A three-state early warning system for the European Union

Abstract

The global financial crisis of 2007-2008 focused the attention of financial authorities on developing methods to forecast and avoid future financial crises of similar magnitude. We contribute to the literature on crisis prediction in several important ways. First, we develop an early warning system (EWS) that provides 7-12 quarters advance warning with high accuracy in out-of-sample testing. Second, the EWS applies region-wide to the leading economies in the European Union. Third, the methodology is transparent – utilizing only publicly available macro-level data and standard statistical classification methodology (multinomial logistic regression, discriminant analysis, and neural networks). Fourth, we employ two relatively novel methodological innovations in EWS modeling: ternary state classification to guarantee a minimum advance warning period, and a fitting and evaluation criterion (the total harmonic mean) that prioritizes avoiding classification errors for the relatively infrequent events of most interest. As a consequence, a policymaker who uses these methods will enjoy a high probability that future crises will be signaled well in advance and that warnings of crisis will not be false alarms.