Walk-forward results for a simple EURUSD trend EA, warts and all

@ea_architectjoined Mar 23, 2025Aug 5, 2026464 views4 replies

Sharing a full walk-forward run because most WFO posts show only the winners. Strategy: deliberately boring — Donchian 55 breakout entry, ATR(20) trailing exit, EURUSD H1, 0.5% risk, no filters.

Setup: anchored walk-forward, 8 windows, 70/30 IS/OOS split, 2016-2025, real ticks, variable spread, 0.3 pip slippage modelled.

Results: 6 of 8 windows profitable OOS. Aggregate OOS PF 1.34, max DD 14%, CAGR ~7%. Two windows (2018 chop, 2020 covid whipsaw) were OOS losers at PF 0.87 and 0.79.

The warts: in-sample PF averaged 1.9, so the OOS haircut was 30%. Parameter stability was mediocre — optimal Donchian length wandered from 40 to 70 across windows. And 7% CAGR with 14% DD is not a fundable strategy; it is a building block.

Verdict: passes my bar (at least 6/8 windows, OOS PF 1.2+) as a portfolio component, fails as a standalone. Posting the full report structure in case the format helps anyone.

4 Replies

@risk_firstjoined May 23, 2025#1 · 1 day ago
0
The 30% in-sample to out-of-sample haircut matches my rule of thumb — I assume 25-40% degradation and I am pleasantly surprised maybe once a year. And thank you for the honest "not fundable, it is a building block" framing. That sentence is what most WFO posts carefully omit.
@pips_and_pythonjoined Jun 20, 2025#2 · 16 hours ago
0
Genuine question on the wandering Donchian length: 40 to 70 across windows would fail my stability bar. How do you distinguish "acceptable parameter drift" from "no robust core"? I have never found a clean rule for where that line sits.
@ea_architectjoined Mar 23, 2025#3 · 11 hours ago
+2
Fair challenge — my heuristic: re-run each window using the neighboring window's optimal parameters. If the OOS result stays within 20%, the drift is tolerable; the strategy is not standing on one exact number. This EA passed 7 of 8 of those swaps. I call it the parameter-neighbor test — not standard practice, just the most honest stability check I have found.
@macro_mikejoined May 8, 2025#4 · 8 hours ago
0
Stealing the neighbor test immediately. And note that 2018 chop and 2020 covid failing is actually informative — those are precisely the regimes a Donchian breakout should hate. Failures in the expected places increase my confidence that the other six windows are a real edge rather than luck.