Empirical Research On Text Stock Selection Factor Based On Stock Reversal Effect

In investing in the stock market,rational trading behavior will make different investment decisions based on different information,and different investment decisions will eventually lead to changes in investment returns.Quantitative investment,as an investment method that makes full use of company and stock information and reflects market changes,is becoming a hot investment method in China.In order to better guide investors to invest using China’s A-share market information,this paper constructs new stock selection factors and conducts empirical analysis.The traditional SUE(Standardlized Unexpected Earnings)factor is based on announced financial data to measure the post-earnings price drift(PEAD)effect and predict the abnormal return of stocks.This paper attempts to construct the text SUE.txt factor from the perspective of pure text,and deconstruct the text to mine alpha information.The empirical data based on the performance forecast and relevant research report texts shows that the SUE.txt factor has a strong stock selection ability,and the splitting and deconstruction of the text by the machine learning model is consistent with the intuitive logic,and the model has high credibility.This article takes the performance forecast and analyst research report collected by Chaoyang Sustainable Sustainable Company as a sample,selects all text data and some stock price data before 2021,uses machine learning to segment and disassemble the text data,and uses the performance forecast to release The stock price of a trading day before and after makes a label for it(rising,oscillating,falling).In this paper,three machine learning models,Logistic,XGBoost and Support Vector Machine Model,are used to classify and predict stocks from the beginning of 2021 to the end of June 2022.According to the classification probability and exponential decay,the text stock selection factor(SUE.txt)is obtained,and the stock selection factor is used.It is empirically tested by means of backtesting.From the results,The SUE.txt factor constructed based on the support vector machine model has some guiding significance for the portfolio,but because it does not have monotonicity,it is impossible to optimize the combinatorial optimization according to the factor value.In contrast,the SUE.txt factor stratification performance based on the XGBoost model and the Logistic model is relatively good,but the factor monotonicity of the XGBoost model fitting is better than that of the Logistic model.At the same time,although the rate of return of this factor is higher than that of the stock market benchmark,it is not significant,and this factor is strongly affected by the stock reversal effect.The group performance is the best.After removing the portfolios that are strongly affected by the stock reversal effect,the SUE.txt constructed by the XGBoost model and the Logistic model perform well.The factor monotonicity of the XGBoost model fitting and the optimal group investment return Both are better than the logistic regression model,and their returns are 20.28%and 20.25%respectively.It can be seen that the SUE.txt factor has obvious guiding significance for investors.