Research On Stock Selection Strategy Based On Random Forest Algorithm And Analyst Recommendation

This paper combines the random forest method with the traditional factor model to predict whether the analyst’s target price can be reached within a certain period of time in the future.According to the prediction results,the stock target with the highest rising probability is selected to establish an effective investment portfolio and achieve a higher The excess return on the market average.First obtain all research reports with target prices from 2015 to 2019 and rated as buy or overweight,and then analyze relevant literature to select market factors,company fundamentals factors,investor behavior factors,research report characteristic factors,There are a total of 81 factors in six categories: stock market style factor and research institution characteristic factor as pre-selected factors.The data is selected from 2015 to2019.Using the dynamic modeling method,using 3 months as a sliding window,using 3,6,and 12 months of research report data as training samples to predict the target price of research reports 3,6,and 12 months after the research report is issued Achievement situation.After that,hyperparameter tuning is performed and the best training duration is selected according to the prediction results under different windows.The top 10 stocks with the highest probability of being positive are extracted from the forecast results of each period as the stock pool and held for 3,6,and 12 months.It is found that they can obtain a positive return higher than the Shanghai and Shenzhen 300,and then use three methods To optimize the model,first in the weight selection,because the original model adopts the method of average shareholding,the first optimization process introduces the assumption of unlimited funds to observe the maximum potential of the model performance,and then expands the adjustment period and label length And perform modeling back-testing again,and finally use the factor rotation method to select the best factor of each period to back-test again.This paper uses machine learning methods to study the accuracy of analysts for the first time,and expands the relevant influence factors on the basis of the existing literature to improve the feasibility of the model.The best total return rate obtained by the model is 323.32%,the annualized rate of return is 35.5%,the annualized excess return rate based on the CSI 300 is 30.75%,and the maximum drawdown rate is 26.36%,which is far ahead of the market benchmark.It is believed that this model has better stock picking performance.In the method of improving the stock holding method of the model,changing the label length and factor rotation,each method can further enhance the effect of the model and improve the investment performance of the original model.