Recursive Forecasting Strategy for Stock Exchanges using LSTM-based and GRU Models

This research explores deep learning-based time series forecasting techniques to predict the S&P 500 Index tracking the stock performance of 500 leading companies listed on stock exchanges in the United States within one year or 365 days into the future. The study employs three recurrent neural network (RNN) architectures as Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU) using PyTorch. Historical S&P 500 index data is preprocessed and structured into supervised learning sequences, with each model trained to capture temporal dependencies and long-range patterns. Model performance is evaluated using metrics such as Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). The experimental results indicate that while all models are capable of learning the underlying market trends, BiLSTM outperforms the others in predictive accuracy due to its ability to leverage both forward and backward context in the data. These findings affirm the applicability of recurrent neural networks in financial forecasting tasks.