| The aim of this work is to forecast events (crises) in emerging markets. We introduce a novel approach in the speculative attack literature: we estimate exchange rate events and interest rate events, not by combining them in an index (as in mainstream empirical literature), but by estimating them simultaneously. We therefore develop and implement a system that produces one-month ahead probabilities of simultaneous events in the exchange rate and in the money markets. The development of the model was completed in three main stages: the selection of variables, the simultaneous estimation, and the implementation of the model. These stages are discussed separately in three chapters.;Chapter 1. Identifying the causes of exchange rate and interest rate events. We run regional as well as global regressions that attempt to estimate separately the two events. An exchange rate event is a dummy that takes the value of 1 if the exchange rate depreciation exceeds 10%. An interest rate event is a dummy that takes the value of 1 if the increase in the key intervention rate exceeds 500 basis points. Three conclusions were drawn from this stage: first, the variables that cause events do not greatly change from one region to another. It is therefore acceptable to pool across regions in order to take advantage of a higher ratio of events to non-events episodes. Second, first-generation variables, mainly domestic credit expansion, are as important in the nineties as they were in the seventies and early eighties. Last but not least, as shown by the regional Asia regression, the Asian 1997 crisis was about financial excess. The lending boom, manifested in a surge of claims on the private sector that was not matched by a parallel surge in GDP growth, was the most significant variable.;Chapter 2. Estimating simultaneously exchange rate and interest rate events . We know from theory and practice that an interest rate increase is an important factor in the determination of an exchange rate event. We also know that the decision of whether to devalue or not has important implications on the interest rate front. Therefore, a logical step would be to estimate simultaneously and not separately the two events. We implement a technique suggested by Amemiya and we find that the simultaneous estimation improves significantly the results in terms of type 1 vs. type 2 error. In both cases, the coefficient on the simultaneous event turns out to be the most significant of all variables. It also has a positive sign in the two regressions, suggesting that the increase in the probability of one event, simultaneously increases the probability of the other. This provides a strong argument against the interest rate defense policy, and it holds even when the definition of a defense is altered to consider different levels.;Chapter 3. Forecasting exchange rate and interest rate events . The model is extended to include several definitions of the exchange rate events. In addition to the 10%, we consider 5%, 15%, 20% and 25% devaluation levels. As the level of depreciation increases, contagion variables gradually loose their significance and fundamentals become more significant. We suggest a method of evaluating the forecasted probabilities by comparing them to action triggers that were obtained from the in-sample maximization of a return function. We then apply and test the model out of sample where it captures both the Russian (1998) and the Brazilian (1999) crises. It performs also well in capturing non-event episodes when applied to turkey. |
