Establishment Research Of Dynamic Multi-factor Stock Selection Model Based On Ensemble Learning Algorithm

In recent years,Chinese stock market has been developing rapidly.As of December 31,2020,there were 4,140 A-share listed companies with a total market value of nearly 80 trillion yuan.How to excavate the stocks with high investment value,which can obtain the excess income,is the primary concern of the majority of investors.For the purpose of analysing the availability of quantitative stock selection strategy in the A-share market,this paper intends to conduct an indepth analysis of the relevant methods for the construction of multi-factor stock selection model,so as to make the portfolio obtain considerable excess returns.In this paper,we first select the constituent stocks of the CSI 300 Index as the stock pool,and select 34 candidate factors according to the information contained in the pool,which is divided into six types: valuation,quality,growth,sentiment,risk and technology,obtain financial and market data from the first quarter of 2011 to the fourth quarter of 2020,including yield and factor exposure.After the lag processing,processing of missing values and outliers,as well as data standardization,the validity and significance of the candidate factors are tested by the method of sorting and the method of IC value sequence,and the Spearman correlation coefficient is used to remove the redundant factors which have strong correlation with another one and weak correlation with the return rate.Then,use the ensemble learning to fit the Nonlinear regression model between the return rate and the factors,and the training window is set up to update the model over time,which solves the problem that the multi-factor stock selection model lacks timeliness in practical application.The parameters of the model are optimized by cross-validation and grid search,and the optimal multi-factor stock selection model is obtained by comparing the models fitted by different ensemble learnin algorithms.By modeling and comparing,the dynamic multi-factor stock selection model based on Light GBM algorithm can fit the effective factors such as operating revenue,net profit growth rate,average daily turnover rate of circulating shares,etc.,the average annualized return of return is18.60% and the five-year cumulative excess return of the CSI 300 Index is 94.95%,which proves the effectiveness of the multi-factor stock selection strategy in the A-share market.In addition,we also found that in the period when the market structure is relatively stable,the effect of historical data and market rules are more obvious,when the multi-factor stock selection model is applied to actual investment,can achieve a more substantial excess return.