FURI | Summer 2025

Comparing LSTMs and GRUs with and without Reinforcement Learning for Stock Price Prediction and Trading Decisions

Gated recurrent units (GRUs) and long short-term memory (LSTM) networks are two popular recurrent neural network architectures used for time-series forecasting in financial markets. This project implements both GRU and LSTM models, with and without reinforcement learning (RL), to evaluate their effectiveness in prediction accuracy and simulated trading outcomes. Using historical S&P 500 data from 2000 to 2025, the models will be assessed based on prediction metrics and backtested trading performance. Findings will inform whether GRUs provide advantages over LSTMs and how RL influences trading decision outcomes, contributing to advancements in quantitative finance.