#29 — H1 Donchian sweep

2026-05-19 (Session 062)

Verdict: negative · ⛔ closed negative — no deployable edge at any parameter

The H4 Donchian breakout edge does not scale down to H1: 0 of 2,400 configs finished OOS-positive; spread dominates the smaller H1 moves.

Hypothesis — what it tests

Whether the H4 Donchian breakout edge survives when scaled down to H1 cadence for a higher trade frequency.

Description

The live H4 Donchian breakout on account 010 raised an obvious question: would the same channel-breakout mechanism work faster, at H1? A brute-force sweep tested 200 configurations per pair (channel length x ATR-trail multiple x max-hold x optional H4 direction filter) across all 12 pairs, plus a separate 1,296-config dual M5+H1 confirmation variant. Every single one of the 2,400 pure-H1 configs was OOS-negative (best: EUR_GBP at -6.37 p/d), and the dual-TF variant added 0 survivors. The diagnosis: H1 ATR is 2-3x smaller than H4 ATR while the spread is the same absolute pip cost, and H1 has more false breakouts — smaller expected moves x more reversals x same spread = systematically negative expected value.

Key result

0/2,400 pure-H1 configs OOS-positive (best EUR_GBP −6.37 p/d); 0/1,296 dual M5+H1 configs survive; H1 ATR ~2-3× smaller than H4 at identical spread cost

Indicators

Donchian channelWilder ATR(14)

Algorithms

grid search parameter sweepwalk-forward IS/OOS validationNumba JIT backtest

Code

research/experiments/donchian_h1/backtest_donchian_h1.py research/experiments/donchian_h1m5/backtest_h1m5.py

Interactive version (search, filters, figures): the experiment explorer.

← #28 ZR ML=4 force-close + CAR-25 #30 RACS reversal-accumulator IC study →

This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.