Abstract
This study provides new insights into banks' credit risk models by exploring features of their credit risk estimates and assessing practicalities of transition matrix estimation and related assumptions. Using a unique dataset of internal credit risk estimates from twelve global A-IRB banks, covering monthly observations on 20,000 North American and EU large corporates over the 2015-2018 time period, the study empirically tests the widely used assumptions of the Markovian property and time homogeneity at a larger scale than previously documented in the literature.
The results show that internal credit risk estimates do not satisfy these assumptions as they show evidence of both path-dependency and time heterogeneity. In addition, contradicting previous findings on credit rating agency data, banks tend to revert their rating actions.