Research On Multi-factor Quantitative Investment Strategy Based On Feature Optimization

Quantitative investment is a trading strategy that uses mathematical statistical models and computer technology to discover the laws of the stock market and apply the laws to investment.Compared with traditional manual investment,quantitative investment is more objective and can effectively reduce investment risks caused by people's subjective emotions.The multi-factor model is one of the important models of quantitative investment.It has the characteristics of high efficiency and strong plasticity,so it is accepted and used by the majority of institutions and individual investors.However,in recent years,due to the continuous changes in the A-share market and the excessive use of multi-factor models by investors,many funds based on multi-factor models have performed poorly.In order to more accurately grasp the direction of market changes,select the most effective factors.In this paper,a hybrid feature selection method is used in the factor screening stage,that is,filter method first,and then wrapper method.In the filter method,the two criteria of factor information coefficient and factor grouping return are used for filtering to obtain a preliminary feature subset.Then use the genetic algorithm as the search strategy of the wrapper method to find the optimal feature.Among them,in the genetic algorithm search,this paper uses the decision tree algorithm as the fitness function and the receiver operating characteristic curve area as the fitness evaluation value to establish a hybrid feature selection model based on the genetic algorithm.This paper selects the stocks of the Shanghai and Shenzhen 300 Index from 2016 to 2019 as the backtest samples to construct a multi-factor model.First,a single factor test is used to select a total of 305 factors in 7 categories,and the candidate factor pool of this article is established.Then the above model is applied to factor screening,and the concept of rotating factor is introduced at the same time,that is,factor data of the previous 6 months are used for factor testing on each adjustment day.Then use random forest algorithm to dynamically model factors and returns,and compare performance with fixed factor models and traditional factor test models.Finally,in order to reduce the risk,this article uses a timing strategy based on the MACD indicator to optimize the model.The results show that the performance of the model proposed in this paper is significantly improved compared to the comparison model,not only in the return index,but also in the risk index.After using the timing strategy,the annualized return of the model can reach 24.46%,Sharpe ratio of 1.81,illustrates the applicability of this model in China stock market.