Forecasting Banking Crises with Dynamic Panel Probit Models

Abstract

Banking crises are rare events, however when they occur they often have dramatic consequences. The aim of this paper is to contribute to the toolkit of early warning models available to policy makers by exploring the dynamics and non-linearities embedded in a panel dataset covering several countries over three decades. The in-sample and out-of-sample forecast performance of several dynamic probit models is evaluated, with the objective of developing a common vulnerability indicator with early warning properties. The results obtained evidence that adding dynamic and non-linear components to the models substantially improves the ability to forecast banking crises.

Keywords: banking crisis, binary data, dynamic probit models, early warning models.

JEL Classification: C12, C22