The Research Of The Copper Futures Price Prediction Research Based On The GLAR And ANN Nonlinear Integrated Model

As an important raw material in the national economy production, copper has an important position in the development of national economy. For the past few years, the frequent fluctuation of copper price influences on the enterprise's production and management. So it has important theoretical and realistic significance to study the copper price prediction.Basing on the domestic and foreign research present situation, this paper put forward to use integrated model to predict the copper price, which described the deficiency of the linear and nonlinear models; From the copper's industry attribute and the metal property, this paper has select the11main factors which affect the copper price. The result of the Grainger Granger causality test show that the other factors are not the Granger reason of the SHFE copper price except the LME copper price.This paper has using EMD method for separating copper price as the long-trend and the high frequency component IMF sequence and using the system clustering method to synthesis for the long-term trend, high frequency part and the low frequency part. Using the Granger test to select the factors which affect the three components, this paper has established the prediction models. The result of the prediction show that the long-trend component has a highly linear, which can be predicted by the linear model, while the explanation for the low frequency part and the high frequency is not high, in particular the linear model has failure for the high frequency. The GLAR and ANN nonlinear integrated model has proposed based on the above analysis. An outstanding merit of this model is that it can play in the linear and nonlinear model of their respective strengths. We then apply this prediction model in SHFE copper price and compare with other forecasting methods:ARIMA model, GLAR model, ANN models, and so on. Empirical analysis shows the GLAR and ANN nonlinear integrated prediction model achieves the best forecasting results.