基于EMD与NARX网络的汇率预测方法研究  

On Exchange Rate Forecasting Method Based on EMD and NARX Neural Network

作　　者：马超[1] 徐瑾辉 蓝斌[1] 侯天诚 欧阳泽拯 

机构地区：[1]广东外语外贸大学金融学院,广州510006

出　　处：《西安文理学院学报（自然科学版）》2015年第2期70-74,共5页Journal of Xi’an University(Natural Science Edition)

基　　金：国家级大学生创新训练计划资助项目(201411846001)

摘　　要：探索构建对汇率进行直接预测的高精度组合预测算法.采用NARX神经网络作为基础模型,并结合经验模态分解模型进行混合预测,提高模型精度.为研究不同时间间隔对预测结果的影响,采用美元兑日元汇率的时间间隔为5分钟与1天的数据进行预测.实验结果显示,时间间隔较短时,模型的预测精度更高.此外,通过对汇率改革前后的人民币汇率的预测发现,汇率改革对EMD-NARX模型的预测结果影响不大,说明模型稳定性较高.We have explored the high-precision combined calculation for direct exchange rate forecast. We adopt the NARX neural network as the original model. In order to enhance the forecast precision, a forecasting method based on empirical decomposition mode is also pro- posed. To explore the effects of various time intervals on the prediction outcomes, the 5-minute and daily data between US and Japanese Yen exchange rates are used respectively in the experiments. The forecasting results indicate that the precision is higher when the time interval is shorter. Besides, by forecasting the exchange rate of RMB before and after the exchange rate reform, we discover that the exchange rate reform has little effect on EMD-Narx model, which means that the model is relatively stable.

关 键 词：汇率预测 NARX神经网络 经验模态分解(EMD) 组合预测 误差分析 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]