Optimizing Credit Gaps for Predicting Financial Crises: Modelling Choices and Tradeoffs

Abstract

Credit gaps are good predictors for financial crises, and banking regulators recommend using them to inform countercyclical capital buffers for banks. Researchers typically create credit gap measures using trend-cycle decomposition methods, which require many modelling choices, such as the method used, and the smoothness of the underlying trend. Other choices hinge on the tradeoffs implicit in how gaps are used as early warning indicators (EWIs) for predicting crises, such as the preference over false positives and false negatives. We evaluate how the performance of credit-gap-based EWIs for predicting crises is influenced by these modelling choices. For the most common trend-cycle decomposition methods used to recover credit gaps, we find that optimally smoothing the trend enhances out-of-sample prediction. We also show that out-of sample performance improves further when we consider a preference for robustness of the credit gap estimates to the arrival of new information, which is important as any EWI should work in real-time. We offer several practical implications.