Research And Implementation Of Quantitative Stock Selection Model Based On Factor Mining

In recent years,with the gradual increase in the reform and innovation of China’s capital market,the combination of finance and science and technology has become more and more close,which has promoted the leapfrog development of quantitative investment.Multi-factor model is one of the most successful strategies in the current quantitative stock selection method.It is favored by the majority of investors because of its mature,practical and flexible characteristics.However,the continuous switching of market styles has led to the failure of a large number of factors,which has a negative impact on the yield of the stock selection model.Factor mining and factor timing are effective means to solve such problems.Traditional factor mining methods rely too much on manual analysis.Therefore,with the dramatic development of data and computing power,this approach is now difficult to work.Although some methods employ intelligent optimization algorithms,they still lack a nonlinear representation of the data.At the same time,in the field of factor timing research,there also exists simple modeling methods,data fusion difficulties and other issues.To address these challenges,a factor mining method based on a deep learning operator and a multi-factor stock selection method based on timing prediction are proposed respectively in this thesis,providing a complete solution to the problem of factor failure.The first method focuses on how to build a factor calculation model to generate new factors.In this approach,four kinds of deep learning operators with different functions are designed as the basic operation units for constructing factor calculation model.Then factor mining algorithm is used to generate a large number of factor calculation models,thus obtaining a large number of deep alpha factor elements through those models.The second method focuses on how to adjust the factor configuration mode in time according to the effectiveness of the factor in the future.First of all,we designed a timing index for a comprehensive evaluation of factor performance,then we combined the momentum effect prediction module with the deep learning prediction module to predict the future performance of the factor.Finally,a stock selection model which can dynamically allocate factors according to the prediction results is realized.Finally,this thesis integrates the above two methods and tests them based on the quantitative platform.Firstly,a large number of deep alpha factors are generated by factor mining method,which constitutes a factor pool to construct a multi-factor stock selection model.Then we conduct daily frequency stock selection prediction,and simulate trading on the quantitative platform according to the prediction results.According to the experimental results,the multi-factor timing stock selection model proposed in this thesis has a strong ability to excess returns,and the work of this thesis provides certain theoretical innovation and practical experience for quantitative investment research.