| With the rapid development of mobile Internet,big data,artificial intelligence and other technologies,people’s investment concepts are also undergoing significant changes.From the previous investment based on investment experience and perceptual intuition,to the present quantitative investment method using computer combined with investment experience and perceptual intuition,making the investment way more scientific and rational.Over the years,in the quantitative investment model,the multi-factor quantitative stock selection strategy is still deeply concerned by investors.With the development of mobile Internet,big data,artificial intelligence and other technologies,factor data has become more and more.In view of the complexity and redundancy of factors in high dimensional data sets,it has become the focus of research to select effective factors from numerous factors.This paper constructs a multi-factor quantitative stock selection strategy based on penalty function method.The core idea of multi-factor quantitative stock selection is to show the intrinsic value and risk of assets as much as possible through multiple effective factors.This paper analyzes the multi-factor quantitative stock selection strategy,and selects 85 factors from the aspects of style factor,trend factor,momentum factor and reverse transformation factor.In the empirical part,the monthly frequency data of the last trading day of each month of the CSI 300 component stocks was used as the forecasting variable,and the monthly return rate was used to produce the target value for modeling.The time interval was selected from January 1,2010 to May 31,2022.The ratio of training set,verification set and test set is 6:1:3,that is,the period of training set is from January 1,2010 to May 31,2017,the period of verification set is from June 1,2017 to August 31,2018,and the interval of test set back test is from September 1,2018 to May 31,2022.According to the factor value of the sample and the estimated value of the rising probability of the model parameters as the score of each stock,select the top 15 stocks with the score value to form an investment portfolio,and allocate funds to buy the stocks in the portfolio according to the proportion of the score.In this paper,after data processing of factor data,based on the cross entropy loss function of three classification logistic regression(LR),penalty functions L1,L2 and Elastic Net were introduced respectively for factor screening and determining factor weights,and LR-L1,LR-L2 and LR-Elastic Net models were established.And the coordinate descent algorithm is used to solve the factor coefficient.Based on the above three models,LR-L1,LR-L2 and LR-Elastic Net strategies are constructed.In the back test,comparative analysis was conducted from profitability and stability to seek to establish a multifactor model to obtain more Alpha excess returns under certain risks.The backtesting results show that after penalty function is introduced,all three strategies can obtain excess returns,but compared with LR-L1 and LR-L2 strategies,the LR-Elastic Net strategy established with Elastic Net has more advantages in profitability and stability. |
