Research on prediction of chaotic exchange rate time series applying dynamic component predicting model

In order to forecast chaotic variable of exchange rate, the paper integrated RBF neural network model, Lyapunov exponent model and Volterra adaptive model into a dynamic component model, weights of which could be adjusted by series themselves. Empirical research on 5 exchange rates showed both error indexes and direction statistics of component model could obtain better results than individuals, especially for JPY/USD and SEK/USD. Moreover, a compare on performance was taken between component model and random walk Model. Both D-M and H-M test refused null hypnosis, showed the component model could obtain obvious advantages than RW Model as expected.