Abstract. The study aims at developing an Early Warning System for predicting balance of payments crises for 17 emerging economies, which constitute a relatively homogenous group, over the period 1975-2012. We construct an index of exchange market pressure, based on monthly depreciations of the nominal exchange rate and declines in reserves, to identify crisis episodes. To construct the index we propose a new weighting scheme using principal components analysis, as an improvement over the conventionally used precisionweighting scheme. Probit regressions are used to identify key macroeconomic indicator variables that can predict the onset of a crisis. These include the ratio of M2 to reserves, short-term debt to reserves, export growth, ratio of total reserves to external debt, change in reserves, openness and overvaluation of the real exchange rate. From alternative specifications, we identify the best model based on various accuracy measures. We use criteria such as area under the Receiver Operating Characteristic, Quadratic Probability Score, Pseudo R2 and Kuiper’s Score to evaluate model performance. Empirical results show the warning system exhibits a high degree of accuracy and performs well. The variables identified show significant ability to signal vulnerability of the external sector of the economy. Policymakers can use the early warning system as the core of a larger set of variables on their radar to take pre-emptive measures to avoid crises or dampen their effects.

Keywords: Early warning system, Foreign exchange, Balance of payments, Crises, Model evaluation.