A Strategy Application Research On Price Limit Up Prediction Of New Shares Within The Dav When It Fell Out Of The Limit Price Based Upon Machine Learning Classification Algorithm

The paper systematicially reviews the development of quantitative investment and the research findings upon the application of machine learning classification algorithm on quantitative investment at home and abroad;otherwise puts emphasis on the concepts and principles of the three supervised classification algorithms,namely,BP Neural Network,Random Forest and Support Vector Machine(SVM);then takes the phenomenon in China A-share market,where the new shares feature keeping at limit up price the whole day since the beginning of 2014;initially,the paper uses the three classification algorithms to predict that whether the price of new shares can limit up again within the day when it fell out of the limit line.After selecting samples of new shares which has achieved it after IPO as classification prediction targets,the paper takes good advantages of the stock market fundamentals analytical framework,and selects self-characteristic variables,such as the scale,valuation and profit,and external market variables in different dimensions and types like market heat,industry popularity and style transformation,then constructs seven sample input characteristics to proceed supervised training and parameter optimization for all the classification models;subsequently,filters optimal parameter models for three algorithms in this research context;accordingly builds an integral set of short-term trading strategies aiming for limiting up again after falling out of the limit line of new shares based on the three classification prediction models;in addition,conducts back testing and simulated trading testing in and out of the samples for new shares trading strategies in different dimensions of parameters.The main conclusions and new shares trading suggestions are as follows:1.From the perspective of trading strategies selection,it will probably result to a negative net value in the long run if making speculation trade directly to buy all the new shares that limit up without filtering;on the contrary,an effective filtering and chasing of new shares featuring limiting up again within the day when it fell out of the limit line,as the paper illustrates,will lead to a higher benefit-to-risk ratio.2.For any new shares,it would be more reasonable to seize a chance where the price is comparatively high to sell in 1-3 trading days instead of holding it for a long time.3.For the prediction performance of the three models,Random Forest,with the optimal parameter,owns the highest prediction accuracy of the three in the samples and beyond,therefore it characters the better overfitting-resistance;for the mistakes prediction type,Random Forest with the optimal parameter owns a relatively lower possibility of type I error than SVM and BP Neural Network,while S VM with the optimal parameter has a significant lower possibility of type ? error than Random Forest and BP Neural Network.4.For the seven characteristic variables chosen in the paper,"industry popularity before falling out of the limit line","index stock heat before falling out of the limit line","performance forecasting in the prospectus" and so on have higher degree of contribution in the variance of the prediction than others.