AN APPROACH OF COMBINING EMPIRICAL MODE DECOMPOSITION AND NEURAL NETWORK LEARNING FOR CURRENCY CRISIS FORECASTING

This paper presents a hybrid model for predicting the occurrence of currency crises by using the artificial intelligence tools. The model combines the learning ability of the artificial neural network (ANN) with the inference mechanism of the empirical mode decomposition (EMD) technique. Thus, for a better detection of currency crises emergence, an EMD-ANN model based on the event analysis approach is proposed. In this method, the time series to be analyzed is first decomposed into several intrinsic mode components with different time scales. The different intrinsic mode components are then exploited by a neural network model in order to predict a future crisis. For illustration purposes, the proposed EMD-ANN learning approach is applied to exchange rate data of Turkish Lira to evaluate the probability of a currency crisis. We find evidence that the proposed EMD-ANN model leads to a good prediction of this type of crisis. Significantly, the model can thus lead to a somewhat more prescriptive modeling approach based on the determination of causal mechanisms towards finding ways to prevent currency crises.