Forecasting chaotic time series of exchange rate based on nonlinear autoregressive model

Exchange rate time series is often characterized as chaotic in nature. The prediction using conventional statistical techniques and neural network with back propagation algorithm, which is most widely applied, do not give reliable prediction results. Exchange-rate time series is also a dynamic non-linear system, whose characteristics cannot be reflected by the static neutral network. The Nonlinear Autoregressive with eXogenous input (NARX) includes the feedback of the network output, therefore can reflect the dynamic property of the system. This paper proved the chaotic property of the exchange-rate time series, calculated the embedding dimension and time delay of the series, and established the exchange-rate forecast model using the NARX network. The result shows that the NARX network has better short-term forecast effect, comparing to the BP network and the SVM model.