Machine Learning-Enhanced Momentum Trading Strategy

Keywords

Abstract

This paper develops a machine learning-enhanced momentum trading strategy that achieves a 19.94% annualized return over 13.5 years (2011-2024). The strategy employs an ensemble of Ridge Regression, Random Forest, and XGBoost models trained on 25 engineered features across multiple time horizons. By combining traditional momentum signals with ML predictions, the approach demonstrates significant alpha generation (+6.7% over baseline) with a Sharpe ratio of 0.83. The framework incorporates volatility targeting, dynamic position sizing, and robust risk management to maintain consistent performance across varying market regimes. Backtesting results show superior risk-adjusted returns compared to traditional momentum strategies, particularly during periods of high market volatility. The paper provides detailed analysis of feature importance, model performance metrics, and transaction cost considerations.

Full Paper

Tip: Use your browser's zoom controls or the PDF controls to adjust the view. You can also scroll within the document or swipe on mobile devices.