An Empirical Study On Multi-factor Quantitative Stock Selection Based On Boosting Algorithm

Quantitative investment is an emerging investment method in the financial mar-ket.It makes investment decisions through mathematical models and computer pro-grams,and has the advantages of discipline,decentralization,and efficiency.In the field of quantitative investment,the multi-factor model has the characteristics of simplicity,high efficiency,and strong plasticity,so it is very popular in the industry and academia.When using a multi-factor model to build stock selection strategies,the process can be roughly divided into two parts:factor selection and regression model construction.Most of the traditional multi-factor stock selection models are more concerned about the financial data of listed companies.They usually choose financial index factors such as market value,price-earnings ratio,and return on net assets to mine the company's fundamental information.But financial data often lags behind.When the market style changes,it is easy to bring greater losses.In response to this problem,this article in-novatively constructed a factor pool containing six categories of factors with reference to the latest practices in the industry.Specific factors include valuation factors,qual-ity factors,growth factors,momentum factors,sentiment factors,and technical factors.In this way,the transaction data on the stock market is effectively used,and we can more quickly and comprehensively analyze the factors that affect the future return of the stock.In addition,traditional multi-factor models use multiple linear regression to fit the relationship between factors and stock returns.As a result,factors that are non-linearly related to stock returns cannot be used effectively.In this regard,this article uses Boosting algorithm to replace the traditional regression model,and builds a multi-factor quantitative stock selection model based on Boosting algorithm.Moreover,compared with the traditional static model,this article uses the rolling training method to train the model and build a dynamic stock selection model,which can better adapt to market style changes.In the process of parameter optimization,this article uses Bayesian opti-mization algorithm,which is faster than grid search and random search,and improves the efficiency of modeling.For the stock selection model constructed in this article,we use the data of the con-stituent stocks of the CSI 300 INDEX and the CHINEXT PRICE INDEX for empirical analysis,and the time span of the data is from 2014 to 2019.We compare the sub-division algorithm under the Boosting algorithm such as Adaboost,GBDT,XGBoost,LightGBM with the traditional linear regression model.The empirical results show that both the Boosting algorithm model and the traditional linear regression model can ob-tain more than the CSI 300 INDEX,and the performance of the Boosting algorithm is generally better than the traditional linear regression model.This shows that the idea of using Boosting algorithm to improve the traditional linear multi-factor stock selection model is feasible.In the end,this paper uses the weighting of return volatility to further improve the profitability of the model,thereby obtaining a more complete investment strategy.