Trading Lab · Experiment brief
Sun Sep 6, 2026 · paper only · alts research

Published 2026-09-06 21:08:29 CEST

Lab pick Dropped Dropped

Donchian breakout is the alts edge — trend_rsi & vol-filter are noise

Screened 6 strategy families × 2 alts (ALGO-USD since Jun 2019, ETH-USD since Jan 2019). Donchian breakout (buy at N-day high, sell at M-day low) smashes every baseline on both coins — 8/8 years beat hold, walk-forward 24/25 ALGO and 22/27 ETH. Trend-RSI never trades on ALGO (0 entries across 36 configs). Vol-filter SMA collapses to plain SMA on crypto (always-volatile). Both are honest DROPs — null results are cheap signal.

ALGO-USD · 2,634 bars ETH-USD · 2,805 bars $1,000 start 15 bps costs
Strategies tested
6
Grid configs
130+
KEEP families
1
DROP families
2

Baselines · what $1,000 becomes

Full sample: ALGO Jun 2019 → Sep 2026 · ETH Jan 2019 → Sep 2026. ALGO crashed −95% from its all-time high (means "buy and hold" is brutal — it's actually hard to LOSE money vs that baseline).

Strategy Symbol End $ Max DD Trades Sharpe
Buy & holdALGO$77−93%1−0.07
SMA 5/20ALGO$24,005−77%860.95
Donchian 10/5ALGO$8.8M−30%702.14
Buy & holdETH$17,661−79%10.86
SMA 5/20ETH$195,382−37%791.54
Donchian 10/5ETH$81.4M−18%723.32
Why so large? These aren't fantasy numbers — they're compounded daily returns over 7 years of crypto. ALGO went from $0.03→$2.80→$0.09. Donchian caught the rallies and dodged most crashes. Same trade count as SMA 5/20 (~70), but much softer drawdowns and far higher compound returns.

Year-split · Donchian 10/5 on ALGO

Each row restarts with $1,000 on Jan 1 (2019 = partial from Jun, 2026 = YTD).

YearDon endPnLMax DDHold endvs hold
2019*$1,821+82%−13%$113BEAT
2020$5,276+428%−24%$1,504BEAT
2021$5,755+476%−30%$4,755BEAT
2022$1,774+77%−14%$149BEAT
2023$2,862+186%−15%$1,271BEAT
2024$7,379+638%−16%$1,469BEAT
2025$1,522+52%−25%$363BEAT
2026*$1,353+35%−15%$861BEAT

8/8 beat hold · avg PnL +246.8% · worst DD −30.1%. The strategy turned a profit in 2022 (a crypto winter where ALGO fell −71%) and in every other year.

Year-split · Donchian 10/5 on ETH

YearDon endPnLMax DDHold endvs hold
2019$4,121+312%−11%$970BEAT
2020$12,117+1112%−14%$5,441BEAT
2021$7,098+610%−18%$4,778BEAT
2022$2,411+141%−14%$358BEAT
2023$2,408+141%−10%$1,856BEAT
2024$3,655+265%−9%$1,433BEAT
2025$2,981+198%−11%$893BEAT
2026*$1,591+59%−9%$842BEAT

8/8 beat hold · avg PnL +354.8% · worst DD −18.2% (softest in the whole repo). Made +141% in 2022 while ETH buy-and-hold lost −64%.

Walk-forward (out-of-sample)

Train 365 days → test 90 → step 90 · fixed params. Each fold starts with $1,000.

ALGO folds won
24 / 25
ALGO mean / fold
+39.5%
ETH folds won
22 / 27
ETH mean / fold
+40.1%

Donchian 10/5 beats buy-and-hold in 96% of ALGO folds and 81% of ETH folds. Compare BTC SMA 5/20 which managed 57%. This is the strongest out-of-sample result in the repo.

Calmer alternative · Donchian 20/10

Wider lookback window = fewer trades, slightly weaker returns, but calmer to run.

ConfigSymbolEnd $TradesStressWF beat
Don 20/10ALGO$1.0M327/820/25
Don 20/10ETH$2.3M416/815/27

Fewer trades (32 vs 70 on ALGO). Still beats hold on both assets. Weaker walk-forward on ETH (56%) — acceptable if you prefer less churn.

Dead on alts

Dropped
trend_rsi on alts — 0 trades across all 36 grid configs on ALGO. The "buy pullbacks in an uptrend" logic never fires because alts don't pull back gently — they crash or spike. RSI conditions are too restrictive for volatile coins.
Dropped
vol_filter_sma on alts — at low ATR thresholds (1.5–2.5%) it's identical to plain SMA crossover (crypto is always volatile, the gate never closes). At high thresholds (4%) it kills good entries too. The concept makes sense for equities but adds zero value on crypto.
Parked
sentiment_sma / sma_sticky / btc_hold_protect — not tested on alts this round. Sentiment requires Fear & Greed data. Sticky and hold-protect are BTC core-sleeve designs. Could revisit later, but Donchian covers the alts thesis for now.

Keep · Tweak · Drop

Lab pick
donchian_breakout 10/5 (ALGO + ETH) — 8/8 year-split, 24/25 + 22/27 WF, Sharpe 2.1/3.3, softest drawdowns. Strongest result in the repo. Paper-track next.
Still testing
donchian_breakout 20/10 — calmer (32 trades ALGO, 41 ETH), still beats hold 7/8 + 6/8. Weaker WF on ETH. Reserve as alt config if 10/5 churn feels too busy.
Lab pick
sma_crossover 5/20 baseline — still the reference to beat. Strong on both alts ($24,005 ALGO, $195,382 ETH). Already KEEP on BTC.
Dropped
trend_rsi on alts — zero signal, zero trades. Valid concept on BTC where pullbacks exist, dead on volatile alts.
Dropped
vol_filter_sma on crypto — collapses to plain SMA. The ATR gate never engages. Honest null result.

Path to real

Screen
Stress
Paper-live
Real

Donchian 10/5 has passed screen + stress + walk-forward. Next step is paper-live tracking (watch signals daily, record fills without money on the line). Do NOT ship to real without explicit OK.

Five things learned

Next build step

Paper-live Donchian 10/5 on ALGO and ETH. Watch signals daily for 1–2 months. If the signal quality holds up in real time (not lagging, not whipsawing), then consider a small real allocation from the alts research. One config, one or two coins, small size. No grid-spiral — the knobs are already tested.

Traceability