Multiyear Risk of Credit Losses in SME Portfolios

Abstract

We model multiyear loss distributions based on credit scores and macroeconomic risk drivers. In a two-step approach, we first model future default probabilities as functions of these risk factors and, second, model processes for the risk factors themselves. As an essential extension to one-year forecasts – used e.g. for calculating a bank’s regulatory capital charges under Basel II – we explicitly consider forecasting errors. These errors are introduced by forecasting future paths of the risk factors. We distinguish between idiosyncratic and systematic forecasting errors and show their effects on individual default probability and portfolio loss distribution forecasts using data provided by Deutsche Bundesbank. It turns out that default probability forecasts are much noisier than portfolio risk forecasts due to the possibility of diversification of idiosyncratic forecasting errors in the latter.