Developing Default Prediction Models Using Machine Learning: An Empirical Study of Vietnamese Enterprises

Abstract

Forecasting bankruptcy risk is a critical task for financial institutions in the realm of credit risk management. Identifying bankruptcy risk allows creditors to screen loan applicants, estimate interest rates, establish lending conditions, and manage credit portfolios. Investors also use bankruptcy probabilities to monitor the credit quality of bonds, for pricing, and to set up investment portfolios. In this study, the authors employed machine learning models based on decision tree algorithms to predict the bankruptcy risk of companies in Vietnam from 2009 to 2020. The results indicate that the Random Forest and Gradient Boosting models significantly outperform the Logistic regression model across all evaluation metrics, such as Confusion Matrix AUC, Accuracy ratio, Precision ratio, Recall ratio, and F1 Score. Among them, the Random Forest model tends to perform better than the Gradient Boosting model on these evaluation criteria. Moreover, the results of the model also suggest important predictive variables in constructing a bankruptcy risk prediction model.