How to predict stock prices of trillion dollar club companies using deep learning analysis? Academic Article in Scopus uri icon

abstract

  • Recurrent Neural Networks (RNNs) have led to a greater global focus on stock price prediction. This research explores the combination of technical analysis indicators and RNNs to forecast stock prices for NVIDIA, a key player in the Artificial Intelligence (AI) revolution. Surpassing Microsoft and Apple in 2024, NVIDIA became the world's most valuable company, driven by demand for its Graphic Processing Units (GPUs) in the AI space. After joining the trillion-dollar club in May 2023, a group with other companies whose market cap is above the trillion-dollar threshold, the company added more than two trillion dollars in market value, solidifying its leadership in the industry. This research applies Deep Learning (DL) algorithms to predict daily closing stock prices, incorporating financial indicators such as simple moving averages, Bollinger bands, moving average convergence divergence, and exponential moving averages as key predictors. The framework of this article includes data processing, feature selection, and Principal Component Analysis (PCA). This work compares the effectiveness of three DL algorithms: i) Long-Short-Term Memory (LSTM) networks, ii) Gated Recurrent Units (GRU), and iii) a hybrid LSTM-GRU architecture that delivered the best performance in the coefficient of determination (R2) and the mean absolute error (MAE). Optuna® facilitates hyperparameter adjustment, enhancing model performance. The findings indicate that the LSTM-GRU model produces the best performance, with a value of 0.8, improving to 48.15% in the R2 test and 31.15% in the MAE compared to the same model without technical indicators. The proposed approach demonstrates that combining technical indicators with recurrent neural networks enhances the model's effectiveness in stock market forecasting practices. © 2025 IISE Annual Conference and Expo 2025, Conference Proceedings. All rights reserved.

publication date

  • January 1, 2025