Using A Nonlinear Combined Model Forecast The Exchange Rate Based On The Support Vector Machine

ABSTRACT:After we review the existing literature, we find that the indicators which used to test the effectiveness of the model are limited in mean square error (MSE), mean absolute error (MAE), mean absolutr percentage error (MAPE). The disadvantage is that these indicators are over pursuiting the prediction accuracy while ignoring the usefulness of the model. So we build a comprehensive evaluation system which can judge the validity of the model from a practical point of view, In this system, we choose the return on the investment as the core indicator, the direciton accurate rate and the standard deviation are choosed as secondary indicators. Secondly, we select the MA5、 MA10、MA20、MA60and MA120as the input vector and build a support vector machine. Compare this model with the time series model, we find the technical indicators forecasting model has a better result based on our evaluation system.This research also make up a blank field of the exchange rate. At last, we select the weekly data of US/CNY exchange rate from2005.07.21to2013.12.31as the sample, the5days moving average (MA5), the10days moving average (MA10), the20days moving average (MA20), the60days moving average (MA60), the120days moving average (MA120), the predicted value of ARIMA and GARCH model are the input vector, and the next weekly data of exchange rate is the output vector, so we get a new nonliner combined model based on support vector machine. Comparing this new nonliner combined model with the single ARIMA and GARCH model, empiricial results show that the effect of the new nonliner combined model is better than the single ones and the investment based on the nonlinear model obtained a positive return of0.83%.