LSTMNetworks versus ARIMA Models for Stock Price Prediction: A Case Study

Abstract

There is no doubt that economy is of central importance at the present time. A strong economy is considered a fundamental asset of the strength of countries. Therefore, analyzing and predicting economy-related time series is necessary. In this paper, we use artificial neural networks, which are considered one of the most successful machine learning techniques, in the field of stock market price prediction. Long short-term memory (LSTM) network is chosen for this task as it is characterized by the ability to retain information for a long period of time and eliminate unimportant information, which reduces the effect of fading that traditional neural networks suffer from. Autoregressive integrated moving average (ARIMA) model has the ability to deal with temporal data that contains upward or downward trends, which makes it useful in analyzing financial data that are affected by temporal factors. However, as indicated in our case study of TESLA daily stock prices in the period from 2010 to 2020, LSTMnetwork-based predictions are more accurate than those obtained using the fitted ARIMA model.