Still testing Lab pick

Slower Donchian exits sit longer in BTC — and give back a lot of dollars

Idea: keep the same “buy the breakout” rule as Donchian 10/5 (price makes a new 10-day high), but sell less often — wait for a slower low, or require a deeper dip below that low. That matches a stay-invested BTC posture better than flipping 88 times. Result: time in the market does go up (43% → 60–77%). Ending coin-units and dollars do not — they fall vs 10/5. Family is Tweak, not a new Keep. Donchian 10/5 stays the BTC paper Keep.

BTC-USD · 2019-01-01 → 2026-09-07 $1,000 start 15 bps round-trip next-bar open · 95% size

Hold 10/10 end

$1.82M

vs Don 10/5

$3.36M

In market

60% vs 43%

End units

23 vs 42

What the rule is, in one sentence

donchian_hold buys when today’s close breaks above the prior N-day high (same as Donchian). It sells only when the close falls through a slower M-day low, optionally by an extra gap. Optional “confirm” waits extra days above the high before buying.

entry_period — lookback for the buy high (kept at 10, same as the Keep).
exit_period — lookback for the sell low (10 / 15 / 20 vs Donchian’s 5).
confirm_bars — extra days the breakout must last (1 = no delay).
exit_gap_pct — extra % below the low before selling (0 = plain channel).

Small grid: 12 combos. Sacred harness unchanged — we did not cheapen fees to look good.

What $1,000 becomes · full sample

Same period, same costs. “In market” = share of days the rule is long (not sitting in cash). “End units” = how many BTC your ending dollars could buy at the last price (purchasing-power score — not a promise you held coins the whole time).

Strategy	End $	Max DD	In market	End units	Trades
Buy & hold	$20,050	−76%	100%	0.25	1
SMA 5/20	$122k	−40%	55%	1.54	80
Donchian 10/5 (Keep)	$3.36M	−18%	43%	42.4	88
Hold 10/10, gap 0	$1.82M	−25%	60%	23.1	57
Hold 10/10, gap 1%	$1.46M	−25%	65%	18.4	48
Hold defaults 10/20, gap 1%	$478k	−45%	77%	6.0	32

Hypothesis check: slower exits did raise time-in-market and cut trades vs Donchian 10/5. They did not raise ending units — 10/5 still ends with ~1.8× the purchasing power of the best hold config, and ~7× the designed defaults. Sitting through more of the dip is the cost. All hold configs still crush buy&hold ($20k, −76% dip) and SMA 5/20.

Small grid (12 knobs)

Ranked by Sharpe (risk-adjusted return). Entry fixed at 10. Every config beat buy&hold’s Sharpe (0.95). Waiting 2 extra days to buy (confirm_bars=2) was a trap — it misses the obvious breakout and ends between $142k and $238k.

Exit / gap / confirm	End $	In mkt	Units	Trades
10 / 0% / 1 (grid winner)	$1.82M	60%	23.1	57
10 / 1% / 1	$1.46M	65%	18.4	48
15 / 0% / 1	$1.16M	69%	14.7	42
defaults 20 / 1% / 1	$478k	77%	6.0	32
confirm=2 (all six)	$142–238k	41–57%	1.8–3.0	23–31

The grid winner with confirm=1 and gap=0 is just plain Donchian 10/10 — not a new sticky trick. Extra stickiness (longer exit, 1% gap) steadily trades dollars for more days in the market.

Year-split stress · Hold 10/10 gap 0

Each calendar year restarts with $1,000. 2026 is year-to-date through Sep 7. Max DD = worst peak-to-trough dip that year.

Year	Rule end	PnL	Max DD	B&H end	vs B&H
2019	$3,778	+278%	−18%	$1,869	BEAT
2020	$5,187	+419%	−13%	$3,868	BEAT
2021	$2,272	+127%	−22%	$1,563	BEAT
2022	$1,774	+77%	−16%	$388	BEAT
2023	$3,048	+205%	−7%	$2,469	BEAT
2024	$2,760	+176%	−12%	$2,143	BEAT
2025	$1,192	+19%	−17%	$937	BEAT
2026*	$1,422	+42%	−9%	$906	BEAT

8/8 beat buy&hold · avg PnL +168% · worst DD −22%. In 2022 it made +77% while buy&hold lost −61%. Trails Donchian 10/5’s year-split (avg +197%, worst DD −18%) but still a clean stress.

Year-split · other knobs

Hold 10/10, gap 1% — also 8/8 beat hold · avg +165% · worst DD −22%. Mild sticky still passes stress.
Defaults 10/20, gap 1% — 7/8 (lost 2023) · avg +118% · worst DD −45% in 2021. Too sticky for a crash-skip story.

Walk-forward (out-of-sample)

Train 365 days → test 90 → step 90 · fixed params. Each fold starts with $1,000. Donchian 10/5’s prior BTC walk-forward was 21/27 (78%).

10/10 g0 folds

20 / 27

10/10 g0 mean

+24.2%

10/10 g1 folds

19 / 27

Defaults 10/20

15 / 27

Grid winner beats buy&hold in 74% of folds (worst fold −13%, worst DD −24%). Gap 1% is 70%. Designed defaults drop to 56% — no better than the old SMA 5/20 walk-forward (~57%). Out-of-sample, extra stickiness is not free.

Keep · Tweak · Drop

Lab pick

donchian_breakout 10/5 on BTC — still the paper Keep. Best dollars, best units, softest dip, 8/8 years, 21/27 walk-forward. The hybrid did not beat it.

Still testing

donchian_hold family — idea is real (slower exits sit longer) but the extra hold costs purchasing power. Grid winner 10/10 is just calmer Donchian, not a new edge. Keep the code; don’t promote it over 10/5.

Still testing

Hold 10/10 gap 0 (or 1%) — 8/8 years, walk-forward 70–74%, fewer sells (57 or 48 vs 88). Reserve if 10/5 churn feels too busy. Not an upgrade on $ or units.

Dropped

confirm_bars=2 as a default — delays the obvious buy; six configs end $142k–$238k. Don’t wait extra days to enter a 10-day breakout on BTC.

Dropped

defaults 10/20 + 1% gap as primary — 77% in market looks “hold-y” but walk-forward 56%, lost 2023, dip −45%. Too much lag.

Path to real

Screen

Stress

Paper-live

Real (needs OK)

This hybrid finished Screen + Stress. It is not the next paper-live book. The queued book is still Donchian 10/5 on BTC. No live orders without an explicit per-trade OK.

Suggested next experiment

Bootstrap paper_live for the existing Keep — donchian_breakout 10/5 on BTC-USD — rather than enlarging this exit-length grid. If “sells too often” still bothers the core sleeve, try a capital split (part stay-invested, part 10/5) instead of blending the exit slower. Slower Donchian exits gave back the crash-skip that made 10/5 a Keep.