Can Corporate Governance Variables Enhance the Prediction Power of Accounting-Based Financial Distress Prediction Models?

We integrated accounting, corporate governance, and macroeconomic variables to build up a binary logistic regression model for the prediction of financially distressed firms. Debt ratio and ROA are found to be the most explanatory accounting variables while the percentage of directors controlled by the largest shareholder (which measures negative entrenchment effect), management participation, and the percentage of shares pledged for loans by large shareholders are shown to have positive contribution to the probability of financial distress. For macroeconomic sensitivities, firms with higher sensitivities to the annualized growth rates of manufacturing production index and money supply (M2) are more vulnerable to financial distress. As to the issue of sampling technique, we find that oversampling of distressed firms is subject to the problem of choice-based sample bias pointed out by Zmijewski (1984). The classification accuracy is overstated consequently. We try to include as many healthy firms as possible in our sample instead of following the traditional 1: 1 or 1: 2 matching principle. The results show that the classification accuracy is mostly significantly improved in our integrated prediction model when the sample is closest to the actual population. For the trade-off between type I and type II errors in the predicted probability classification, we maximize the sum of classification accuracy for both groups of firms (the healthy and the distressed). It is found that an estimated probability of financial distress of 0.2000 represents the optimal cutoff point for predicting financial distress. Under such a cutoff scheme, our integrated model produces an in-sample classification accuracy of 80.7% for distressed firms and 93.2% for healthy firms. For out-sample prediction, 90% of the distressed firms and 85.4% healthy firms in 2001 are correctly identified using an integrated model built upon samples from 1998 to 2000.