Stock Price Prediction Based On Time Series And Neural Network Model

Stock price forecasting has always been an issue of concern for stock investors.In reality,it would be very meaningful to use the empirical analysis method to select a suitable model for stock movement forecasting from the many forecasting models.This is because a reasonable estimation of stock market movements through forecasting can minimize the negative impact caused by stock market fluctuations,help financial regulators monitor the investment market,and assist the government in formulating monetary policies,thus effectively promoting the economic development of China.The stocks used for the study in this paper are the 50 constituent stocks of the SSE 50 index,and 1500 trading days of data between December 08,2016 and February 10,2023 are selected for each stock for the study,using traditional financial time series GARCH models and RNN,LSTM,GRU,Bi-LSTM,Bi-GRU ConvLSTM,and TCN,which are seven neural network models,combined with five stock indicators,namely,total market capitalization,turnover ratio,P/E ratio,P/N ratio,and P/S ratio,for each stock on each working day to train and predict the daily closing price of each stock.In this paper,we first use the GARCH model to predict ICBC(the largest of the 50 constituent stocks in terms of market capitalization)to affirm the feasibility of the model in stock prediction.Then seven neural network models are used to forecast stock prices,and the loss function MAE and MSE are used to analyze the prediction results of each model.Through comparative analysis,it is found that the overall prediction effect of RNN model is better.However,it was found that these seven neural network models had poor prediction results for all four stocks in the prediction process.So this paper tries to combine the traditional time series GARCH model with the RNN neural network model to construct a GARCH-RNN model to further predict these four stocks.