| This study evaluates the performance of auditors' opinions in predicting bank failures, as defined as the closing of the bank by regulators. The study uses an indicator variable for a modified audit opinion defined as (a) a modified opinion for going-concern issues, (b) a modified opinion for uncertainty due to regulatory matters, (c) a qualified opinion because of departure from generally accepted accounting principles, (d) adverse opinion because of departure from generally accepted accounting principles, or (e) a disclaimer. Using a logistic regression model employing financial ratios proxying for the components of the regulators' CAMEL rating and an indicator variable for audit opinion modification, the additional explanatory power (i.e., the information content) of the auditor's opinion and the predictive ability of the model was investigated. In addition, a cross-validation of the model was provided by test of the Thomson (1991) model, with and without the indicator variable for the audit opinion modification. The study compares the predictive ability of the models tested across varying cost ratios of type I to type II errors.; The descriptive statistics of the sample shows significant differences between failed and nonfailed banks. Failed banks were significantly different in receiving a modified audit opinion, using local CPA firms, being independent single banks, being smaller, and having higher loan concentrations, lower earnings ability and capital than nonfailed banks. In testing the logistic regression and the Thomson (1991) models, the coefficients of the audit opinion variable were not significant for the year preceding failure. However, the models with the indicator variable for modified audit opinion had the highest likelihood ratio index, which is analogous to R{dollar}sp2{dollar} and can be described as the amount of uncertainty explained by the model. The chi-square tests for these models indicate that the modified audit opinion variable has significant explanatory power and has incremental explanatory power beyond the set of financial ratio variables.; The classification accuracy, predictive ability and misclassification costs (costs of type I and type II errors) of the models were tested using the McNemar Change Test. Overall, the models with the audit opinion variable had the lowest cost and the best classification accuracy. The audit opinion models were significantly and consistently less costly than the naive, or benchmark model. |
