Modeling and Estimating the Credit Cycle by a Probit-AR(1)-Process

Abstract

The loss distribution of a credit portfolio is considered within the framework of a Bernoulli-mixture model where in each rating grade the stochastic Bernoulli-parameter follows an autoregressive stationary process. Changes in the loss distribution are discussed when the unconditional view is replaced by a conditional view where information from the last period is taken into account. This relates to the lively debate among practitioners whether regulatory capital should incorporate point-in-time or through-the-cycle aspects. Calculations are carried out in a model estimated with real data from a large retail portfolio.