Comparative Analysis Of Factor Selection Model Based On Regression Method And Scoring Method

The multi-factor stock selection model has always been a very important research area in quantitative investment.The multi-factor stock selection model is one of the important stock picking models in quantitative investment.It can comprehensively consider many factors affecting stock returns,and finally draw a stock picking result.In different market conditions,the multi-factor stock selection model will be relatively stable,because there are always some factors that will work.Therefore,in the quantitative research,investors and researchers have developed many different types of multi-factor stock selection strategies.Based on the traditional linear multi-factor stock selection model constructed by regression method,this paper adds the equal weight method and two unequal weight scoring stock selection models(IC mean weighting and IC_IR weighted scoring method),so that the backtest results A significant improvement has been made.Based on the study of market experience and economic logic,and considering the availability of data,this paper screens out seven major factors that are theoretically highly explanatory for stock returns.A total of 31 candidate factors,most of these factors reflect the financial and operational factors of the company's fundamentals.Then from January 2010 to November 2018,an empirical study was conducted on 31 factor data of all Shanghai and Shenzhen 300 constituent stocks.Firstly,31 factors were tested for single factor validity,and the redundancy factor was removed by the large class analysis method.Finally,five effective linear factors were obtained,namely: pe_ttm(P/E ratio(TTM)),yoy_or(Revenue growth rate,roe_ttm2(ROE(TTM)),pct_chg_per(interval range),pq_avgturn2(interval daily turnover rate).Using the five effective linear factors to construct a multi-factor stock selection model based on regression method,a multi-factor stock selection model was constructed.During the backtest,we obtained an excess return relative to the benchmark CSI 300,with an annualized Alpha of 3.5%.In order to further compare the results of different multi-factor stock selection models,the scoring method stock picking model is added in this paper.The equal weight and unequal weight scoring methods are used to score the effective factors of the stock respectively,then weighted.Get a total score,then sort the stocks from high to low according to the total score,and finally filter out the top stock construction portfolio.Finally,from the backtest results,it can be seen that compared with the benchmark CSI300,the equal weight scoring stock selection model has the highest excess return rate during the backtesting period,and the multi-factor stock selection model based on the regression method is second,while the other two types are different.The annualized rate of return obtained by the weighted scoring method is very unsatisfactory.The equal weighting methodis 11.4% for the annualized Alpha relative to the CSI 300,and the Sharpe ratio is 0.75.Compared with the multi-factor stock selection model,the annual volatility of the equal weight scoring method has increased from 3.5% to 11.4% compared with the CSI 300.Therefore,among the four stock selection methods,the equal weight scoring method can use the linear effective factor to score a batch of higher quality stocks,and thus obtain a higher excess return rate than other stock selection models,so in practical applications.In our case,we prefer to use the equal weighting method for stock selection.