Rethinking Operational Risk in the AI Era: A Comparative Perspective

Abstract

Operational risk has become increasingly important due to recurrent financial scandals and significant loss events, yet it remains comparatively underexplored in the literature. At the same time, artificial intelligence is gaining prominence as a transformative tool in risk management. This paper compares traditional parametric and machine-learning approaches for operational loss severity modeling using the SAS OpRisk database. The analysis combines descriptive examination of loss characteristics, parametric fitting with the Lognormal and Gamma distributions, and supervised prediction based on structured explanatory variables. The predictive framework includes Ridge regression, Decision Tree, Random Forest, Gradient Boosting, and a hybrid Ridge–Random Forest model. The findings suggest that operational loss severity reflects both linear and non-linear structures and that hybrid modeling can improve predictive performance. The research highlights the complementary roles of traditional parametric fitting and machine-learning methods in operational risk severity analysis