Coal Price Index Forecast by a New Partial Least-Squares Regression

摘要

Deviation of coal price has great influence on growth of China's economic. Daily coal price indexes in Qinhuangdao were collected. Past twenty days were used to predict next day index. The principal components of twenty days were extracted. The function between output variable and components was fitted by linear, quadratic and exponential model. This improved traditional partial least-squares regression. Traditional method such as multivariate linear regression and polynomial regression were coming into comparing with our method. Improved quadratic partial least-squares obtained the smallest relative errors in mean and variance for ten reserved indexes. Those ten errors had minimum 0.3%, median 3.3% and maximum 9.7%. The ideal forecast precision certified that quadratic partial least-squares was suitable for coal price indexes.