Empirical Reserch On Multi-factor Stock Selection Based On Nonparametric Methods With Lasso Procedure

With the development of China's capital market,more and more investors begin to study multi-factor stock selection model,trying to obtain excess returns.Therefore,it is necessary to find a suitable factor selection method for China's stock market and effective factors for China's stock market based on the research results of foreign multi-factor models.In this paper,we propose a nonparametric method with Lasso procedure to determine characteristics which provide important information on expected return.Firstly,we build stepwise linear regression model,linear regression Lasso algorithm model and non-parametric estimation with group Lasso procedure model to screen out the key factors.The factor selected by Lasso procedure of non-parametric estimation has the least number of overlaps with the other two models,which shows that non-parametric estimation can break through the restrictive framework of multiple linear regression model to screen effective factors.Then we compare the out of sample performance of these three models.The group Lasso model of nonparametric estimation has a significant increase in yield.In the end,the rolling regression method is adopted to optimize the non-parametric estimation Group Lasso model,which can improve the return rate without significantly increasing the risk.