Abstract:With the development of economy, stock investment has come into public view. How to choose constituent stocks to track the stock index has been paid more and more attention by the people. Based on this, aiming at the problem of stock index tracking, a method of variable coefficient product model for variable selection is proposed. Based on B-spline function approximation technique, this method combines LPRE (Least Product Relative Error) criterion and group SCAD (Class Clipped Absolute Devation) penalty function to apply to the variable coefficient product model.The implementation steps of solving the estimation are given by Newton iterative algorithm and local quadratic approximation. In order to verify the effectiveness of the proposed method, the results of SCAD penalty method with variable coefficient product model (LPRE-S) and the least square SCAD penalty method with variable coefficient model (LS-S) were compared by numerical simulation. In order to verify the practicability of the proposed method,LPRE-S estimation method and LS-S estimation method are compared for the tracking and forecasting effect of dividend index in Shenzhen Stock Exchange.The results show that the ratio of the LPRE-S estimation on the method to select the real model is almost close to 1, which can better achieve the purpose of variable selection and has a good prediction effect in the stock index tracking.