ARIMA融合神经网络的人民币汇率预测模型研究

Research on RMB Exchange Rate Forecasting Model Based on Combining ARIMA with Neural Networks

本文在深入分析了单整自回归移动平均(ARIMA)模型与神经网络(NN)模型特点的基础上,建立了ARIMA融合NN的人民币汇率时间序列预测模型。其基本思想是充分发挥两种模型在线性空间和非线性空间的预测优势,即将汇率时间序列的数据结构分解为线性自相关主体和非线性残差两部分,首先用ARI-MA模型预测序列的线性主体,然后用NN模型对其非线性残差进行估计,最终合成为整个序列的预测结果。通过对三种人民币汇率序列的仿真实验表明,融合模型的预测准确率显著高于包括随机游走模型在内的单一模型的预测准确率,从而证实了融合模型用于汇率预测的有效性。这一结果也表明,人民币汇率市场并不符合有效市场假设,可以通过模型对汇率未来走势做出较准确预测。

Based on analysis of the autoregressive integrated moving average （ARIMA） and neural networks （NN） models, this paper presents an ensemble approach to RMB exchange rate time series forecasting which combining ARIMA with NN. The RMB exchange rate time series are considered to be composed of a linear autocorrelation structure and nonlinear structure. ARIMA is used to model the linear component of exchange rate time series and the NN model is applied to the nonlinear residuals component prediction. The results of RMB exchange rate forecasting show that the proposed model, which integrates the unique strength of the two models in linear and nonlinear modeling, has the more forecasting accuracy than that of single model （including random walk model） . It provides evidence against the efficient market hypothesis and suggests that there exists a possibility of predicting it into the future.