A comparison of Bayesian, Hazard, and Mixed Logit model of bankruptcy prediction

Abstract

The purpose of this study is to examine the impact of the choice of cut-off points, sampling procedures, and business cycles on the forecasting accuracy of bankruptcy prediction models. A misclassification can result in an erroneous prediction resulting in prohibitive costs to firms, investors, and the economy. A salient feature of our study is that our analysis includes both parametric and nonparametric bankruptcy prediction models. A sample of firms from the Bankruptcy Research Database in the U.S. is used to evaluate the relative performance of the three most commonly used bankruptcy prediction models: Bayesian, Hazard, and Mixed Logit. Our results indicate that the choice of the cut-off point and sampling procedures affect the rankings of the three models. We show that the empirical cut-off point estimated from the training sample result in the lowest misclassification costs for all three models. When tests are conducted using randomly selected samples, and all specifications of type I costs over type II costs are taken into account, the Mixed Logit model performs slightly better than the Bayesian model and much better than the Hazard model. However, when tests are conducted across business-cycle samples, the Bayesian model has the best performance and much better predictive power in recent business cycles. This study extends recent research comparing the performance of bankruptcy prediction models by identifying under what conditions a model performs better. It also allays the concerns for a range of users groups, including auditors, shareholders, employees, suppliers, rating agencies, and creditors’ with respect to assessing corporate failure risk.