Extreme Bounds Analysis in Early Warning Systems for Currency Crises

Selecting early warning indicators to predict currency crises is not straightforward, because there are several mechanisms that lead up to currency crises and these mechanisms change over time and are not always visible, obvious or linear in nature. Traditionally, early warning indicators are selected from theory, stylized facts or meta studies. The alternative is to use data-driven processes, such as the Extreme Bounds Analysis (EBA) that can handle unbalanced panels better than other methods. We use the selection of indicators determined by the EBA in an Early Warning System (EWS) and compare the performance with a benchmark EWS, where the indicators are chosen from a meta study of the literature. We find for a sample of 33 emerging economies that the determinants of currency crises confirm a large number of the early warning indicators from the literature, but also exhibit differences, which provide more insight in crisis mechanisms. Applying the determinants in an Early Warning System, we find that the EBA-selection outperforms the benchmark in the in-sample period (1990-2010). In the out-of-sample period (2011-2018) the EBA-selection performs better in the majority of the specifications.