A Two-Country Study of Default Risk Prediction Using Bayesian Machine-Learning

This paper tackles the prediction problem of firm default based on financial accounts and other firm features. We propose to exploit a novel machine-learning algorithm, the Bayesian Additive Regression Tree with Missingness Incorporated in Attributes (BART-MIA), which has been recently shown to outperform many traditional algorithms in analogous prediction tasks. We address the issue from an international perspective to assess its performance in both Netherlands and Italy over recent years. Despite the structural differences in the financial accounts in the two countries, we find the BART-MIA can take advantage of countrylevel missingness patterns and outperform state-of-the-art econometric and machine-learning models