Still testing Dropped

Hold-protect softens crashes — and shrinks the BTC pile

btc_hold_protect stays long until a deep drawdown, then waits for recovery. Defaults (dd=25%, reentry=10%) cut max DD to −28% vs hold −76%, but end with fewer BTC units than buy&hold and lose 6/7 calendar years. Fast SMA 5/20 still wins on $ and units. Grid: wider re-entry helps — keep the family, don’t ship defaults.

BTC-USD dd 25 · re 10 fast 5/20 $1,000 main 669099c+

vs hold (yrs)

1 / 7

WF beat hold

6 / 23

HP end units

0.077

Fast end units

0.612

Full sample · $ and BTC-units lens

One continuous run: 2020-01-01 → 2026-09-05 · start $1,000.

Strategy	End $	Max DD	Trades	In mkt	BTC units
Hold-protect 25/10	$6,160	−28%	7	42%	0.077
Fast 5/20	$48,827	−40%	70	54%	0.612
Buy & hold	$10,558	−76%	1	100%	0.132

Units lens: Defaults end at 0.077 BTC-equivalent vs hold 0.132 and fast SMA 0.612. Crash softness bought with too much cash time (only ~42% in market) — fights BTC core stay-invested style.

Year-split · defaults

Each row restarts with $1,000 on Jan 1 (2026 = YTD).

Year	HP end	PnL	Max DD	Hold end	vs hold
2020	$3,293	+229%	−28%	$3,868	LOST
2021	$1,506	+51%	−24%	$1,563	LOST
2022	$681	−32%	−34%	$388	BEAT
2023	$2,469	+147%	−20%	$2,469	LOST
2024	$1,903	+90%	−26%	$2,143	LOST
2025	$864	−14%	−24%	$937	LOST
2026*	$840	−16%	−24%	$914	LOST

1/7 beat hold · avg PnL +65.1% · worst DD −33.7%. Soft 2022; miss bulls elsewhere.

Walk-forward (fixed params)

Train 365 → test 90 → step 90 · 23 folds · $1,000 each.

Mean / fold

+12.1%

Mean end $

$1,121

Worst DD

−23.6%

Beat hold

6 / 23

Positive mean, but beat-hold rate only 26% (fast SMA was ~57%). Shock absorber, not OOS winner.

Small grid · re-entry is the knob

dd ∈ {20,25,30} × reentry ∈ {5,10,15} · ranked by Sharpe · same full sample.

Params	End $	Max DD	In mkt	Units
dd20 / re15 ★	$16,783	−46%	55%	0.210
dd20 / re10	$10,112	−26%	41%	0.127
dd25 / re15	$9,428	−28%	44%	0.118
defaults 25/10	$6,160	−28%	42%	0.077
buy & hold	$10,558	−76%	100%	0.132

Grid read: 8 configs beat B&H Sharpe. Winner 20/15 beats hold on $ and units (0.210) but still ≪ fast SMA 0.612. Wider re-entry > tighter.

Keep · Tweak · Drop

Dropped

Defaults (25/10) as BTC core rule — fewer units than hold; 1/7 year wins; only ~42% time-in-market.

Still testing

Hold-protect idea family — right shape (accumulate + crash exit); stress grid winner + optional F&G gate.

Lab pick

Fast SMA 5/20 as paper baseline — still best $ & units on this path; live core remains TWEAK.

Lab pick

Buy&hold as core reference — 100% time-in-market; deepest DD (−76%).

Dropped

Go live — paper only; no explicit OK.

Five insights

Defaults cut DD to −28% vs hold −76%, but end $6.2k vs $10.6k and units 0.077 vs 0.132.
Only 41.6% time-in-market — cash after exits; 10% re-entry is sticky the wrong way in bulls.
Year-split: only 2022 clearly beats hold; other years leave upside on the table.
Walk-forward beat-hold 26% vs prior fast SMA ~57% — not an OOS winner.
Grid: reentry=15 lifts units above B&H (0.210); still ≪ SMA 5/20 (0.612) — tweak, don’t crown.

Next build step

Re-run stress + walk-forward on grid winner dd_pct=20, reentry_pct=15. Optionally gate re-entry with Fear & Greed (fng_floor). Keep scoring ending_units + time-in-market for BTC core.

Traceability

experiments/runs/stress_20260906T165623Z — year-split defaults
experiments/runs/walkforward_20260906T165620Z — WF 365/90/90
experiments/runs/20260906T165628Z/ — full-sample HP + SMA 5/20
experiments/runs/20260906T165629Z/ — full-sample B&H
experiments/runs/20260906T165632Z/ — small grid
briefs/2026-09-06-btc-hold-protect.md · experiments/results.tsv