Nonlinearity in Exchange Rates and Forecasting

Exchange rates are subject to large and frequent fluctuations in mean and variances making it very difficult to model and forecast. In this paper, a series of tests for nonlinearity and chaos in exchange rates is conducted using the daily data on the market rates in Iran for the period 1991-2005. The tests for nonlinearity are BDS and ANN tests, and the tests for chaos are autocorrelation and Lyapunov exponents. The tests results suggest that the exchange rates and their rates of change follow complex nonlinear and stochastic processes. In the second part of the paper, an ANN model is designed to forecast the exchange rates. The results show that ANN outperforms both ARIMA and GARCH models in forecasting the exchange rates, but generates the same results as the alternative models in forecasting the rate of change of the exchange rates.