Study On The Forecast Of The Closing Price Of China's Main Stock Index

With the outbreak of COVID-19 in 2019 and the global economic downturn,China becomes the first major economy with positive economic growth in 2020,and its GDP will also achieve a historic breakthrough of 100 trillion yuan.It is against this background that the movements of the Chinese stock market are not only influenced by their own factors,but also closely related to the stock markets of other countries.In particular,trade frictions between China and the United States have intensified since 2018.All of these events will have an impact on the stock market,and in the new environment,it is inevitable that forecasts for China's stock market will take the state of the U.S.stock market into account.The main stock index is the representative of a country's stock market.This paper takes the rise and fall of the closing prices of four major stock indexes in China as the research object.Specifically,the following work is carried out:(1)Through the Granger causality test of the closing price data,it is concluded that the three major stock indexes in the United States are the Granger causality of the major stock indexes in China.(2)According to the pre-processing of basic stock data,the closing price of the stock index to be predicted and the closing price of several selected U.S.stock indexes are taken as the data of the day minus the previous day,and the difference of the closing price minus the opening price as the characteristic variable,get the raw data.When using the supervised learning method,we further processed and symbolized all variables,and comprehensively considered whether the historical data of closing spread and trading volume were large enough in the research stage to obtain new variables.(3)The importance of selecting the output variables of random forest algorithm and XGBoost algorithm is further explained,and the importance of the major stock index in the United States as the characteristic variable is further explained.By comparing the SVM with random forest and XGBoost,the evaluation indexes of the training set and the test set are calculated respectively.By comparing the prediction effects of different stock indexes under the same method and different methods under the same stock index,it is concluded that the XGBoost algorithm has the best performance under various evaluation indexes comprehensively,which can provide some reference for investors.In the24 experiments,XGBoost algorithm has the best simulation effect on the training data of Taiwan Weighted Index,with the highest accuracy of0.6764,corresponding F1 value of 0.7062,and Matthews correlation coefficient of 0.3462.(4)Using deep learning LSTM method: we adopted after processing the original data,through to the closing price or fall predict the value of the corresponding forecasting results are obtained,and according to the predicted values of positive and negative and the actual value of the positive and negative contrast to judge or predict whether accurate,the results show that the accuracy is not high,especially for the hang seng index.In conclusion,according to the experiments we designed,the XGBoost algorithm is the most suitable.