A Multinomial Approach to Early Warning Systems for Debt Crises
38 Pages Posted: 10 Jul 2006
Date Written: May 2006
This paper develops an early warning system for sovereign debt crises, broadly defined as episodes of outright default, failure of a country to be current on external obligations and substantial access to IMF resources. It estimates a multinomial logit model that makes it possible to differentiate between three regimes labelled 'tranquil', 'pre-crisis' and 'adjustment'. The model includes a large set of macroeconomic variables and is able to predict, in-sample, 78 percent of onsets of crisis while sending false alarms in 34 percent of tranquil cases; its out-of-sample performance is very similar, with 70 percent of entries into crisis correctly predicted and 20 percent of tranquil cases triggering false alarms.
Keywords: emerging markets, early warning systems, debt crises, default
JEL Classification: H63, E66
Suggested Citation: Suggested Citation