TLDR: A new study uses a Bayesian model to analyze Commodity Trading Advisor (CTA) returns, finding that while short-term trends alone underperform, combining them with raw market exposure significantly boosts risk-adjusted performance and drawdown protection, outperforming traditional long-term trend strategies.
Commodity Trading Advisors (CTAs) are investment managers who use systematic strategies, often relying on trend-following rules. These rules can operate across various timeframes, from capturing major, long-term directional market moves to identifying quick, short-term momentum signals in fast-moving markets. Despite extensive research on trend following, the specific benefits and interactions of short-term versus long-term trend systems have remained a subject of debate.
A recent research paper, titled “Re-evaluating Short- and Long-Term Trend Factors in CTA Replication: A Bayesian Graphical Approach,” delves into this controversy. Authored by Eric Benhamou, Jean-Jacques Ohana, Alban Etienne, Béatrice Guez, Ethan Setrouk, and Thomas Jacquot, the study offers a fresh perspective by dynamically breaking down CTA returns. It uses a sophisticated Bayesian graphical model to separate returns into short-term trend, long-term trend, and market beta factors. Furthermore, the paper demonstrates how blending these different time horizons can significantly influence a strategy’s risk-adjusted performance.
Understanding the Approach
The researchers aimed to answer several key questions: Can CTA-style strategies be replicated using only liquid futures? How do short- and long-term trend factor risk-returns compare over the past decade? Does mixing horizons improve drawdowns and diversification? And are short-term trends truly valuable, or should investors rely solely on longer trend horizons?
To address these questions, the study introduced a novel approach. They engineered “horizon-specific trend factors” using a concept inspired by a “lookback straddle,” which essentially captures the maximum price excursion over a period. For short-term trends, they used lookback windows of 10, 20, 40, and 60 days, designed to capture rapid breakouts and reversals. For long-term trends, a 500-day window was employed, reflecting classic two-year trend followers. In addition to these trend factors, they included a “market beta” factor, which is simply the raw daily return of each contract, representing baseline market exposure.
The core of their methodology is a Bayesian graphical model. This model acts like an advanced filter, simultaneously tracking how much a CTA strategy is exposed to short-term trends, long-term trends, and overall market movements at the individual contract level. This dynamic approach allows for a more nuanced understanding of how different factors contribute to CTA performance over time.
Also Read:
- Detecting Anomalies in Bitcoin Trading: A Look at Statistical vs. Machine Learning Models
- Understanding Market Swings: A New Model for Bull-Bear Dynamics
Key Findings and Implications
The empirical results of the study provide compelling insights:
-
Short-Term Trend (STT) Value: While the short-term trend factor on its own showed lower risk-adjusted returns (Sharpe ratio of 0.20), it proved to be crucial for diversification. Its convex payoff profile helps cushion portfolios during sudden market swings and allows for quicker adaptation to changing market conditions, acting as a form of “insurance” against regime shifts.
-
Power of Horizon Blending: The research confirmed that combining short-term and long-term trends (STT+LTT) is significantly more efficient than relying on a single horizon. A simple 50/50 allocation between STT and LTT notably improved the Sharpe/Max Drawdown efficiency ratio, leading to smoother equity curves and smaller peak-to-trough losses.
-
Market + Short-Term Trend Dominance: Perhaps the most significant finding was that pairing raw market returns with the short-term trend factor (MKT+STT) yielded the strongest overall performance. This combination achieved the highest Sharpe ratio (0.49), the best Return/Max Drawdown (0.48), and the shallowest maximum drawdown (14.9%). This dominance was consistent across various evaluation periods, including recent live market conditions and longer back-tests.
The study concludes that short-horizon trend signals, despite sometimes causing “false starts” or whipsaws, offer a critical asymmetric payoff that balances the more linear risks of long-horizon rules. The Bayesian filter developed in this paper provides portfolio managers with a real-time view of how each trend horizon contributes to returns and drawdowns, enabling more informed allocation decisions. This research highlights that a blend of short-term trend signals and raw market exposure is not just a recent phenomenon but a consistent source of convex returns and drawdown protection over time.
For more detailed information, you can access the full research paper here.
