Aggregate Bankruptcy Rates and the Macroeconomic Environment. Forecasting Systematic Probabilities of Default

Recent empirical research has stressed the importance of economy wide factors in the assessment of default risk, for instance for bond portfolios or portfolios of bank loans. Macroeconomic risk is essentially systematic, as it is difficult to reduce through diversification. Adequate forecasts of the links between macroeconomic factors and default risk, often defined as aggregate credit risk, are therefore important in a large number of risk management applications. This paper considers eleven alternative ways to model the aggregate bankruptcy rate, which is a proxy for aggregate credit risk, on Belgian data for the period 1986- 2002. Based on these models, forecasts of the bankruptcy rates for 2003-2006 are made and compared. Four or five-variable Almon lag models result in the best in-sample fit and have the best forecast performance, but much more straightforward alternatives – such as an ARMA model or a one-variable model based on real GDP growth – do almost equally well. Four or five-variable vector error correction models are shown to have good in-sample fit, but very poor out-of-sample forecasting power. Overall we find that the prediction power of simple models is hard to beat.