Approaches to Default Probability Estimation of Credit Rating Agencies’ Rating Scales

Abstract

Under limited data, the classical cohort method for the creation of migration matrices does not fully reflect the dynamics of the credit quality of the objects within the sample. This problem is exacerbated for objects of lower credit quality less represented in the sample. This paper investigates a continuous time approach to the creation of migration matrices. A continuous time migration matrix considers migrations between the credit quality of objects on a given horizon on a daily basis, and thus not only the initial state of the default object, but also its movement between credit quality categories up to the moment of default. We demonstrate that the classical cohort method is inferior to the continuous time method both on simulated data and in the analysis of the real migration statistics of the credit ratings of Russian companies. The cohort method overestimates the probability of default across the entire credit rating scale. The continuous time method consistently surpasses the cohort method in accuracy and efficiency starting from the second year of observations and allows the mitigation of the problem of data scarcity.