Machine Learning-Based Adaptive Time Series Momentum Strategies in Equity Index Futures: A Comparative Analysis Between S&P 500 and CSI 300 Futures Markets

This paper employs machine learning techniques based on market volatility to identify and construct trading signals for both short-term and long-term Time Series Momentum (TSM) strategies. Through a comparative study of China's CSI 300 Index and the U.S. S&P 500 Index, we conduct an empirical analysis from a cross-market perspective. The findings reveal that the performance of time series momentum strategies is jointly determined by their signal responsiveness and the prevailing market volatility regime. Using the Random Forest algorithm, this study effectively identifies critical thresholds for regime switching between low-volatility and high-volatility states in index futures markets. The empirical results demonstrate that during high-volatility periods, short-term TSM strategies significantly outperform their long-term counterparts, whereas the opposite holds true in low-volatility environments. Further analysis indicates that the short-term momentum alpha can be attributed to market timing ability. Our findings provide important theoretical and practical implications for optimizing trend-following strategies in commodity and financial futures markets through machine learning approaches.