Multi Factor Stock Selection Model Based On Machine Learning

In recent years,quantitative investment has developed rapidly,gradually occuping an important position in the capital market,which has promptly become a new method in the international investment community.The paper illustrates a new IC weighting method in the multi-factor stock selection model which refers to the half-life used in physics and medical science.The weight of half-life is used to measure the effect of IC on the weight of the factor in the most recent time,which can be called half-life IC weighted.In this paper,important factors are selected by combining XGBoost method in machine learning,and then important factors are weighted by half-life IC weighting.Then,common IC weighting,support vector machine and neural network are used for stock selection,and the four methods are compared by annualized stock returns.Select the data of 8 categories of factors generated during the period of early 2010 to early 2019,including value,growth,scale,trading,sentiment,share,quality and risk from the specific stock pool formed by constituent stocks in HS300 and additionally use XGBoost method which is based on machine learning to select key factors.Use half-life IC weighted method to weight the key factors and select the top 10% ranked stocks.The result indicated that the final annualized return is 26.86%,24.81% higher than the 2.05%annualized return of the HS300 index.Secondly,the paper indicates that using common IC weighted,support vector machine,and neural network,combined with the important factors selected by XGboost,which obtained annualized of 8.17%?10.98%?3.39%,that illustrates the machine learning method can also be used to obtain far higher returns than only referring to benchmark HS300 index.The research concludes that the half-life IC weighted method is the best choice compared to the other various method,which can be put forward the new thinking for quantitative picking stocks.