Machine Learning Algorithms for Predicting ETF Directional Movements Using Technical Indicators and VIX

Abstract

Predicting exchange-traded funds (ETFs) is challenging due to their diversified portfolios, exposure to market volatility, and short-term noise. This study compares eight machine learning models—Logistic Regression (LR), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Naïve Bayes (NB)—for forecasting the direction of daily ETF returns. Two horizons, next-day (T+1) and five-day (T+5), are examined. The analysis also tests the incremental value of incorporating the CBOE Volatility Index (VIX). Results show that performance varies across ETFs and horizons: linear models and LSTM perform best on large-cap indices (SPY, DIA), while RF leads for the small-cap index (IWM). At T+1, top models reach 56–58% accuracy; at T+5, LSTM improves markedly to 64.5% on SPY and remains strongest on QQQ and DIA, while most other models decline. Overall, horizon effects are model-specific rather than uniformly positive, and adding VIX provides only marginal, statistically insignificant gains (<2%). The findings suggest that model selection should depend on ETF and horizon, while broad volatility indicators such as VIX offer limited value for short-term forecasts.