Stock Price Prediction Using ARIMA and LSTM Models: An Application to CSI 300 Closing Prices

Abstract

In the development of the real economy, the performance of the stock market can well reflect its development status, so it is of great value to select effective forecasting methods to make a relatively accurate prediction of stock prices. The research and forecast object of this paper is the daily adjusted closing price of CSI 300 index. The sample data is derived from China Securities Data Website and is composed of six characteristic data for 261 days from July 1, 2022 to July 26, 2023. This paper compares the prediction ability of Autoregressive Integrated Moving Average (ARIMA) model and Long Short Term Memory (LSTM) networks. For the deviation function, this paper chooses MAE and RMSE two methods. By comparing the prediction ability of the models, the following conclusions are drawn: The daily closing price of CSI 300 has autocorrelation and long-term memory, which enables the prediction of future price based on historical data. Compared with ARIMA model, LSTM has stronger prediction ability, but the prediction has a lag, which is not conducive to reflecting the drastic trend changes in the short term. More characteristic information needs to be added to characterize to obtain more accurate results.