Research On Multi-factor Stock Selection Model Based On XGBoost

With the continuous improvement of the development of my country's securities market and the dual promotion of the development of quantitative investment technology,quantitative investment has gradually attracted the attention of domestic investors,especially institutional investors,and many institutional investors have gradually begun to establish quantitative investment as their main investment method.Fund products.Nevertheless,the proportion of quantitative investment in our securities investment market is still quite small.One is the small scale of the quantitative investment market and the unstable performance of quantitative strategies;the second is that there is a certain degree of convergence in quantitative strategies.Therefore,research on new quantitative strategy models is particularly necessary for the development of quantitative investment in China.This article uses the financial,dividend,scale,valuation,quality and other factors of the 300 constituent stocks in Shanghai and Shenzhen from 2011 to 2020,effectively uses the trading data on the stock market,and selects multiple factors based on the XGBoost algorithm in the statistical learning method.The construction of the stock model uses a rolling training method to construct a dynamic stock selection model,which has obtained excess returns above the benchmark.At the same time,this paper compares the XGBoost,random forest and support vector machine algorithms theoretically and empirically.The results show that the XGBoost model is better than the support vector machine and the random forest control model from the perspective of the correct rate ACC and AUC value.From the perspective of measuring the combined return rate index and risk index,although the information ratio and calmar ratio of the XGBoost model are higher than other models,the maximum retracement rate is higher than that of the support vector machine.In order to reduce the maximum retracement rate of the XGBoost algorithm,this article adds a stop loss condition to optimize the model to reduce the impact of the overall market macro factors and reduce interference.The test results show that the optimal portfolio return rate before adding the stop loss condition is92%,And the maximum retracement rate reached 42%.After adding the stop loss optimization condition,the maximum retracement rate has dropped significantly(from 42% to 39%),and the annualized rate of return has risen from 11.87% to17.40%.Optimization The effect is obvious.In summary,the multi-factor stock selection model based on XGBoost can achieve higher returns,and the backtest effect is relatively stable.Although the maximum drawdown rate is relatively large,it can also be overcome by stop loss optimization,thereby obtaining a more complete investment Strategy.