Multi-factor Stock Selection Strategy Research And Empirical Analysis Based On Machine Learning

Multi-factor stock selection model is one of the classic model of quantitative investment.Its main idea is based on the theory of financial gain some factors related to stock yield data,through the analysis of the mathematical algorithm fitting the relationship between the multiple factor and yield.Then all kinds of investor can take advantage of a well-constructed model to establish a stock selection strategy and invest for a high yield.With the continuous development of artificial intelligence,machine learning algorithm can be widely used to deal with high-dimensional data problems,which can get better results.Therefore,this paper builds a new multi-factor stock selection model and strategy based on machine learning,conducts back-test on the strategies obtained by different methods,and compares the back-test results.Firstly,this paper introduces the theoretical basis of the multi-factor stock selection model,the theory and optimization method of Softmax regression to deal with the multi-classification problem,and the concept and steps of establishing the machine learning ranking model using Ranknet algorithm.Secondly,the factor data and stock return rate data from January 1,2011 to December 31,2017 were obtained from Ricequant quantitative trading platform.Factors were screened through factor validity analysis,and principal component analysis was used to process data to avoid multicollinearity caused by strong correlation among factors.Finally,three multi-factor stock selection models are constructed.The first one is to use regression method to fit the linear relationship between stock return rate and multi-factor.The second one is to divide the rate of return into ten categories according to the size,and establish a multi-classification model by Softmax regression algorithm,then compare the differences between different optimization algorithms.The third one is to implement Ranknet algorithm based on neural network to predict the ranking of stock returns.Then use the constructed model to write the multi-factor stock selection strategy,and carry out empirical analysis and back-test from January 1,2018 to October 30,2020.By comparing the back=test results of different models,it is shown that the model established by Softmax regression has a higher return rate,and the ranking model established by Ranknet algorithm is more accurate.The return on the back-test is slightly lower than the Softmax regression model,but it faces less risk of loss.