Comparison of Binary Logit Model and Multinomial Logit Model in Predicting Corporate Failure

A critical issue in the prediction of corporate failures is, whether to categorize sample firms in a binary fashion into failed firms and non-failed firms or to classify failed firms according to multiple financial difficulties. As most previous studies only employ the binary approach in their forecast, this work compares both the binary logit model and the multinomial logit model to determine whether or not the accuracy of forecasting corporate failures can be improved by further classifying financially-failed firms. The binary logit model recognizes slightly-distressed events and bankruptcy-and-reorganizations events both as corporate failure, while the multinomial logit model distinguishes between levels of corporate failure events as slightly-distressed firms and bankruptcy-and-reorganization firms. The empirical results show that the misclassification errors and error costs of the binary logit model are smaller than those of the multinomial logit model, suggesting that the binary logit model performs superior to the multinomial logit model in predicting corporate failure. The comparison results imply that the slightly-distressed firms and bankruptcy-and-reorganization firms are similar in characteristics. The occurrence of slightly-distressed events is already on the verge of bankruptcy, signifying major financial failure in the company operations. In such case, investors and debtors should be especially alert to withdraw their investments or terminate their loans to prevent loss.