasticity commonly found in the distribution of firm assets. We expect it to be positively related to the Sheshunoff rating.
Banks explicitly provide direct measures of risk for their assets. Risk should be controlled since it is likely to be a determinant of the liquidity of a bank's portfolio, which is one of the CAMEL rating factors, and has been theoretically and empirically documented as a factor explaining auditor choice (Titman and Trueman, 1986; Datar et al., 1991). The Basle Agreement (accepted by the Federal Reserve in 1992) provides a methodology for computing risk-weighted assets. This methodology focuses only on credit risk in calculating risk-based assets. Other types of exposure such as interest rate, liquidity, and funding risks, as well as asset quality problems, are not factored into the risk-based calculation. Risk-based asset reporting is a mechanism that weights the relative value of assets held in a bank's portfolio by the risk level associated with these assets. A basic premise in the Basle Agreement is that the riskier a bank's assets, whether they are on or off the balance sheet, the more capital is requir ed to support them.
According to the Basle Agreement, risk-based assets are determined by assigning the bank's assets to one of four risk categories. Category 1 includes cash and cash equivalents, category 2 includes short-term claims maturing in one year or less, category 3 includes family residential mortgages or public sector bonds, and category 4 includes commercial and consumer loans. We treat reported risk-based assets as a direct measure of the bank's riskiness. We scale riskbased assets by total assets (RBA/TA) to get a percentage measure of assets discounted due to loan uncertainty. Thus, lower percentages of risk-based assets (lower percentages implies greater discounts) would represent greater levels of risk and would adversely affect regulator's evaluation of bank financial condition. We expect the variable RBA/TA to be negatively related to higher Sheshunoff RATINGs.
Finally, we use the number of bank BRANCHES to measure the complexity of bank operations. Firm complexity has been shown in previous studies to be correlated with auditor class (DeAngelo, 1981; Simunic, 1980; Craswell et al., 1995). Geographically dispersed bank branch offices make it more difficult for regulators to assess the safety and soundness of the bank's assets. If bank complexity reduces the regulator's inclination to assign favorable CAMEL ratings, then we expect the number of bank BRANCHES to be negatively related to the Sheshunoff RATING.
RESULTS
Descriptive Statistics
Table 1 reports means and standard errors for the sample partitioned by auditor type (Big 5 versus Non-Big 5). The sample is composed of 199 firms choosing a Big 5 auditor and 53 firms choosing a Non-Big 5 auditor. The sample observations are a random selection from the Sheshunoff database, and include both publicly-held and nonpublicly-held banks.
The mean Sheshunoff RATING is higher for the Non-Big 5 group than the Big 5 group. The difference, however, is marginally significant for the national measure only (p-value = .093). While these comparisons are not consistent with our argument, they are univariate comparisons only, which may not measure the incremental effects of one factor if the effects of other factors are not controlled. The economic effects of bank condition could easily swamp the effects of audit quality. Thus, controlling for bank economic condition is critical before making inferences about the effects of auditor choice on regulatory evaluation. Also, a univariate analysis is not sophisticated enough to capture the two-step effect developed above in equations (1) through (6). Consequently, we place no confidence in the univariate tests and provide a multivariate analysis in the next section based on our system of equations.
Although we make no predictions about the CAMEL factor inputs to the Sheshunoff rating, it is interesting to note that the mean capital adequacy ratio is statistically higher for the NonBig 5 group than the Big 5 group (.0978 versus .0884, respectively; p-value = .003 for the difference). This difference may reflect size effects. Differences in the other CAMEL factor inputs (asset quality, earnings quality, and liquidity ratio) are not statistically significant across auditor type.
Total assets (ASSETS) and the number of branches (BRANCHES) are statistically higher for the Big 5 group than the Non-Big 5 group, which is consistent with previous research that finds Big 5 auditors associated with bigger and more complex clients (Palmrose, 1986; Simunic, 1980). The proportion of assets discounted due to risk (risk-based assets as a percentage of total assets) is significantly greater for the Non-Big 5 group, suggesting that they carry more credit risk than the Big 5 group. This is consistent with previous research that finds higher client-specific risk is associated with low quality auditors (Titman and Trueman, 1986).
We perform a Pearson correlation to assess the potential for multicollinearity among the independent variables (results not presented). With the exception of the Rho coefficient for BRANCHES and InASSETS, none of the correlations among the variables which will be used as independent variables in the multivariate regression is greater than .36. The correlation between BRANCHES and InASSETS is .66; however, variance inflation factors for all of the independent variables are less than 1.4, indicating that harmful collinearity probably does not exist.
In summary, the univariate statistics suggest that auditor choice is significantly associated with differences in regulatory assessments. Although these univariate comparisons are not consistent with our predictions for audit quality, a multivariate analysis that controls for other effects on the dependent variable may provide different results. Two of the four control variables are significantly associated with the dependent variable, national RATING, and three of the four control variables are significantly associated with the dependent variable, peer group RATING.
Multivariate Results
Coefficient estimates from multivariate regressions are presented in Tables 2 and 3. In Table 2, the dependent variable RATING is defined as the bank's percentile ranking among all national banks as determined by the Sheshunoff rating service. The overall model is significant (zero slopes F-statistic = 24.52, p-value [less than].01) and has significant explanatory power with an adjusted [R.sup.2] of 31.9%. Coefficients on control variables for bank characteristics, InASSETS for bank size and BRANCHES for bank complexity, are not significant. However, coefficients on control variables representing performance, ROE for profitability and RBA/TA for asset quality, are significant at traditional levels in the expected direction. ROE and RBA/TA have positive signs, indicating that higher returns and less risky loan portfolios have a favorable effect on regulator's evaluations.
FEED