Autor: |
Zhenhua Zhang, Zezheng Ouyang, Chao Ma, Jinhui Xu, Longxin Li, Qing Wen |
Předmět: |
|
Zdroj: |
International Journal of Simulation -- Systems, Science & Technology; 2016, Vol. 17 Issue 47, p1.1-1.7, 7p |
Abstrakt: |
Currency exchange rate is affected by many factors, which are often uncertain, thus exchange rate time series is a typical non-stationary time series. However, only stationary series forecasting methods are used on the research of exchange rate forecasting. Based on the non-stationarity that exchange rate possesses, a novel combinational forecasting model suitable for nonstationary time series is proposed in this paper. First, we adopt the NARX neural network as the original forecasting model. And then, we operate a novel prediction model combining empirical mode decomposition (EMD) with NARX neural network to improve the forecast precision. Finally, we proposed a combination model according to NARX model and EMD-NARX model and use two examples to demonstrate the prediction effect. To study the difference of prediction results in different time intervals, we use the 5 min exchange rates and daily exchange rates of US dollar against Japanese yen. The forecasting results indicate that the precision is likely to be higher when the time interval is shorter. Moreover, by forecasting the exchange rate of RMB before and after the exchange rate reform, we find that the exchange rate reform barely affects EMD-NARX model, showing the relatively high stability of the model. [ABSTRACT FROM AUTHOR] |
Databáze: |
Complementary Index |
Externí odkaz: |
|