Planning Of The Shanghai Stock Exchange Index Up And Down Forecasting Scheme Based On The Voting Fusion Algorithm

In recent years,with the rise of data mining and machine learning,some quantitative investment fields have begun to integrate investment strategies into machine learning methods,and achieved good research results.On the one hand,the advantage of using these methods overcomes the defect that the traditional method can’t consider many factors at the same time.On the other hand,thanks to the extensive use of a number of complex models represented by neural networks,support vector machines and combinatorial lifting algorithms,it is possible for researchers to better fit the chaotic trend of stocks.Recently,due to the impact of trade war,China’s stock market plummeted,and the whole securities market was in a low mood.The daily trading volume of the stock market also hit new lows.Since the us government announced that it was going to tax Chinese goods,the Shanghai composite index has been falling from about 3,300 in the early stage to around 2,600 so far,with a large number of listed companies’ stock prices falling below the issue price and investors suffering heavy losses.If we can effectively judge the future trend of the stock market,then we can reduce the loss of investors to a certain extent and avoid the risk caused by investors’ irrational investment sentiment and behavior.Based on this,this paper uses 55 technical factors of Shanghai composite index in the past five years from 2014 to 2018.At first,six different machine learning methods,including Logit regression,SVM,Naive Bayes,random forest,neural network and XGBoost,are used to predict the rise and fall of the index on the second day.Then the combination of optimal fusion models under different fusion methods of Voting and Stacking was explored.The results showed that both of the two fusion methods could improve the prediction accuracy of the single model to a certain extent.But the overall effect of Voting fusion is stronger than Stacking fusion.In the hard voting mode,the combination of the best fusion model is the fusion of SVM,neural network and XGBoost.In the soft voting mode,the combination of the best fusion model is the combination of Logit regression,random forest and neural network.In Stacking fusion,the accuracy of fusion model obtained by taking naive bayes and neural network as the input model of the first layer and Logit regression as the output model of the second layer is higher than that of single model.Finally,this paper constructs the trading strategy and carries on the backtest using the model forecast situation.The results show that the fusion model not only improves the overall prediction accuracy of the model,but also improves the overall profitability of the model.