Corporate default forecasting with machine learning

• Published in: Expert systems with applications 2020-12, Vol.161, p.113567, Article 113567
• Main Authors: Moscatelli, Mirko, Parlapiano, Fabio, Narizzano, Simone, Viggiano, Gianluca
• Format: Article
• Language: English
• Subjects: Credit scoring; Default; Gradient boosting machine; Machine learning; Random forest; Regression analysis; Statistical analysis; Statistical models
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2020.113567

•Machine learning models entail gains in accuracy relative to statistical models.•Due to non-linear relationship between some of the variables and the default outcome.•The gains are greater for clusters of borrowers more difficult to predict.•The gains decrease when high quality information is used in the training dataset.•Credit allocation based on ML models would imply lower default rates. We analyze the performance of a set of machine learning models in predicting default risk, using standard statistical models, such as the logistic regression, as a benchmark. When only a limited information set is available, for example in the case of an external assessment of credit risk, we find that machine learning models provide substantial gains in discriminatory power and precision, relative to statistical models. This advantage diminishes when confidential information, such as credit behavioral indicators, is also available, and it becomes negligible when the dataset is small. Moreover, we evaluate the consequences of using a credit allocation rule based on machine learning ratings on the overall supply of credit and the number of borrowers gaining access to credit. Machine learning models concentrate a greater extent of credit towards safer and larger borrowers, which would result in lower credit losses for their lenders.