Studing On Nonparametric Autoregressive Model And Its Application In Prediction Of Exchange Rate

Nonparametric autoregressive model have gained much attention recently, due primarily to the fact that they can describe some nonlinear features exhibited by many data itself in applications. At the same time, nonparametric methods is certainly flexible in reducing modeling biases so that they become the most important methods of studying nonparametric model. The focus of this thesis is comparing some nonparametric methods of nonparametric autoregressive model and establishing the nonparametric autoregressive model on CNY/USD exchange rate after the reform of RMB exchange rate genesis on July 21.2005. The main contributions are as follows:1. Based on several nonparametric estimation theories, several estimation expressions of conditional heteroscedastic function are constructed. The polynomial spline estimation of nonparametric autoregressive model is illustrated better by simulations.2. In empirical part, the paper establishes the ARMA model on CNY/USD exchange rate; then we establish the general nonparametric autoregressive model which utilize polynomial spline estimation; but we found that the residual sequence has heteroscedastic property by related test of residual squared sequence, so we establish the nonparametric autoregressive conditional heteroscedastic model which can reflect the heteroscedastic property; at last, after the comparison of the ARMA model and the nonparametric autoregressive conditional heteroscedastic model about CNY/USD exchange rate, the results show that CNY/USD exchange rate has the characteristic of heteroscedasticity, nonparametric autoregressive conditional heteroscedastic model can describe the pattern of CNY/USD exchange rate better and predict more accurately than ARMA model.