When Momentum and Reversal Coexist: What Weekly Commodity Futures Data Reveals

When Momentum and Reversal Coexist: What Weekly Commodity Futures Data Reveals

The textbook view of return predictability is neatly segmented: returns reverse at short horizons (weekly to monthly) and exhibit momentum at intermediate horizons (3 to 12 months). This framework, synthesized across decades of equity market research from Lehmann (1990) through Jegadeesh and Titman (1993), has become canonical in asset pricing. But new evidence from commodity futures markets suggests this clean separation is incomplete.

Ding, Kang, Yu, and Zhao (2026) use a simple but powerful decomposition to show that momentum and reversal operate simultaneously at the weekly horizon. By separating commodity returns into a speculator-flow component and an orthogonal residual, they find that the flow component reverses (consistent with liquidity provision) while the residual component exhibits momentum (consistent with trend-chasing behavior). The practical implication: systematic commodity traders have been conflating two distinct signals into one noisy measure, and separating them produces a meaningfully better trading signal.

The Decomposition: Flows vs. Information

The key insight relies on CFTC Commitments of Traders (COT) data, which reports weekly positions of non-commercial speculators (a proxy for institutional trend followers) across 26 commodity futures markets from 1993 to 2025.

The authors construct a weekly net trading measure, Q, defined as the change in non-commercial net long positions scaled by open interest. They then regress weekly commodity returns on Q cross-sectionally and define the residual as R(nonQ), the component of returns orthogonal to speculator trading pressure.

This decomposition is conceptually clean: Q captures the price impact of speculative demand, which temporarily pushes prices away from fundamentals and subsequently reverses. R(nonQ) captures everything else, including information diffusion and the component of returns that attracts trend-following capital in subsequent weeks.

The Evidence: 6.2% Annualized from Weekly Momentum

The paper's central finding is striking in its magnitude. A one-standard-deviation increase in R(nonQ) in week t predicts an 11.6 basis point increase in returns in week t+1, translating to 6.2% annualized. This exceeds the unconditional average commodity return of 4.7% per year.

Several features of this momentum signal stand out for practitioners.

First, it applies to the entire cross-section of commodities, not just a subset with specific characteristics. Unlike equity-market short-term momentum, which Medhat and Schmeling (2022) find concentrated in high-turnover stocks, the commodity version is pervasive across metals, energy, agriculture, and livestock.

Second, the momentum effect strengthens when volatility is low and when the expected profitability of trend-following is high. This is consistent with Hong and Stein's (1999) model: when markets are calm, speculators are more confident in chasing trends, and their collective behavior generates return continuation.

Third, the R(nonQ) signal enhances traditional intermediate-term (3-12 month) momentum. A probit analysis shows that being a short-term R(nonQ) winner significantly increases the probability of being a winner over longer horizons. Aggregating weekly R(nonQ) signals into intermediate-term momentum construction substantially improves performance.

Why It Matters: Two Signals Hiding in One

The practical import of this research goes beyond academic interest. Most systematic commodity strategies use raw past returns as inputs for both momentum and mean-reversion signals. This paper demonstrates that raw returns conflate two economically distinct forces: a flow-driven reversal that reflects temporary price pressure, and an information-driven continuation that reflects trend-following capital deployment.

For commodity trading advisors (CTAs) and systematic macro funds, the implication is that COT data, updated weekly and freely available from the CFTC, contains actionable information for signal construction. The decomposition is not complex: regress returns on net flow changes, take the residual, and use it as a momentum signal. The reversal signal uses Q directly.

Regime Dependence and Signal Dynamics

The paper provides granular evidence on when the momentum signal is strongest. The trend-chasing behavior that drives R(nonQ) momentum intensifies under specific conditions.

The duration structure also matters. R(nonQ) positively predicts subsequent speculator trading flow for up to three weeks (t+1 through t+3), after which the trend-chasing effect unwinds. Beyond the short horizon, the mechanism shifts: R(nonQ) forecasts returns over the intermediate window (1-12 months) through what appears to be gradual information diffusion rather than trend-chasing.

Reframing Short-Term Reversal

The paper also reframes what "short-term reversal" actually means. In the equity literature, short-term reversal strategies are constructed using past returns as the signal. The authors show that in commodity markets, the reversal effect is more precisely characterized as trading-based reversal rather than return-based reversal.

When Q (past speculator flow) and past returns are both included as predictors, Q dominates. Past returns add little incremental predictive power once flow information is controlled for. This suggests that equity-market short-term reversal strategies, which lack high-frequency position data, may be using a noisy proxy for what is fundamentally a liquidity provision effect tied to order flow, not to past prices per se.

Limitations and Implementation Constraints

Several constraints limit real-world implementation.

COT data is released with a three-day lag (Tuesday positions, published Friday afternoon). This means the weekly momentum signal from R(nonQ) cannot be acted upon until the following Monday at earliest, introducing execution lag. The paper's backtests implicitly assume next-Tuesday execution, which is realistic given the lag structure.

Transaction costs in commodity futures are low relative to equities, but the weekly rebalancing frequency generates meaningful turnover. The 6.2% annualized signal is a gross figure that does not account for bid-ask spreads, slippage, or roll costs. Net returns will be lower, particularly for less liquid agricultural and livestock contracts.

Capacity is constrained by the size of commodity futures markets. The 26 commodities in the sample vary enormously in liquidity, from deep markets like crude oil and gold to thinner markets like oats and lumber. A realistic allocation would need to size positions according to market depth.

The sample period (1993-2025) covers both secular commodity bull markets (2000s) and extended bear markets (2014-2020). The paper does not conduct formal out-of-sample testing beyond sub-period robustness checks.

Practical Takeaways

The coexistence of momentum and reversal at the weekly horizon carries several analytical implications for systematic commodity traders.

The traditional framework of "reversal at short horizons, momentum at intermediate horizons" appears to be an oversimplification that arises from using raw returns as the sole predictor. When speculator flow is controlled for, both effects are visible at the same frequency.

COT data from the CFTC provides a publicly available, weekly-frequency source for decomposing commodity returns into flow and non-flow components. The decomposition is straightforward: the flow component (Q) serves as a reversal signal, while the orthogonal residual (R(nonQ)) serves as a momentum signal.

The momentum component of R(nonQ) has historically produced annualized returns of approximately 6.2% from a one-standard-deviation signal, which is economically meaningful relative to average commodity returns of approximately 4.7% per year. This signal has shown greater strength during low-volatility environments and periods when recent momentum strategies have been profitable.

Aggregating weekly R(nonQ) signals into intermediate-term momentum construction has historically improved the performance of traditional 3-12 month commodity momentum strategies. The short-term and intermediate-term momentum effects appear to be connected rather than independent phenomena.