Construction Of Multi-factor Model And Research On Quantitative Stock Selection Strategy

After more than 20 years of development,the Chinese stock market has entered a relatively mature stage.Selecting more valuable and growing stocks from the huge stock pool has become a problem that attracts much attention from investors.This is also a meaningful academic issue that is of great practical significance.In this situation,the research on stock selection strategies is particularly important.Excellent stock selection strategies can not only spread risks,but also bring considerable benefits.Based on this,this paper uses the multi-factor stock selection strategy,a kind if stock selection strategy in the filed of quantitative,to study the stock selection problem in the A-share market.Based on the traditional multi-factor model,this article optimizes the multi-factor model,gives a more reasonable investment portfolio scheme,and provides investors with the opportunity to create more income.This article is mainly built from several parts,the introduction of introduction method,single factor validity test and empirical test of multi-factor model.Before introducing the relevant theories and methods of the multi-factor model,this article gives a detailed introduction to the current stock market situation and the researches relevant to the multi-factor stock selection model.Besides,it explains methods and conclusions of studies in detail of the development process of the multi-factor stock selection model.Since then,this article has introduced the methods used in each stage in the process of constructing a multi-factor stock selection model,start with the singlefactor validity test method,then turns to effective but redundant factor test and processing method,then multi-factor model construction framework,and finally the method of weighting in building a multi-factor selection model.After the theoretical method is introduced,the CSI 500 stock pool is used to test the validity of 239 factors in nine broad categories of the Quantitative Quantization Platform.At the same time,valid factors with strong correlation or multiple colinearity are merged or eliminated.Then,the traditional static equal weight factor-addition method,dynamic equal weight factor-addition method based on factor timing theory,and dynamic ICIR value weight factor-addition method based on factor timing theory are used to synthesize the effective factors.Finally,the score-based multi-factor stock selection models are constructed,and the CSI 500 stock pool are used to test the model effect.In the method introduction section,this article details the methods often used in each step of the multi-factor stock selection model establishment process.The establishment of multi-factor stock selection model is mainly divided into several steps: single-factor validity test,to deal with effective but redundant factor,and multi-factor model construction.For the single factor validity test,there are three methods: ICIR value analysis,regression and group test.For the part of dealing with effective but redundant factors,correlation between factor sequences and whether there is multicollinearity are examined.There are three methods for constructing multi-factor models: ranking,regression,and scoring.Among them,the multi-factor stock selection model based on the scoring method is used the most.For the single factor validity test,this paper considers IC value analysis and group test to check the validity of the factors.For the factor data in the CSI 500 stock pool from December 2009 to November 2019,factors with an average IC value sequence greater than 0.05 and an IR value greater than 0.5 are selected each month.The factors selected according to the sequence of ICIR values have a better ability to explain the rate of return and the ability to obtain excess returns in a stable manner.Thereafter,the grouping test is used to test the discrimination and monotonicity of the return on stock portfolios selected according to different factor values.At the same time,a long-short investment portfolio is constructed to compared with the return of the CSI 500 Index.After the effective factor list is obtained,factors belong to the same class are synthesized to eliminate the correlation.Finally,based on the scoring method,the CSI 500 index stock pool from December 2009 to November 2019 are used to construct multi-factor models with three different factor weighting methods.They are static multi-factor model of equal weight method,dynamic multi-factor model of equal weight method based on timing theory,and dynamic ICIR weighted multi-factor model based on timing theory.From the results,the annualized return of the portfolio selected by the dynamic ICIR weighted multi-factor model reached 16%,and the annualized return of the long-short portfolio reached 28%.The annualized return of the dynamic equal weight multi-factor model portfolio reached 14%,and the annualized return of the long-short portfolio reached 27%.However,the annualized return of the portfolio selected by the static equal weight multi-factor stock selection model was only 9%,and the annualized return on longshort portfolios has reached only 4%.The optimized dynamic multi-factor models based on timing theory have better performance than the benchmark index.