Machine Learning for IRBA Rating Models

What role can machine learning play in the development and calibration of IRBA rating models? The article provides a practice-oriented overview of current ML methods — from gradient boosting through random forests to neural networks — and evaluates their suitability for use in regulated banking processes.

In addition to pure predictive performance, aspects of model validation, regulatory acceptance and the explainability of ML decisions are also examined. The article is aimed at professionals in credit risk management who are evaluating the use of modern methods in their IRBA models.