Bank failure prediction in the COVID-19 environment

Abstract

The paper delivers a multi­state, continuous, non­homogeneous Markov chain to present a COVID­19 stressed probability of default (PD) model for banks. First it analyzes the theoretical and methodological considerations of bank failure. Then it provides a comprehensive review of earlier empirical bank failure models published in literature. It makes the case for a multi­state model design, which has numerous advantages over the conventional binary classification techniques. A formal description of Markov chain modeling is followed by the detailed presentation of empirical model development. Eventually it estimates PDs for a five­year forecast horizon with the developed model reflecting COVID­19 crisis impacts.