AUTHOR=Nagl Matthias , Nagl Maximilian , Rösch Daniel 

TITLE=Quantifying uncertainty of machine learning methods for loss given default

JOURNAL=Frontiers in Applied Mathematics and Statistics

VOLUME=Volume 8 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/applied-mathematics-and-statistics/articles/10.3389/fams.2022.1076083

DOI=10.3389/fams.2022.1076083

ISSN=2297-4687

ABSTRACT=Machine learning has increasingly found its way into the credit risk literature. The approaches mainly focus on better forecasting credit risk parameters and are often shown to be superior to standard statistical models. The quantification of the prediction uncertainty is typically not analyzed in the machine learning credit risk setting.  However, this is imminent for risk managers and regulators likewise as its quantification increases the transparency and stability of machine learning methods in risk management and reporting tasks. We fill this gap by applying the novel approach of deep evidential regression to loss given defaults (LGDs). We evaluate aleatoric and epistemic uncertainty for LGD estimation techniques and apply explainable artificial intelligence (XAI) methods to evaluate main drivers. We discover that aleatoric uncertainty is considerably larger than epistemic uncertainty. Therefore, the majority of uncertainty in LGD estimates appears to be irreducible as it stems from the data itself.