Year: 2025
Type: preprint
Abstract: This study evaluates the effectiveness of the Seasonal AutoRegressive Integrated Moving Average with eXogenous regressors (SARIMAX) model for forecasting the daily closing prices of the S&P 500 index. The model incorporates a set of exogenous variables, including commodity prices, U.S. Treasury yields, cryptocurrency prices, major stock indices, and unemployment rates to capture key macroeconomic signals and investor sentiment influencing market dynamics. A rolling origin backtesting framework is employed to assess predictive performance over a 27-month period spanning January 2023 to March 2025. Empirical results demonstrate that the SARI-MAX model achieves high predictive accuracy during periods of market stability, with R 2 values exceeding 0.80 in multiple intervals. However, performance declines during periods of heightened market volatility, particularly during mid-2023 and early 2024. These findings underscore the importance of incorporating diverse exogenous indicators and applying stationarity transformations, while also highlighting the limitations of linear models under volatile market conditions. The results suggest that SARIMAX provides a strong baseline for financial time series forecasting.
Authors: Rasmus Erlemann, Vincent Liantonio, Soumya Banerjee, Noah Rahman, Michael William Porto
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W4413963416