Donchian breakout on crypto: does it hold up on new coins?

A channel breakout is one of the oldest trend rules with a name: buy when the price makes a new high for the last N bars, sell when it makes a new low. The channel is named after Richard Donchian, and N is usually 20.¹ The "Turtle" traders of the 1980s were taught a breakout system of this kind.²

The version tested here adds a trend filter, a momentum band, stops measured in ATR and a short side, and its numbers were tuned on the BTC perpetual, then on a handful of other coins. On those coins it looks good. This article asks the question tuning cannot answer: what does the same rule do on coins it has never seen? The test for that was written down before it was run, and the answer is mixed.

The rule

• Long: on a 4h bar whose high reaches the 20-bar Donchian high (the highest high of the last 20 bars, the bar itself included), with the close above its 200-bar EMA and RSI(14) between 62 and 80. The trade opens at the next bar's open.
• Short: the mirror image — the low reaches the 20-bar low, the close is below the EMA 200, RSI(14) is between 20 and 38 — and only while the last funding rate is zero or positive, so the short is not the side paying it.
• Exits: a stop 2 ATR(14) from the entry; a trailing stop 3 ATR behind the best close since the entry; for shorts, a take-profit at 3R — three times the distance to the initial stop.
• Size: each trade risks 1% of the account at its stop (the position is 1% of equity divided by the distance to the stop), 3x leverage, isolated margin, at most one long and one short open at once.

Two versions run throughout. v2 is the rule above — the BTC configuration, strategy #104. The alt version (strategy #107) differs in one thing: longs trail 5 ATR instead of 3, which the earlier research found helped on most altcoins but not on BTC; with the wider trail a short signal sometimes came while a long was still open, and 78 of its 10,228 trades opened next to one (v2 never had two open). As saved, #104 also stops opening trades after a 25% drawdown and resolves stops inside a 4h bar from 5-minute candles. Both are off in the cross-coin runs (strategy #154): the drawdown guard blocks entries until the end of a run once it trips, and 5-minute candles exist on the platform for only a few perpetuals. On BTC the numbers come out the same either way.

Data and method

• Market: Binance USDT-M perpetuals, 4h candles, 2020-01-01 → 2026-10-01 — or from a coin's first bar if it listed later, and to its last bar if it was delisted. 10,000 USDT per run, fills at the next open, 0.05% fee and 0.03% slippage per side, funding charged and paid.
• Where the numbers came from: the rule is its author's own configuration, tuned mainly on BTC. In an earlier session about 60 variants were then tried on it by dry run — a choppiness filter, Donchian lengths of 10, 40 and 55, other RSI bands, stops of 1.5 and 3 ATR, 2% risk and more — first on BTC, ETH, SOL, BNB and XRP, then on LINK, ADA, DOGE, LTC, AVAX and DOT, and finally on 23 perpetuals. Two changes survived: the 3R take-profit for shorts (v2) and the 5 ATR long trail (the alt version). Every coin that research touched counts as seen: 23 perpetuals, recovered from that session's log.
• Benchmark: holding the perpetual — a 1x long with 95% of the account, closed and reopened at the start of every month so its size follows the account, same costs, funding paid (strategy #155). Why it is re-sized monthly: what the carry cost a perpetual long. It holds from the first bar; the breakout first waits 199 bars, about 33 days, for its 200-bar EMA.
• Runs: BTC as saved #2109; v2 #2148–#2170 on the seen coins and #2171–#2190 on the new ones; the alt version #2191–#2213 and #2214–#2233; the benchmark #2234 (BTC) and #2235 (RUNE). The other benchmark results, the two halves of the period and v1 are dry runs. All four strategies are private, so their embeds show the summary and the equity curve.

How to read Sharpe ratio, drawdown and the rest: backtest metrics explained.

On BTC, where it was tuned

Backtest

BTC 4h Trend Breakout (research)

BTCUSDT · 4h · Futures · 2020-01-01 → 2026-10-01

Return

+249.40%

Max drawdown

11.99%

Sharpe

1.38

Win rate

37.8%

Trades

325

Open the report →

On the BTC perpetual v2 made +249.40% in 325 trades, with a Sharpe ratio of 1.38 and a max drawdown of 11.99%; the profit factor was 1.75. The first version, without the short take-profit, made +234.80% with a Sharpe ratio of 1.24 and a 13.23% drawdown in 302 trades; a dry run today gives the same. The fresh run of v2 repeats the one from 2026-10-22, +249.40%, to the cent.

Holding the BTC perpetual over the same window made more — +329.98% — with half the Sharpe ratio, 0.67, and a 78.32% drawdown; the monthly long paid 25,999 USDT of funding along the way. The breakout spent 48.3% of the time in the market and never risked more than 1% of the account on a trade, which is where its smaller drawdown comes from.

Backtest

Perp long 1x, re-sized monthly (benchmark)

BTCUSDT · 4h · Futures · 2020-01-01 → 2026-10-01

Return

+329.98%

Max drawdown

78.32%

Sharpe

0.67

Win rate

54.3%

Trades

81

Open the report →

One number argues for caution even here. The platform's trials registry counts every attempt on the same symbol, timeframe and window — 283 on BTCUSDT 4h for 2020–2026 on 2026-10-30, from all strategies — and corrects the Sharpe ratio for them. The deflated Sharpe ratio of run #2109 — the probability that its true Sharpe ratio is above zero once those attempts are allowed for — was 0.18: after that many attempts a Sharpe of 1.38 on one coin is not, by itself, evidence of an edge.

The 23 coins it was tuned around

On the coins it was tuned around the breakout looks robust. v2 made money on 20 of 23, the alt version on 22 of 23, with median Sharpe ratios of 0.54 and 0.67 against 0.56 for holding the perpetual. The alt version's Sharpe ratio was above holding on 16 of 23 coins, v2's on 12. Both had a smaller drawdown than holding on all 23: a median of 20.68% for v2 and 24.83% for the alt version, against 94.57%. BTC had the highest v2 Sharpe ratio of all 43 coins in this article — the parameters lean towards the coin they were built on.

The alt version ended ahead of v2 on all 23 coins, and its Sharpe ratio was higher on 19. Most of the profit came from longs — 99,397 of v2's 166,173 USDT summed over the 23 coins — although the shorts made money on more of them: on all 23, against 16 for the longs (for the alt version, 23 and 21).

The out-of-sample check

Every coin above has been looked at, so none of them can say whether the rule works elsewhere. The holdout was fixed before any of its runs:

1. Pool: the USDT-M perpetuals trading on 2021-01-01, ranked by turnover over the 30 days before, delisted coins included — the platform's point-in-time list (GET /market/universe), which ranks 52 perpetuals on that date.
2. Exclude the 23 seen coins; take the first 20 of what is left by turnover. A coin without 4h candles would be replaced by the next one — none was needed.
3. Run v2 and the alt version unchanged, plus the benchmark, over the same window.
4. Pass bar: a version holds up if it makes money on at least 14 of the 20 coins and its median Sharpe ratio is at least half its median on the 23 seen coins.

The rule picked GRT, EOS, EGLD, COMP, WAVES, CRV, YFII, SNX, MKR, SXP, OMG, AXS, BZRX, RUNE, DEFI, REN, BAL, ALPHA, SRM and OCEAN. Thirteen of them have been delisted since, which is what picking from 2021 rather than from today does — see survivorship bias, measured.

v2 failed both parts of the bar: it made money on 13 of 20 new coins and its median Sharpe ratio fell from 0.54 to 0.17. The alt version passed the first part and missed the second by a hair: 15 of 20 coins in profit, but a median Sharpe ratio of 0.329 against a bar of 0.333 — half of its 0.666 on the seen coins. By the rule written before the test, neither holds up.

What did carry over is the drawdown control. The median max drawdown on the new coins — 21.57% and 24.02% — was about the same as on the seen ones, and the worst was 43.48% (v2 on SNX). With 1% risk per trade the rule loses slowly when it loses. In each version three trades were liquidated, one each on AAVE, SXP and ALPHA: in a violent market the 2 ATR stop sat further away than the 3x liquidation price. Each cost about 1% of the account or less, because the margin behind a 1%-risk position is small.

Backtest

Alt 4h Trend Breakout (research)

RUNEUSDT · 4h · Futures · 2020-01-01 → 2026-10-01

Return

+150.94%

Max drawdown

15.07%

Sharpe

1.08

Win rate

33.6%

Trades

268

Open the report →

Backtest

Donchian 4h v2, no guard, no intrabar (research)

SNXUSDT · 4h · Futures · 2020-01-01 → 2026-10-01

Return

-42.31%

Max drawdown

43.48%

Sharpe

-0.78

Win rate

28.0%

Trades

322

Open the report →

Beating a coin that fell 90%

Holding the perpetual lost money on 19 of the 20 new coins, with a median of -87.93%; only AXS gained. Both versions of the breakout ended above holding on the same 19 coins. That says more about the coins than about the rule: a strategy that is out of the market about half the time and risks 1% per trade will beat almost anything that fell 90%. Holding the perpetual could lose even where the coin rose: RUNE's price ended 20.92% higher, but holding its perpetual lost 29.10% and paid 27,244 USDT of funding; the alt version made +150.94%.

Backtest

Perp long 1x, re-sized monthly (benchmark)

RUNEUSDT · 4h · Futures · 2020-01-01 → 2026-10-01

Return

-29.10%

Max drawdown

98.68%

Sharpe

0.54

Win rate

41.1%

Trades

73

Open the report →

On Sharpe ratio, the breakout beat holding on only 7 of 20 new coins (v2) and 10 of 20 (the alt version). Holding had a median Sharpe ratio of 0.18 on coins that lost a median 87.93%: the ratio is built from the average daily return, and on a coin whose price swings about 100% a year or more the average day can be slightly positive while the compounded result is a near-total loss. Neither comparison is the pass bar; the bar compares the rule with itself.

Where the money came from

On the seen coins longs carried the breakout. On the new ones they did not: v2's longs lost 8,136 USDT summed over the 20 coins and made money on only 7 of them, while its shorts made 32,125 USDT and made money on 15. The new coins were mostly coins that fell, and the short side, with its 3R take-profit, is what kept the totals positive. The alt version's wider long trail turned its longs into a small overall profit (22,957 USDT, 10 of 20 coins) and beat v2's Sharpe ratio on 16 of the 20 — the one change picked on altcoins did carry over as an improvement over v2, just not far enough to pass. So did the change that made v2: against the first version, the 3R take-profit for shorts raised the Sharpe ratio on 22 of the 23 seen coins and on 17 of the 20 new ones (dry runs of v1, which made money on only 9 of the new coins).

The 7 new coins that still trade did better than the 13 delisted ones — for the alt version, a median Sharpe ratio of 0.46 against 0.21 — but even the survivors fell short of the 0.67 on the seen coins.

The two halves

Every coin and both versions were re-run as dry runs on each half of the window, split at 2023-01-01. Each half warms its indicators up on its own first 199 bars, about 33 days.

On the coins it was tuned on, the rule held up over time. Both versions made money on all five of the first coins in both halves, and the alt version on 20 of the 23 seen coins in each. The gap between seen and new coins is there in both halves as well — for the alt version a median Sharpe ratio of 0.76 against 0.18 in 2020–2022 and 0.69 against 0.29 in 2023–2026 — so the weak holdout is not one bad stretch of time. YFII, BZRX and SRM were delisted before 2023, which leaves 17 new coins in the second half.

On BTC, v2 did better in the second half than in the first: a Sharpe ratio of 1.67 against 0.99.

Conclusion

• On the BTC perpetual it was tuned on, the breakout made +249.40% with an 11.99% max drawdown — less than holding (+329.98%), with twice the Sharpe ratio and about a sixth of the drawdown.
• On 23 coins seen while it was tuned, the alt version made money on 22 with a median Sharpe ratio of 0.67.
• On 20 coins picked by a rule fixed beforehand it made money on 15, but its median Sharpe ratio halved to 0.33 and v2's fell to 0.17. Neither met the bar set before the test. What travelled was the risk control — drawdowns stayed near 20–25% — not the returns.

Other trend rules on crypto: moving-average crossovers and the golden cross on the top 20 coins.

Notes

1. "Donchian channel", Wikipedia: upper band = the highest high over the last N periods, lower band = the lowest low, N usually 20. Read on 2026-10-30. ↩

2. "Richard Dennis", Wikipedia: the Turtle experiment began in 1984 and taught buying when prices rose above their recent range. Read on 2026-10-30. ↩