Study On Quantization Strategy Of New Shares Based On Machine Learning Algorithm Such As GBDT

In recent years,the field of quantitative investment has attracted wide attention and has become a very popular research direction.Using mathematics,statistics,machine learning and other technologies,quantitative investment can fully tap the laws of historical price evolution of various assets,and use these laws to form investment strategies and obtain high returns.Quantitative investment has been widely concerned by institutional investors and individual investors in recent years.Quantitative investment rose in the 1960 s,when with the study of measurement tools,investors began to try to quantify market prices.However,China's financial and related derivatives market started relatively late.Quantitative investment has been widely valued by the market since 2010.At present,there are relatively few investment and financial products related to quantification in the market.At the same time,due to the restriction of market trading rules,quantification investment strategies are not as rich as those in Europe and the United States and other countries.Therefore,the study of quantification investment strategies has a huge space for development in China's market.A-share market is a mature and large-scale market in China at present,so how to construct a quantitative investment strategy suitable for A-share market and use these strategies to guide investors has very important practical and theoretical significance.Since 2010,new shares have a more obvious premium issuance characteristics,after the listing of prices tend to rise sharply,so there are many strategies to hit new shares.However,restricted by circulation,new shares are often hard to find.At the same time,because of the big fluctuation of the new stock after listing,the price fluctuation of the new stock in a period of time is different from other stocks.Therefore,it is of great practical significance to study the price characteristics of the new stock and look for trading opportunities.Statistically speaking,after a boom in the early stage of the listing,new shares often appear in the process of price fall and fall back,and then rise again.And not all new stocks are equally likely to rise.After 30 trading days,more than 50% of new stock prices will fall.Therefore,how to construct a set of quantitative trading strategy for the sub-IPO,guide investors in the sub-IPO trading for scientific stock selection has become very important.In recent years,machine learning has developed rapidly.Many researches have focused on the application of machine learning in quantitative investment.Many algorithms of machine learning,such as Support Vector Machine(SVM),Gradient Boosting Decision Tree(GBDT),have a good effect on the prediction of nonlinear problems,and for the rise and fall of secondary new shares can be transformed into classification problems in machine learning.Therefore,this paper attempts to use SVM,GBDT and other machine learning models to predict the rise and fall of the next IPO,and build trading strategies.In this paper,quantitative research methods are used to extract market data of listed A shares from open databases such as Jukuan and Tushare from January 1,2006 to December 31,2017.Through statistical analysis,we find that the volatility of new shares is different from other stocks in the 30 trading days after listing.Through the single factor test,it is found that total equity,total market value,turnover rate,P/E ratio,P/E ratio,market-to-net ratio,market-to-market ratio and other factors have a significant relationship with the rise and fall of new shares in the 30 trading days after listing.Through the establishment of SVM and GBDT model to predict the IPO,GBDT forecast AUC to 0.91,SVM forecast AUC to 0.77,have a certain forecasting ability,and GBDT forecast ability is better than SVM model.Based on statistical analysis and modeling analysis,three trading strategies are constructed,including short-term callback trading strategy,sub-IPO multi-factor trading strategy and machine learning short-term trading strategy.The yield of short line callback trading strategy is 423.9%,which has a good effect.The yield of trading strategy based on sub-IPO factor is 14.7%.In order to compare the effect of multi-factors,the profitability of single factor is tested.It is found that the effect of single factor is relatively poor if independent factor is used to select stocks.However,the multi-factor model is used to select stocks comprehensively,and the annual return is nearly 15%.Promote.In the short message trading strategy based on machine learning,SVM and GBDT are used to make decisions respectively.The yield of trading strategy based on SVM is 121.36%,but the capital return curve fluctuates greatly,with the maximum withdrawal of 35.6%.The GBDT-based trading strategy has a smooth capital curve with a yield of 128.7% and a maximum withdrawal of 9.8%.Compared with the SVM-based trading strategy,the GBDT-based trading strategy has 7 percentage points higher returns,while the maximum withdrawal has decreased by nearly 26 percentage points.After modeling and testing the listed stock data of A-share market after 2010,the results show that GBDT and SVM model can predict the sub-IPOs to some extent,and the returns are stable.The method of this paper provides a way to further build quantitative strategies.