Signal-only research · No auto-trading · Not financial advice
20-Day Breakout + 10-Day Trailing Stop
Daily scan for new 20-day highs (long) and lows (short), ranked by volatility-normalised strength, at most 5 positions per side, exit on a 10-day trailing stop. 5 years of data (2021-08 → 2026-08): 43 Bybit USDT perps, 97 US stocks, 38 index ETFs, 18 commodity ETFs, net of fees, slippage and funding.
0 · Audit corrections 2026-09-28
The first published version of these results was materially overstated. The engine and metrics were fixed and everything below was regenerated:
- Position cap bug — "top 5 per side" was checked after adding, so one extra position got in every day with a signal (up to 33 open at once). Now a hard cap of 5 per side.
- Chandelier look-ahead — the stop used today's ATR, which already contains today's range. Now uses the prior bar's ATR.
- Funding not modelled — perp funding over each holding period is now charged or credited (longs pay in bull markets).
- Closed-trade drawdowns — account results compounded R-multiples one trade after another. Now a daily mark-to-market portfolio holding all positions at once (1% risk per trade, notional capped at 25% of equity).
- Universe selection — the crypto list is today's top 60 by volume; 27 of the 43 coins listed after 2021 and are in the list because they rallied. Results are now shown for both the 43 and the 16 listed by 2021 (both still survivor-only).
- Walk-forward — configs were picked on the full sample, then "tested" on sub-periods of it. Now selected on each train window only.
- Also: ETFs were double-counted inside "stocks", the ATR "initial stop" was overwritten on day one (never a stop), hard stops filled at the stop price through gaps, take-profit was assumed to hit before the stop on the same bar, and the entry bar had no stop.
| Headline (crypto) | Published before | Now · 43 coins | Now · 16 listed by 2021 |
|---|
| BASE 20/10 · net sum of trade returns | +2,098% | +1,502% | +738% |
| BASE 20/10 · account equity after 5y (1% risk) | ×2.84 | ×2.19 | ×1.62 |
| BASE 20/10 · CAGR | +23.2% | +17.0% | +10.1% |
| BASE 20/10 · max drawdown | 19.0% (closed trades) | 23.1% (MTM) | 20.5% (MTM) |
| BASE 20/10 · MC P(profit) | 96.9% | 94.3% | 84.2% |
| CHAND4 40/15 · account equity after 5y | ×6.48 | ×2.40 | ×2.26 |
| CHAND4 40/15 · CAGR | +45.3% | +19.2% | +17.7% |
Bottom line: a trend-following edge on crypto perps survives the fixes, but it is roughly half of what was published, it depends heavily on 2023 and 2025, and the "optimised" chandelier variant is not reliably better than simpler settings.
1 · Historical Backtest 5Y · 2021-08 → 2026-08 · NET
20-day breakout, 10-day trailing stop, max 5 open positions per side. Net sum = equal-weight sum of per-trade % P&L after 21 bps round-trip and funding — a trade-edge metric, not an account return. CAGR / MTM DD = the daily mark-to-market account at 1% risk per trade.
| Asset class | Trades | Gross sum | Net sum | Funding | PF | CAGR | MTM DD | Verdict |
|---|
| Crypto-perps · 43 (2026 top-volume) | 1,227 | +1,875% | +1,502% | -116% | 1.33 | +17.0% | 23.1% | EDGE |
| Crypto-perps · 16 listed by 2021 | 1,020 | +1,038% | +738% | -86% | 1.23 | +10.1% | 20.5% | WEAK |
| Stocks (US) | 1,349 | +385% | +102% | — | 1.04 | -1.5% | 34.0% | NONE |
| Index ETFs | 1,074 | -26% | -252% | — | 0.75 | -9.4% | 41.0% | NONE |
| Commodity ETFs | 1,174 | -381% | -627% | — | 0.73 | -11.4% | 55.9% | NONE |
Only crypto shows an edge, and on the 16 coins that existed at the start it is about 60% of the full-universe figure. Stocks are roughly flat; index and commodity ETFs lose money. Verdict = EDGE when CAGR > 0 and ≥ 90% of daily-bootstrap sims are profitable.
2 · Monte Carlo — Trade-Edge Distribution 10,000 SIMS · NET
Block-bootstrap of the realised net trade list, summed at equal weight. This measures how robust the per-trade edge is to resampling — not account returns (see §3). It only resamples in-sample trades, so it cannot capture selection bias or regime change.
| Asset class | Sum P5 | Sum P50 | Sum P95 | P(sum > 0) |
|---|
| Crypto-perps · 43 (2026 top-volume) | -150% | +1,471% | +3,382% | 93.0% |
| Crypto-perps · 16 listed by 2021 | -687% | +691% | +2,461% | 78.0% |
| Stocks (US) | -437% | +98% | +740% | 60.9% |
| Index ETFs | -458% | -256% | -28% | 3.5% |
| Commodity ETFs | -1,182% | -641% | -15% | 4.5% |
| All combined (excl. early subset) | -966% | +811% | +2,864% | 76.1% |
3 · Account-Level Results MTM · 1% RISK / TRADE · NET + FUNDING
A single account holds every open position at once, sized so the initial stop risks 1% of equity (notional capped at 25% of equity), marked to market daily. The Monte Carlo block-bootstraps that account's daily returns (20-day blocks, 10,000 sims).
Crypto 43 · CAGR
+17.0%
Equity after 5y×2.19
Max DD (MTM)23.1%
MC CAGR P5 → P95-0.6% → +38.1%
P(profit)94.3%
Crypto 16 (early) · CAGR
+10.1%
Equity after 5y×1.62
Max DD (MTM)20.5%
MC CAGR P5 → P95-5.5% → +27.8%
P(profit)84.2%
Stocks · CAGR
-1.5%
Equity after 5y×0.93
Max DD (MTM)34.0%
MC CAGR P5 → P95-8.6% → +8.2%
P(profit)43.1%
CAGR distribution (5th / 50th / 95th percentile)
| Asset class | Trades | Equity ×5y | CAGR | MTM DD | Peak gross | MC P5 | MC P50 | MC P95 | P(profit) | MC DD P50 |
|---|
| Crypto-perps · 43 (2026 top-volume) | 1,227 | ×2.19 | +17.0% | 23.1% | 1.1× | -0.6% | +16.3% | +38.1% | 94.3% | 27.0% |
| Crypto-perps · 16 listed by 2021 | 1,020 | ×1.62 | +10.1% | 20.5% | 0.9× | -5.5% | +9.4% | +27.8% | 84.2% | 29.4% |
| Stocks (US) | 1,349 | ×0.93 | -1.5% | 34.0% | 1.6× | -8.6% | -0.9% | +8.2% | 43.1% | 25.2% |
| Index ETFs | 1,074 | ×0.61 | -9.4% | 41.0% | 2.2× | -15.2% | -9.2% | -2.4% | 1.5% | 43.3% |
| Commodity ETFs | 1,174 | ×0.55 | -11.4% | 55.9% | 1.8× | -20.1% | -11.7% | -1.6% | 2.8% | 53.6% |
| One account, all classes | 4,824 | ×0.68 | -7.5% | 64.9% | 4.5× | -25.8% | -7.7% | +16.3% | 28.6% | 61.8% |
Even the full crypto universe has a 5th-percentile CAGR around zero, and trading every asset class in one account loses money — the stock/ETF books dilute the crypto edge and add drawdown.
4 · Parameter Sensitivity CRYPTO · NET · MTM ACCOUNT
30 lookback × trail combinations, sorted by account CAGR (top 8 + the 20/10 spec in bold). 30/30 are profitable on the 43-coin universe and 29/30 on the 16 early coins — the edge is broad rather than a single point. Picking the top row is still in-sample; see the walk-forward in §6.
43 coins (2026 top-volume)
| Lookback | Trail | Trades | Net sum | PF | CAGR | MTM DD |
|---|
| 30 | 15 | 930 | +2,000% | 1.56 | +28.1% | 26.7% |
| 40 | 15 | 829 | +2,015% | 1.63 | +26.8% | 25.2% |
| 20 | 15 | 1,094 | +1,778% | 1.43 | +22.5% | 26.4% |
| 10 | 20 | 1,234 | +1,396% | 1.31 | +20.9% | 27.0% |
| 40 | 20 | 790 | +2,004% | 1.67 | +20.9% | 31.6% |
| 40 | 10 | 914 | +1,669% | 1.49 | +20.4% | 20.4% |
| 30 | 20 | 877 | +1,907% | 1.57 | +19.8% | 28.7% |
| 40 | 30 | 690 | +2,019% | 1.75 | +19.8% | 27.6% |
| 20 | 10 | 1,227 | +1,502% | 1.33 | +17.0% | 23.1% |
16 coins listed by 2021
| Lookback | Trail | Trades | Net sum | PF | CAGR | MTM DD |
|---|
| 40 | 15 | 644 | +1,329% | 1.65 | +15.6% | 22.7% |
| 55 | 15 | 526 | +1,392% | 1.81 | +14.9% | 19.6% |
| 40 | 10 | 694 | +1,242% | 1.57 | +13.9% | 19.6% |
| 55 | 10 | 568 | +1,318% | 1.73 | +13.7% | 16.0% |
| 30 | 15 | 762 | +1,001% | 1.42 | +11.6% | 23.5% |
| 20 | 15 | 946 | +821% | 1.27 | +11.1% | 23.0% |
| 30 | 10 | 805 | +1,023% | 1.41 | +10.9% | 21.4% |
| 15 | 10 | 1,140 | +814% | 1.23 | +10.5% | 21.9% |
| 20 | 10 | 1,020 | +738% | 1.23 | +10.1% | 20.5% |
5 · Yearly & Long/Short CRYPTO 43 · 20/10 · NET TRADE SUMS
| Year | Trades | Net sum | Win rate | PF | Longs (n · net) | Shorts (n · net) |
|---|
| 2021* | 69 | -168% | 8.7% | 0.41 | 39 · -141% | 30 · -27% |
| 2022 | 212 | -183% | 19.3% | 0.75 | 107 · -361% | 105 · +178% |
| 2023 | 189 | +815% | 23.8% | 2.63 | 92 · +820% | 97 · -4% |
| 2024 | 282 | +127% | 14.9% | 1.12 | 155 · +341% | 127 · -214% |
| 2025 | 252 | +824% | 19.4% | 1.74 | 142 · +753% | 110 · +72% |
| 2026* | 223 | +86% | 17.5% | 1.10 | 115 · +83% | 108 · +3% |
| Side | Trades | Share | Net sum | P&L share | Win rate | PF | Avg R |
|---|
| LONG | 650 | 53.0% | +1,495% | 99.5% | 16.9% | 1.54 | +0.13R |
| SHORT | 577 | 47.0% | +7% | 0.5% | 19.4% | 1.00 | +0.02R |
* partial years. Results are concentrated: 2021–22 lose money and most of the P&L comes from 2023 and 2025.
6 · Exploration Tests CRYPTO · NET · MTM
- A2 take-profit — on 20/10, no TP: CAGR +17.0%, DD 23.1%, Calmar 0.73; best TP (5R): CAGR +18.7%, DD 16.1%, Calmar 1.16. With mark-to-market drawdowns, a far take-profit trims drawdown more than return, reversing the earlier "TP never helps" conclusion — but this is one more in-sample grid.
- B8 chandelier 4×ATR on 40/15 — 43 coins: CAGR +19.2%, DD 23.9%, Calmar 0.80 vs base CAGR +26.8%, DD 25.2%, Calmar 1.06. 16 coins: CAGR +17.7%, DD 22.4%, Calmar 0.79 vs base CAGR +15.6%, DD 22.7%, Calmar 0.69. Lower multipliers (2–3×) are much worse. No consistent advantage.
- B6 ATR initial stop (now a real stop) — a tighter stop means a bigger position at 1% risk: 1.5× ATR gives CAGR +45.6%, DD 38.7%, Calmar 1.18. Calmar barely moves: it is a leverage dial, not an edge.
- C12 concentration — TOP10 40/15: 43 coins CAGR +37.2%, DD 25.9%, Calmar 1.43, 16 coins CAGR +16.7%, DD 27.4%, Calmar 0.61. TOP3: CAGR +15.2%, DD 21.9%, Calmar 0.69. The TOP10 edge does not carry to the early universe.
- C14 BTC-correlation cap — cap 1: CAGR +4.1%, DD 10.3%, Calmar 0.40 vs base 20/10 CAGR +17.0%, DD 23.1%, Calmar 0.73. Cuts drawdown by cutting return.
- E20 walk-forward (43 coins) — choosing the best train-window Calmar each year (W1=CHAND4 40/15, W2=BASE 40/10, W3=BASE 40/10) compounds to ×2.11 over the 3 test years, vs ×2.46 for the untuned BASE 20/10. Parameter selection does not add value out-of-sample.
- E20 walk-forward (16 early coins) — choosing the best train-window Calmar each year (W1=CHAND4 40/15, W2=CHAND4 40/15, W3=BASE 40/10) compounds to ×1.73 over the 3 test years, vs ×1.87 for the untuned BASE 20/10. Parameter selection does not add value out-of-sample.
Direction: the plain breakout on crypto perps, long-biased, with no parameter tuning beyond the spec until a forward test says otherwise. Details on the
chandelier re-evaluation page.
7 · Methodology & Caveats READ ME
- Signal: long when today's high > max(high of prior N days); short when today's low < min(low of prior N). Ranked by (close − N-day SMA) / ATR(14); at most 5 positions open per side. Entry at the next bar's open.
- Exit: trailing stop from preceding bars only; the initial stop is live on the entry bar. Stops that gap fill at the open. When a stop and a take-profit are both touched on one daily bar, the stop is assumed first.
- Costs: 5.5 bps taker + 5 bps slippage per side (21 bps round-trip). Perp funding from Bybit history is charged over each holding period. Stock/ETF shorts ignore borrow cost.
- Account: daily mark-to-market, 1% of equity risked per trade at the initial stop, position notional capped at 25% of equity.
- Survivorship & selection: crypto symbols are 2026's top 60 by volume that still trade — delisted coins (LUNA, FTT, …) are missing and later listings were selected for their success. The 16-coin "listed by 2021" view removes the second effect, not the first. Treat all crypto numbers as optimistic.
- Multiple testing: dozens of variants were run on the same 5 years. Only the walk-forward in §6 selects without seeing the test data.
- Signal-only: research, not financial advice.