Implications of macroeconomic conditions on Romanian portfolio credit risk. A cost-sensitive ensemble learning methods comparison

Abstract

Credit risk assessment represents a key instrument in the decision-making of the banking and financial institutions. In this article, we present a framework for credit risk strategies to improve portfolio efficiency under a change of macroeconomic regime. The aim is to compare the accuracy of several ensemble methods (AdaBoost, Logit Boost, Gentle Boost and Random Forest) on a default retail Romanian loan portfolio under different risk adversity scenarios, a priori and posteriori the financial distress. Using cost-sensitive ensemble learning models, we concluded that regime-based credit strategy can offer a better alternative in both terms of loss allocated to the strategy as well as defaulters’ classification accuracy.