A method for reducing credit scores’ sensitivity to economic conditions

Abstract

The cyclicality of credit risk capital requirements has been a matter of concern for banking regulators, supervisors and the industry for years. The sensitivity to economic conditions of the probability of default (PD) grades to which credit exposures are assigned is often one of the most relevant sources of such cyclicality. Moreover, it is often assumed that a grade assignment method with a high differentiation capacity inherently leads to a high sensitivity to economic conditions. In order to challenge this assumption and foster further research – but with no intention of setting any expectation or recommendation for financial institutions – this article explores a methodology aimed at limiting the sensitivity to economic conditions of a pre-existing score while maintaining its differentiation ability, by adding a module to it. This module subtracts an amount which reflects the estimated effect of economic conditions. This allows the original and the adjusted scores to coexist and be used for different purposes. After testing that the methodology works on a synthetic dataset, its effectiveness is confirmed on a real dataset obtained from Banco de España internal sources. The results indicate a significant reduction in the variability of PD and risk weights when comparing a PD calibration of the original score with a PD calibration of the adjusted score.