| Due to the non-stationarity and nonlinearity of stock prices,stock price forecasting has always been a big problem.Fundamental analysis,sentiment analysis and technical analysis are the most commonly used index extraction methods in the field of stock price forecasting.Fundamental analysis requires the selection of many indicators within a few months or years,involving macroeconomic policy factors,microscopic market structure factors,and factors such as the company’s internal operating conditions.Emotional analysis obtains text information on social platforms through web crawler technology,and integrates the acquired text information into an indicator that reflects the degree of emotion.Technical analysis believes that all factors affecting stock prices will eventually be reflected in prices through transactions,so technical analysis focuses on using historical prices and trading volumes to predict future price trends.This article attempts to perform technical analysis of key stock indices and constituent stocks,which can help investors identify current trends and risks in stock prices.This research is only based on the second-hand data collected from the JoinQuant quantitative trading platform and Tushare Pro.In order to make the model prediction effect more convincing,the selected stock indices are Chinese mainland stock indices and 6 international stock indices,8 constituent stocks of the CSI 300 Index and 12 constituent stocks of the S&P 500 Index.The time span of the selected transaction data is from July 1,2016 to December 31,2021,and the time interval is 1 day and 15 minutes.The trading indicators include the opening price,closing price,highest price,lowest price,and trading volume of the stock..In this paper,we design an ICA-GRU model that optimizes price forecasting,first using the independent component analysis method to reduce the dimensionality of the above five indicators and extract independent components.Afterwards,the stock price is predicted by a long short-term memory neural network(LSTM)and a gated recurrent unit neural network(GRU).In addition,the effect of using two kinds of neural networks for stock price prediction under the same parameter structure is also compared.Our experiments show that(1)both composite models can effectively predict stock prices,(2)after comparing the experimental results of ICA-LSTM and ICA-GRU,it is found that the ICA-GRU model has a better prediction effect. |
