The Best Investment Writing Volume 5: Campbell Harvey and Michele Mazzoleni, Research Affiliates – Breaking Bad Trends

Episode Summary

Executive Summary: Campbell Harvey and Michele Mazzolini present a paper on “breaking bad trends” in momentum investing, arguing that trend-following suffers most when fast and slow signals disagree at turning points. Using 55 futures, bond, commodity, equity, and FX markets, they show turning points predict weaker static trend performance and explain recent underperformance. A dynamic strategy that adapts to corrections and rebounds improves results, especially after trend breaks.

Main Topics: Why trend following struggles at turning points (Priority: 5/5): The speakers explain the trade-off between slow lookbacks that miss reversals and fast lookbacks that react to noise. Turning points are the Achilles heel of static momentum strategies because they often lead to bad bets when trends break. Definition and measurement of turning points (Priority: 5/5): Turning points are defined as months when slow and fast momentum signals disagree. The paper counts these events annually by asset and links them to subsequent trend-following performance across a broad cross-section of markets. Evidence that turning points hurt static trend performance (Priority: 5/5): Across individual assets and multi-asset portfolios, more turning points are associated with lower risk-adjusted returns. The relationship is distinct from volatility and economically large. Why recent trend performance has worsened (Priority: 4/5): The authors argue that the increase in turning points in recent years helps explain why monthly trend-following has underperformed relative to prior decades. Dynamic trend-following strategy (Priority: 5/5): They propose a real-time implementable strategy that classifies markets into bull, bear, correction, and rebound states, then blends slow and fast signals differently after turning points to improve outcomes. Comparison with static blends and moving-average approaches (Priority: 3/5): The paper distinguishes its approach from simple static signal blends and moving-average crossovers, arguing the dynamic method uses observable state information to adjust exposure more intelligently.

Key Arguments: Trend following works well in sustained uptrends or downtrends, but breaks down when trends reverse. Long lookback windows are slower and can miss turning points; short windows react faster but are noisier. Turning points are observable from past returns and can therefore be used in real time to guide portfolio adjustments. The number of turning points is negatively related to risk-adjusted trend-following performance across many asset classes. This relationship is not explained by return volatility; turning points carry separate information. Recent years have had more turning points, which helps explain weaker recent trend-following returns. A dynamic strategy that shifts weight between slow and fast signals after corrections/rebounds can recover some of the losses from static trend following. The authors’ method is designed to be implementable ex ante, not merely a backtest. Static blends of momentum signals help, but they do not match the dynamic strategy’s ability to exploit post-turning-point returns.

Data Points: Number of markets: 55 - Futures, forward, and swap markets across equities, bonds, commodities, and currencies Asset classes: 4 - Equity indices, bond markets, commodities, and currency pairs Equity indices: 12 - Included in the sample universe Bond markets: 10 - Included in the sample universe Commodities: 24 - Included in the sample universe Currency pairs: 9 - Included in the sample universe Sample start: January 1971 - Beginning of available data for some assets Sample end: December 2019 - End of the data sample Static lookback window: 12 months - Baseline trend-following signal Fast lookback example: 1 month or 2 months - Used to illustrate and implement faster momentum signals Multi-asset portfolio volatility: 10% annualized - Normalized volatility level used in reported portfolio comparisons Annual return impact: -9.2 percentage points - One-standard-deviation increase in average turning points lowers annual portfolio return by about this amount R-squared: 0.72 - Fit of the relationship between turning points and annual multi-asset trend returns Slope: -0.21 - Downward fitted relationship between average turning points and portfolio returns Static trend return over 30-year period: ~7.5% annualized - Average return for the multi-asset static trend portfolio Static trend return in most recent decade: ~1.8% annualized - Recent-decade average for the static trend portfolio Dynamic trend return in most recent decade: ~4.3% annualized - Recent-decade average for the dynamic trend portfolio Turning-point frequency in recent decade: 6 of the most recent 10 years in top one-third - Recent years rank high in average turning points Total asset-year observations: 1,561 - Used in Exhibit 1 analysis of individual assets High turning-point threshold: 6+ turning points per year - Typical static trend returns become negative Very high turning-point threshold: 8+ turning points per year - Most static trend returns are negative; average Sharpe below -1

Pivotal Quotes: "the Achilles heel of trend investing" — Campbell Harvey: Describing why turning points are a central weakness of trend-following strategies "turning points can help explain the deterioration of trend-following performance in the more recent years" — Campbell Harvey: Summarizing the paper’s explanation for weaker recent performance "dynamic trend following can harvest returns after turning points returns that might have been lost under standard trend following" — Campbell Harvey: Conclusion describing the benefit of the proposed strategy

Implications: For investors, momentum need not be purely static. Monitoring disagreement between fast and slow signals may improve trend strategies, especially in choppy markets with frequent reversals. For the industry, dynamic overlays could help reduce recent underperformance of CTA-style trend systems.