Forecasting Coke's Price by Combination Semi-Parametric Regression Model

Abstract

According to the characteristics of consumption of coke, the article utilized combination semi-parametric regression method rather than usual regression analysis to forecast the coke's price. Coke' price was divided into two parts. Parameter part was analyzed through error correction model and BP artificial neural network. Nonlinear part was fitted by core function. Nonparametric sector was described by coke yield. 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 model estimation. The estimation results of real case demonstrate that error correction-semi-parametric regression model and BP artificial neural network-semi-parametric regression model not only reduced boundary estimation error but also strengthen economical interpretation. It is an effective method to forecast coke's price, which can largely raise the estimation precision of coke's price.