A comparative analysis of machine learning models for corporate default forecasting

• Autor(a) principal: Seum, Alexander Michael
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.14/41738
• topic: Credit risk; Default forecasting; Machine learning; Random forest; Domínio/Área Científica::Ciências Sociais::Economia e Gestão

This study examines the potential benefits of utilizing machine learning models for default forecasting by comparing the discriminatory power of the random forest and XGBoost models with traditional statistical models. The results of the evaluation with out-of-time predictions show that the machine learning models exhibit a higher discriminatory power compared to the traditional models. The reduction in the sample size of the training dataset leads to a decrease in predictive power of the machine learning models, reducing the difference in performance between the two model types. While modifications in model dimensionality have a limited impact on the discriminatory power of the statistical models, the predictive power of machine learning models increases with the addition of further predictors. When employing a clustering approach, both traditional and machine learning models exhibit an improvement in discriminatory power in the small, medium, and large firm size clusters compared to the previous non-clustering specifications. Machine learning models exhibit a significantly higher ability to classify micro firms. The findings of this research indicate that the machine learning models exhibit superior discriminatory power compared to the traditional models across the different specifications. Machine learning models can be used to forecast the potential impact of corporate default of non-financial micro cooperations on the Portuguese labour market by estimating the number of jobs at risk.