Did established Early Warning Signals predict the 2008 crises?

Over the past 60 years, a voluminous literature has painstakingly developed crises theories and their associated Early Warning Signals. The hallmark of this literature is the consistency with which selected Early Warning Signals, such as the level of reserves and exchange rate appreciations, are presumed to predict different types of crises across countries and time. The diversity of crises theories that motivate Early Warning Signals presents, however, a challenge to empirical implementations. Given that the true model of Early Warning Signals is unknown, omitted variable bias contaminates estimates and confidence levels when the uncertainty surrounding a particular theory has been ignored. After addressing model uncertainty in Early Warning Signal regressions, using an extended version of Frankel and Saravelos (2012. J. Int. Econ. 87, 216–231) dataset, we do not find a single Early Warning Signal that alerts to all dimensions of the 2008 crisis. Instead, distinct sets of Early Warning Signals identify different dimensions of the crisis: banking, balance of payments, exchange rate pressure, and recession.