Empirical Study Of Quantitative Investment Transactions Based On Random Forest Algorithm

With the rapid development of China's economy,the Chinese stock market is constantly developing and improving.For investors,how to choose high-quality stocks for investment and get a good return on investment has become the top priority in stock investment.Price is a stock that is volatility around value.Stocks with undervalued value and potential for future development are the target of investors' pursuit.This paper analyses the characteristics of value-based investment strategy and growth-based investment strategy,combines their respective advantages,and based on the idea of GARP,constructs a set of index system which can reflect both the stock value and its growth.The system includes 23 indicators,such as market value(MV),price-earnings ratio(PE),market-to-net ratio(PB),asset-liability ratio(DTA),return on investment capital(ROIC),liquidity ratio(CR),inventory turnover rate(IT),net asset growth rate per share,etc.It can reflect the operating ability,profitability,growth ability and debt-paying ability of the selected stocks,and evaluate and screen stocks from various aspects and perspectives.This paper describes the theoretical basis of the random forest algorithm and multi-factor model.Then,using the established index system to design the experimental model based on the two algorithms,and establish the stock selection model based on Stochastic Forest algorithm and multi-factor model respectively.In view of the requirements of the random forest algorithm for the number of stocks in a single industry and industry,and the important position of the real estate industry in the entire Chinese stock market,this paper uses the real estate industry stocks in the WIND classification as an experimental sample,selected from January 2017 to December 2018,and selects the monthly rise and fall of real estate industry stocks from January 2017 to December 2018 as the experimental sample.The stock selection model constructed in this paper is based on monthly data.The basic idea is to screen the high-quality stocks in the real estate industry every month,so as to buy the selected high-quality portfolio at the beginning of each month,sell all the stocks held at the end of each month,and repeat six times.The whole experiment is divided into six rounds.Each round of training set contains 18 months of monthly data.The test set is monthly data from July 2018.For example,the second round of training set is from February 2017 to July 2018,and the test set is from August 2018.By analogy,six rounds of experiments were completed.By analyzing and comparing the results of six rounds of simulation experiments,the following conclusions can be drawn:(1)The stock selection model based on multi-factor model and random forest algorithm has good applicability in the real estate industry.The selected high-quality portfolio can achieve excess returns over the Shanghai-Shenzhen 300 index and the average level of the real estate industry in the test period.(2)In the comparison of the two models,by comparing the return rate,variance of return rate and Sharp ratio of the two portfolios,it can be seen that the portfolio selected by the random forest model is superior to the multi-factor model in both income level and income stability.Through the simulation experiment in this paper,we can prove that the stock selection model based on GARP and random forest algorithm has better performance and applicability in the real estate market.In the future,we can continue to improve the stock selection index system based on GARP,and further study and improve the random forest algorithm,so as to eliminate the impact of inter-industry differences,and apply this model to the entire Chinese stock market,so as to screen out the portfolio with investment value.