Introduction

Momentum strategies—buying recent winners, selling recent losers—are popular but suffer from reversals: momentum crashes when sentiment shifts. Filtering momentum signals by sentiment (ignoring momentum in stocks with deteriorating sentiment) improves strategy robustness. Machine learning adapts sentiment filters dynamically as sentiment regimes evolve.

Sentiment-Filtered Momentum

Traditional momentum: buy stocks with highest 252-day returns. Sentiment-filtered momentum: buy stocks with highest returns AND improving sentiment. Use NLP sentiment scores from news and social media. Filter out momentum in stocks with declining sentiment (red flags for reversals). Backtest on 2015–2024 data: filtered strategy has higher Sharpe ratios and lower drawdowns.

Adaptive Filtering

Instead of fixed sentiment thresholds, use machine learning to optimize which sentiment measures matter most in current regime. In bull markets, sentiment filters add less value; in bear markets (sentiment divergences common), filters add significant value. Adapt filter sensitivity based on market regime.

Conclusion

Adaptive sentiment filtering improves momentum strategy robustness and out-of-sample performance by avoiding sentiment-driven reversals.