Predicting short-term coke price by nonlinear semiparametric regression method

To improve short-term coke price prediction accuracy, a semiparametric regression idea and linear and Nonlinear semiparametric regression models are specified in this Article, on the basis of the coke consumption characteristics, the coke consumption market data for January 1997 to April 2009, as well as the regression analysis method. In the linear estimation results, the parameters of semiparametric regression model are the price of cast iron, nonparameters are coke output. Cross Valuation Method is applied for choice of optimum window frame. Parabola kernel is chosen for kernel function. Semiparametric model prediction method is lease square estimation method. The cases of coke>40mm indicate that Nonlinear semiparametric regression estimation is more accurate, it´s an effective tool for predicting short-term coke price.