Turtle / Donchian Breakout
Donchian-channel breakout trend-following
Passes 9 of 11 gates. Profit factor 1.61, Sharpe 0.90, placebo PASS, deflated-Sharpe 0.997. It fails only on max drawdown (−49%) and instrument concentration (43%). A genuine trend edge — it survives, pending a risk re-scope before any capital.
- Category
- Trend
- Window
- 2014–2026 (12.5y)
- Instruments
- Diversified futures-style basket
- Timeframe
- Daily (Donchian breakout)
- Tested
- 2026-06-24
A real edge that fails on risk, not signal
The Turtle / Donchian-channel breakout — buy the N-day high, ride the trend, exit on the opposite channel — is the strongest survivor in the program to date. Over 12.5 years (2014–2026) it passes 9 of 11 pre-registered gates: profit factor 1.61, Sharpe 0.90, positive in 92% of calendar years, a deflated Sharpe of 0.997, and a placebo PASS (real PF 1.61 vs the random-breakout 95th percentile of 1.48 — only 0.8% of permutations beat it). The 2× cost stress still clears at PF 1.55.
It fails exactly the two gates that trend-following famously strains: max drawdown −49% (gate: ≤12%) and instrument concentration 43% (gate: ≤40%). Both are risk-shape failures, not edge failures — which is why the status is SURVIVING, not RETIRED.
Gate scorecard — 9 / 11
| # | Gate | Result | Pass |
|---|---|---|---|
| 1 | ≥ 100 trades | 590 | ✅ |
| 2 | PF ≥ 1.20 net | 1.61 | ✅ |
| 3 | Sharpe ≥ 0.6 | 0.90 | ✅ |
| 4 | Max DD ≤ 12% | −49.0% | ❌ |
| 5 | Positive in ≥ 60% of years | 92% | ✅ |
| 6 | Bootstrap 95% LB Sharpe > 0 | 0.58 | ✅ |
| 7 | Placebo: real PF > p95 | 1.61 vs 1.48 | ✅ |
| 8 | 2× cost stress PF > 1.0 | 1.55 | ✅ |
| 9 | Deflated Sharpe positive | DSR 0.997 | ✅ |
| 10 | No instrument > 40% of P/L | 43% | ❌ |
| 11 | Walk-forward OOS ≥ 0.9× IS | holds | ✅ |
What survival means here
This is the honest middle ground the framework exists to surface. The edge is statistically real and survives the kill-shot placebo and the deflated-Sharpe correction for multiple testing — the things that retire most strategies. What it does not yet have is a risk profile a small account can hold: a −49% drawdown will end most live deployments long before the trend pays off, and 43% of the P/L riding on one instrument is fragile.
The disciplined next step is a re-pre-registration with lower per-instrument risk and a broader basket, evaluated on a fresh window — not a post-hoc tweak to make this run pass. Survival earns a Round 2, not capital.
Verdict: SURVIVING (9/11). Real edge, wrong risk shape. The trend signal is genuine and placebo-confirmed; the −49% drawdown and 43% concentration must be re-scoped before this could clear all eleven gates and earn a paper allocation.
Charts & evidence
Frequently asked
Does the Turtle trading system still work in 2026?
The signal still carries a real edge. Run over 2014–2026 through a pre-registered 11-gate battery, the Turtle / Donchian-channel breakout system posts a profit factor of 1.61, a Sharpe of 0.90, and a deflated Sharpe of 0.997, and it beats a random-breakout placebo. It passes 9 of 11 gates — the two failures are a −49% max drawdown (the classic trend-following deep-drawdown profile) and 43% profit concentration in one instrument. It is a survivor under evaluation, not yet a deployment-cleared strategy.
Why is the Turtle system not marked Validated despite passing 9 of 11 gates?
Because two of the gates it fails are real risk constraints, not technicalities. A −49% max drawdown exceeds the 12% gate by a wide margin, and 43% of profit comes from a single instrument (the 40% concentration limit). The trend edge is statistically genuine — placebo-confirmed, deflated-Sharpe positive — but the risk profile needs to be re-scoped (lower per-instrument risk, broader basket) before it could clear all eleven and earn capital.
Methodology: The Validation Gauntlet — pre-registered spec, 11-gate battery, real market data.
Full reproducible report: backtests/turtle/results.json in the source repository.
Author: Brent Akamine (Founder, Vinovest). Backtests are not investment advice.