Still testing

SMA 5/20 on Bitcoin: softens crashes, still sells too often

We stress-tested the short/fast moving-average rule (5 & 20 days) year by year and in rolling 90-day windows. It often beats “just hold BTC” on paper — especially in ugly years — but it regularly steps out of Bitcoin. For your stay-invested style, that’s a tweak signal, not a green light.

BTC-USD fast=5 · slow=20 $1,000 / year main 3a1878f · PR #3

Years vs hold

6 / 7

Avg year PnL

+89.9%

Worst DD

−32.5%

WF beat hold

13 / 23

Year table · plain $

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

Year	SMA end	SMA PnL	Max DD	Hold end	Hold PnL	vs hold
2020	$4,258	+326%	−17%	$3,868	+287%	BEAT
2021	$1,661	+66%	−32%	$1,563	+56%	BEAT
2022	$727	−27%	−33%	$388	−61%	BEAT
2023	$1,714	+71%	−18%	$2,469	+147%	LOST
2024	$2,545	+154%	−13%	$2,143	+114%	BEAT
2025	$1,107	+11%	−16%	$937	−6%	BEAT
2026*	$1,280	+28%	−9%	$914	−9%	BEAT

Worst years: 2022 still lost ~$273 but held up far better than buy-and-hold (~$388 left). 2023 made money (+71%) but trailed a strong hold year (+147%). Deepest strategy dip inside a year ≈ −33%.

Walk-forward (fixed 5/20)

Train 365 days → test 90 → step 90. 23 folds, each starting at $1,000.

Mean return

+11.6%

Mean end $

$1,116

Best / worst

+57% / −13%

Worst DD

−21.3%

Beats buy-and-hold in 57% of folds — useful shock absorber, not an obvious consistent winner.

Keep · Tweak · Drop

Still testing

Live BTC core rule — softens crashes but sells BTC too often vs stay-invested style.

Lab pick

Paper research on this family — traceable baseline; 6/7 years beat hold.

Dropped

Go live — not an obvious winner. Paper only.

vs just holding BTC

Usually wins the year comparison, especially crash/chop years (2022, 2025, 2026 YTD).
Strong bull stretch in 2023: holding wins big.
Cost of the edge: being out of BTC in some of the best stretches — fights “grow the BTC pile.”

ALGO same rule (optional)

5/7 years beat hold · avg +87% · worst DD −55%. Deeper pain — alts research territory, not BTC core. 2026 YTD still red.

Next build step

Try a stickier BTC rule (longer averages or “exit only on deep fear / big drawdown”) so the default is stay invested. Re-run year-split + walk-forward and score time-in-market and BTC units held over time — not just dollars.

Traceability

experiments/runs/stress_20260906T162745Z — BTC year-split
experiments/runs/walkforward_20260906T162749Z — BTC walk-forward
experiments/runs/stress_20260906T162805Z — ALGO year-split
briefs/2026-09-06-sma520-stress.md · experiments/results.tsv