Predicting private company failure: A multi-class analysis

Abstract

This study utilizes an advanced machine learning method known as TreeNet® (Salford Systems, 2017) to predict a variety of private company failure states, ranging from binary settings (i.e. failed vs non-failed) to more complex multi-class settings with up to five states of failure. Based on a large global sample, TreeNet® proved to be a significantly better predictor of private company failure than conventional models such as logistic regression. While the out-of-sample predictive performance of TreeNet® is best in binary settings, the model also produces strong area under the ROC curve (AUC) results for the multi-class models. We also find that the predictive performance of financial variables is significantly enhanced when combined with external risk factors such as macro-economic variables and other non-financial measures. The results of this study have several implications for the private company failure literature and the usefulness of machine learning methods in accounting and finance more generally.