Predicting coke price by semiparametric regression method based on cointegration analysis

Abstract

According to the characteristics of consumption of coke, based on the analysis of Co-integration theory, the article utilized the model of semi-parametric regression to forecast coke' price by giving the data of coke' consumption market from January 1997 to April 2009 and so as to raise estimation precision of short-term coke' price. The test results of coke and iron price series for both Augmented Dickey Fuller (ADF) and co-integration analysis demonstrated that there was a co-integration relationship between the coke and iron price series. And then, the article suggested the parametric sectors was replaced by error correction operator. The result of linear analysis proved that the yield of coke was selected in non-parametric part. The semi-parametric regression combined with error correction operator was constructed. The article utilized cross validation method to select optimum bandwidth, choosing the Parabola kernel for the Kernel function. The least squares estimation was selected in new modeling estimation. The estimation results of real case was demonstrated that the semi-parametric regression model based on error correction operators not only reduced boundary estimation error but also strengthen economical interpretation. It is effective method to forecast coke' price in short-term.