Predicting S&P 500 Closing Prices Using a Feedforward Neural Network: A Machine Learning Approach

Abstract

The Standard and Poor’s 500 index is a crucial benchmark for investors, financial analysts, and policymakers to assess stock market performance and make informed investment decisions. Accurate S&P 500 closing price prediction is vital for efficient portfolio management and risk mitigation. This study explores the application of neural networks, specifically the Feedforward Neural Network, to predict the S&P 500 closing price. The author used a dataset from 1950 to 2024, including variables such as open, high, low, close, and volume as the initial step. Various training-testing ratios are tested to evaluate the models' performance. The results highlight that an eighty to twenty split yields the best predictive accuracy, with the lowest Mean Absolute Error and Mean Absolute Percentage Error. Additionally, this study compares the Feedforward Neural Network's predictions to polynomial regression models and investigates cluster-wise fitting techniques to enhance accuracy. The findings demonstrate that neural networks can significantly improve the predictive power of S&P 500 closing prices, particularly when combined with advanced regression techniques, providing valuable insights for academic research and practical financial decisions.