Ensemble machine learning models in financial distress prediction : evidence from China

Abstract

Corporate distress signals are important for both institutions and banks when evaluating firms’ performances. This paper evaluates five different models in predicting the distress for listed companies in China based on 22 dimensions of financial data from 2014 to 2022. The models include three ensemble machine learning models: Adaboost, Bagging, and Random Forest, as well as a single machine learning model Decision Tree, along with a benchmark Logistic Regression. The comparative analysis found Random Forest to be the most promising method with the highest accuracy ratio and lowest Type I and Type II errors. This paper concludes that ensemble learning models could be an easy-to-replicate and highly efficient tool for institutions and banks to evaluate and predict potential distress in firms.