基于因子偏离度的GBDT多因子选股模型

GBDT Multi-Factor Stock Selection Model Based on Factor Deviation Degree

为了避免股票市场中因子之间复杂非线性关系引起的多因子选股模型过拟合现象,基于因子偏离度对股票因子数据进行分析,筛选影响股票收益率的有效因子,通过梯度提升树对股票影响因子的权值进行不断调整和分析,建立一个DEV-GBDT量化选股模型,再根据基于因子偏离度的GBDT多因子选股模型的预测结果进行模拟交易,以沪深300指数成分股2010年1月1日—2019年7月31日数据为例进行实证分析。实验结果表明,DEV-GBDT选股模型的年化收益率达26.14%,比传统GBDT选股模型提高8.61%。基于因子偏离度的GBDT多因子选股模型能有效识别股市影响因子,提高股票预测准确度,帮助投资者获得超额收益。

In order to avoid the over-fitting phenomenon of multi-factor stock selection model caused by complex nonlinear relations among factors in stock market,the stock factor data is analyzed based on the factor deviation degree,and the effective factors affecting the stock return rate are screened.The weight of stock impact factors is continuously adjusted and analyzed through GBDT.A DVE-GB⁃DT quantized stock selection model is established,and then simulated trading is conducted according to the predicted results of GBDT multi-factor stock selection model based on factor deviation degree.The data from January 1,2010 to July 31,2019 of the CSI 300 in⁃dex are taken as an example for empirical analysis.The experimental results show that the annual return of DVE-GBDT stock selection model is 26.14%which is 8.61%higher than the traditional GBDT stock selection model.GBDT multi-factor stock selection model based on factor deviation can effectively identify the impact factors of the stock market,improve the accuracy of stock prediction,and help investors to obtain excess returns.