A Quantitative Investment Strategy Based On MIDAS-XGBoost Method

Attracted by the high rate of return in the financial market,the research on quantitative investment theory has been extended to various fields,from finance to statistics,computer science,informatics and other disciplines.Traditional quantitative investment to carry the financial model and multiple factors(such as stocks of financial factor,price factor,etc.)model to predict revenue is given priority to,and then rely on the continuous improvement of computer performance,in the field of machine learning algorithms are quantitative investment play a more important role.It can through the historical data training automation portfolio,help investors to make scientific decisions.At present,existing researches include LSTM,SVM,random forest,XGBoost and their corresponding improved algorithms.In addition,in terms of the selection of factors,with the deepening of quantitative investment research,many researchers based on traditional multi-factor investment find that it is increasingly difficult to predict investment through traditional factors,and it is difficult to obtain excess returns.Therefore,many researchers begin to embrace industry fundamental factors.That is,the factors that can reflect the current situation of the whole industry,such as sales,output,inventory,etc.,are expected to make a combination strategy to obtain higher returns in the secondary market.However,in terms of practical research and operation,the mapping of industry factors to individual stocks and how to deal with the different frequencies of industry factors and individual stock factors are all points that can be studied and innovated in detail.Among the various sectors of the secondary market,the new energy sector is hot and has become the "growth engine" in recent years.New energy refers to green energy that replaces energy transmission.Developing new energy industry vigorously is an inevitable choice for China to achieve "carbon neutral and carbon peak",and is also a key link for China’s industrial upgrading.Therefore,this paper selects the new energy vehicle industry,and 296 stocks are selected from more than 400 stocks in the new energy vehicle sector in flush according to the actual business occurrence and planning as the research object of this paper for investment.The design of investment strategy in this paper is mainly divided into two stages.One is to select special industry factors to predict the profitability of each end of the industrial chain through the fundamental research of the new energy industry.First,build a new energy vehicle industry stock pool according to the business scope of the new energy vehicle industry.Then,according to the industrial chain theory,divide the enterprises in the stock pool into upstream enterprises,midstream enterprises and downstream enterprises.Select two representative industry factors in the upstream,midstream and downstream respectively.Also obtain the average stock return of midstream enterprises and downstream enterprises in a rolling three-month interval.Considering that the average rate of return is the quarterly frequency and the industry factor is the monthly frequency,this paper selects the mixed frequency sampling regression model(MIDAS)to use the upstream,middle and downstream industry factors as the average rate of return for 17 periods from 2018 to the end of 2021 of their respective industries.The research on each end of the industrial chain in the first stage above is to adopt different quota proportions for the stocks of the three industry ends according to the level of the return rate in the later stage.Second,the construction of multi factor stock selection model.In this stage,the individual stock factors of the new energy vehicle industry are selected,that is,the sub variables under indicators such as profitability,operating capacity,and valuation that can reflect the company’s financial and operating conditions are taken as variables after data preprocessing.At the same time,the quarterly returns of individual stocks are divided into four categories from high to low.With the four category labels as the target,XGBoost algorithm is used to forecast the stocks in the optional stock pool from 2018 to the end of 2021.After the preliminary establishment of the model,select the combination of parameters that best meet the needs of the model by adjusting the parameters,and select the stock with the highest score of each section from the results of the running model composed of this final parameter for strategic back testing.Of course,in terms of the proportion of capital allocation,reference should be made to the industrial end forecast in the first stage.Finally,RSRS timing strategy is added to optimize the above strategies.Finally,the combination strategy of stock selection and timing is superior to single stock selection strategy,with annual return of 48.22% and excess return of 277.36%.