Stock price prediction using LSTM neural networks: Techniques and applications

Abstract

The prediction of stock prices has garnered significant attention due to the potential financial gains and complex questions involved. This paper elaborates a comparison between the Long Short-Term Memory (LSTM) model, optimised using the early-stopping method, and the conventional mathematical method Autoregressive Integrated Moving Average Model(ARIMA), which is conducted using the S&P500 from 2022, May 01 to 2024, May 01. The results indicate that the LSTM surpasses ARIMA. To be more specific, LSTM achieves a 92% reduction in error rates compared to ARIMA. In addition, when the optimised LSTM is implemented in 6 different stocks, the results indicate a negative correlation between the volatility and accuracy of the stocks. This study demonstrates the advantages of optimised LSTM for predicting stock prices, and the importance of market volatility as a crucial aspect that significantly impacts the accuracy of stock price forecast.