Modelagem e previsão de índice S&P 500 utilizando redes neurais recorrentes

Abstract

The present study aims to develop a predictive model for the S&P 500 index using Long Short-Term Memory (LSTM) Recurrent Neural Networks, due to their ability to capture long-term temporal patterns in financial time series. The index comprises the 500 largest publicly traded companies in the United States and serves as a key indicator of the performance of the U.S. stock market and, by extension, the global economy. It is widely used as a benchmark by institutional investors, fund managers, and economic analysts around the world. The methodology adopted in this work involved the collection and preprocessing of historical S&P 500 data, time series decomposition, and the modeling and forecasting of the index. The results demonstrate that the LSTM model was able to capture the dynamics of the S&P 500 index with reasonable accuracy, effectively tracking its trends and variations in the test set. The graphical comparison between actual and predicted values revealed strong alignment during periods of stability and growth, although performance declined during times of heightened volatility. The time series decomposition reinforced the presence of a long-term upward trend, stable seasonality, and a significant influence of macroeconomic events, highlighting the complexity of the series and the importance of considering its structural components in predictive modeling.