Empirical Research on Daily Direction Prediction of S&P 500 ETF Based on Machine Learning and Multivariate Technical Indicators

Abstract

This paper examines whether machine learning models can predict the next-day direction of SPY, an exchange-traded fund that tracks the S&P 500 Index. Using daily market data from 2010 to 2026, the study constructs 21 technical and cross-asset features, including returns, moving-average ratios, volatility, momentum, RSI, MACD, volume changes, and related market signals. Four supervised learning models are compared: Logistic Regression, Random Forest, XGBoost, and Support Vector Machine. The results show that XGBoost achieves the highest test accuracy among the tested models, with an accuracy of 0.5360 and a ROC-AUC of 0.5393. However, the model does not outperform the majority-class baseline accuracy of 0.5690. These findings suggest that technical indicators contain limited predictive information for next-day SPY direction, but their business value is weak as a standalone trading strategy. The model is more appropriate as a market monitoring or risk-awareness tool than as an automatic investment system.