Deep Learning Methods for Stock Prediction: Comparative Analysis of LSTM, RNN, GRU and Optimizer Performance

Abstract

Stock price prediction is of the utmost importance for financial analysis as accurate forecasting can have a major impact on an investment strategy and risk management. In this work, we investigate the performance of LSTM, RNNs, and GRUs for stock price prediction by providing a comparative analysis of these models in predicting eBay Inc.’s stock price. In addition, we propose the use of recent optimizer algorithms like NAdam and AdamW to improve the models’ performance. The RNN improved from an R2 of 91.4% to 93.4% when using NAdam instead of Adam, showing the effectiveness of our method. We additionally show that GRU with the Adam optimizer algorithm consistently outperforms other models across all metrics.