This paper develops a logistic regression model in an in-house credit assessmentsystem (ICAS) framework for predicting corporate defaults in the Greekeconomy. We consider the impact of the COVID-19 pandemic and the associatedgovernment financial support schemes, aiming to protect against financialvulnerabilities, on the probability of default of non-financial firms, as well asthe relevant sectoral and firm-size effects. In developing the ICAS framework,we address methodological issues such as the predictive performance of statisticalversus machine learning approaches and the imbalanced dataset problem,indicating ways to evaluate such models with strong predictive power. Ourfindings suggest that the effect of the financial support measures dominatesthe pandemic shocks, thus substantially reducing the probability of firms’default, while the size- and industry-based models show that firms in themicro and services sectors benefited the most. Furthermore, using a randomforest model, our findings highlight the trade-off between the transparency oftraditional statistical models and the predictive value of machine learningmodels.
