Quantitative Stock Selection Model Based On XGBoost

Quantitative investment,after a long time of development,has become a popular idea and mature method of investment in foreign markets.In China,though the concept of quantitative investment was introduced into the securities market not long ago,it is enjoying a fast development.At the same time,Quantamental Investing has gradually become a mainstream investment philosophy,marking a professional development trend toward maturity on the home market.In this changing era of market and ideas,Quantamental Investing is sure to have a prospect of further development.Referring to"Value Strategies Based on Machine Learning" written by J.P.Morgan's team and published in August 2017,this paper proposes a statistical modeling method to predict the company's price-to-book ratios by using multi-dimensional stock characteristic factors.The research first tests the single factor efficiency of the price-to-book ratio in the A-share market to demonstrate the rationality of the model.The next step is to predict the price-to-book ratio in next period by using the XGBoost algorithm.Then a"fair value" index is established according to the predicted ratio.An assessment is made on the component stocks of the Shanghai and Shenzhen 300 through the index.The component stocks are categorized into high group and low group.Then the high group is supposed to be going long on and the low group,going short on.Eventually an investment portfolio is constructed with equal proportion weight.The annual rate of return outside the samples is 38%,higher than the benchmark rate of return of the Shanghai and Shenzhen 300 index in the same period.In this paper,a hybrid model of XGBoost + Lasso + SVM is innovatively used to predict the model.Compared with the original strategy,the improved model strategy is obviously enhanced.In this paper,an event-driven fundamental-quantitative model is constructed based on the improved event-driven effect model.Performance forecast is the first-hand information for listed companies to release on the market,which has a leading effect on the stock prices.In this paper,an empirical verification is conducted on the positive and negative events in the performance forecast,after which,the stocks with negative events in the portfolio are excluded according to the event definition.The rate of return of the improved investment portfolio is slightly higher than the original strategy.Finally,this paper gives prospects for the value stock selection based on machine learning.After empirical analysis,a tentative idea is proposed to calculate a new "fair value" index according to the P/E ratio.