Comparison between autoregressive integrated moving average and long short term memory models for stock price prediction

Pi Rey Low (Carnegie Mellon University)
Eric Sakk (Morgan State University)

IAES International Journal of Artificial Intelligence (IJ-AI)
Vol 12, No 4: December 2023

This study compares the forecasting accuracy in stock price prediction of twowidely established models - a more traditional autoregressive integratedmoving average (ARIMA) model and a deep learning network, the long shortterm memory (LSTM) model. They perform exceptionally well in time series data analysis and are applied to ten different stock tickers, comprising exchange-traded funds (ETFs) from different market sectors for the purpose of this study. The parameters in both models were optimised and this process revealed several differences from existing literature with regards to the optimal combination of parameters in both models. Upon comparing their performances, despite being more accurate when making point predictions, the ARIMA was outperformed significantly by LSTMs in terms of long-term predictions. Point predictions made by ARIMA were found to have similar accuracies as the long-run predictions made by LSTMs.