Predicting Currency Crises With a Nested Logit Model (Revised July 2002)

This paper provides a unique methodology - a multi-state nested logit model - to evaluate the currency crisis probabilities and the probabilities of successful defense by the central bank given speculative attacks. This model provides better forecasts compared to the benchmark signal indicator approach and it provides important policy implications: it sheds light on the relative importance of international illiquidity, financial fragility and excessive fiscal deficits in triggering speculative attacks. The findings in this paper also provide empirical evidence on the importance of the "overborrowing syndrome" in triggering speculative attacks. Thus a policy advice to central banks is to strengthen the supervision in the banking sector to reduce the currency maturity mismatch in the assets and liabilities side.