Thesis
Dual-momentum is one of the oldest systematic ideas in quantitative finance: rank holdings by trailing return, own the leaders, exit when trend breaks. Headmars' implementation applies a 60-day lookback across a 24-name universe of large-cap U.S. equities spanning technology, financials, healthcare, consumer staples, energy, and industrials. The logic is deliberate — big, liquid names where momentum is academically documented, not a thin-market artifact.
Backtest Summary
Over 451 days ending in late May 2026, the strategy compounded $10,000 to $12,350 — a 23.5% total return and 12.5% annualised CAGR. The full-period Sharpe of 0.95 is respectable for an equity strategy without leverage. Maximum drawdown reached 15.67%, which is contained given the macro volatility the period included.
The number that demands attention is turnover: 2,638%. That is not a typo — it reflects 136 round-trips across 451 days, meaning the strategy rotates its entire notional book roughly every two weeks on average. Fee drag ($1/trade flat in this simulation) is modest at $136 total, but any live implementation with realistic bid/ask spread would erode returns meaningfully at this churn rate.
The win rate of 28.79% confirms the strategy's profile: it loses small frequently and wins large infrequently. That is textbook momentum behavior — you're buying after strength is already visible, so many entries get stopped out on reversals before the next leg. Sizing discipline and tight exit rules matter more here than entry timing.
Cross-Validation Folds
The four-fold walk-forward tells a nuanced story:
| Fold | Period | Return | Sharpe | Max DD | Trades |
|---|---|---|---|---|---|
| 1 | Aug 2024 – Jan 2025 | +5.85% | 1.23 | 3.92% | 20 |
| 2 | Jan 2025 – Jul 2025 | −7.31% | −1.05 | 17.15% | 60 |
| 3 | Jul 2025 – Dec 2025 | +25.59% | 3.32 | 4.04% | 16 |
| 4 | Dec 2025 – May 2026 | +13.34% | 2.15 | 7.43% | 42 |
Three of four folds are positive, and Fold 3 is exceptional — 25.6% in five months with a Sharpe above 3 and a drawdown under 5%. Fold 4 (the out-of-sample holdout) posted +13.34% and an OOS Sharpe of 2.15, which is the most forward-looking signal in the entire report.
Fold 2 is the outlier. Sixty trades, a −7.31% return, and a 17.15% drawdown suggest the strategy was caught in a choppy, rotation-heavy tape where trend signals whipsawed repeatedly. High-turnover momentum strategies are inherently vulnerable to sideways volatility; this fold is a clean stress test of that risk.
Validation Status: Failed
Despite the attractive headline numbers, dual-momentum did not pass Headmars' validation gate. The key metric is the Deflated Sharpe Ratio (DSR) of 0.476 — below the 0.50 threshold. The DSR penalises strategies for the number of parameter trials tested (6 here) and non-normal return distributions. In plain terms: with six configurations tried before settling on this one, there is insufficient statistical confidence that the 0.95 Sharpe isn't partially a selection artifact.
The Probabilistic Sharpe Ratio (PSR) of 0.893 is more encouraging — it says there is an 89% probability the true Sharpe exceeds a benchmark Sharpe of zero. That is meaningful signal. The DSR failure is a warning about the search process, not a verdict that the strategy is worthless.
Recent Activity
The strategy has been fully in cash since at least June 8. Six consecutive daily runs — June 8 through June 15 — each report zero executions and a $10,000 cash balance. The 60-day momentum screen is not surfacing conviction buys right now, which may reflect a period where no name in the universe has broken out cleanly above its peers. Flat-book behavior is correct and expected when the signal is absent.
Outlook
Dual-momentum earns a watchlist — not deploy rating in its current form. The out-of-sample Fold 4 and PSR both suggest genuine predictive content. But the DSR failure, the Fold 2 drawdown, and the elevated turnover all point to the same remediation: reduce the number of trials before retesting, consider a minimum holding period to dampen churn, and re-run validation after at least one additional fold of live data accumulates.