Modelling exchange rates during currency crisis using neural networks

This paper presents an artificial neural network (ANN) approach to the forecasting of exchange rate movements during periods of currency crises characterized by excessive volatility. The models are built using the feedforward ANN structure trained by the backpropagation algorithm. Exchange rate data from the Lebanese currency crisis period of 1985-1992 is used for training, testing and evaluation of the models. Forecasting performance measures represented by well established error functions are presented for the three models. The best model shows that artificial neural networks are able to forecast exchange rates during periods of extreme fluctuations.