Does an academic Donchian trend ensemble beat holding BTC and ETH?
Short answer: Rebuilt on Binance daily data with no leverage, the nine-lookback Donchian ensemble made 0.8% on BTC and 7.9% on ETH over the 18 months the paper could not have tuned on, while a one-line 200-day trend filter made 30% and 42% with lower drawdowns. Since 2018 it compounded at about half the rate of holding. The paper's volatility target is the reason: it leaves a spot account 50 to 65% in cash.
A 2025 paper from three academic authors tested trend following on 21,616 crypto assets from 2010 to 2025 and reported a net Sharpe ratio above 1.5 for an ensemble of daily Donchian breakouts with volatility-scaled positions. It is the best-documented trend claim we have seen for crypto: real costs, a survivorship-free universe, fixed lookbacks. So we asked the question a small account cares about: does the rule pay in a spot wallet with no leverage?
How we tested
- The rule was written and timestamped before any data was pulled. Nothing was tuned.
- Nine sub-strategies, one per lookback (5, 10, 20, 30, 60, 90, 150, 250 and 360 days), each with a ninth of the capital. Buy at the next open when the daily close exceeds the highest close of the previous L days. Exit at the next open when the close falls below a trailing stop that only rises (the midpoint of the L-day high and low).
- Position size at entry targets 25% annual volatility from the trailing 90-day figure, capped at 100% of the sub-strategy’s capital. The paper allows 200%; spot cannot.
- Cost 0.15% per side. Long only. Cash earns nothing.
- Benchmarks with the same costs: holding, and a daily 200-day EMA filter (in when the close is above it, out otherwise).
- The paper’s data ends in March 2025. Verdict on April 2025 to September 2026, the part the paper could not have seen, with the full window since September 2018 reported alongside.
- Pass rule, per coin: return over worst drawdown above both benchmarks, and a daily excess return over holding whose 95% lower bound stays above zero.
Results, April 2025 to September 2026
| Net | Worst drawdown | Time invested | Round trips | |
|---|---|---|---|---|
| BTC ensemble | +0.8% | 14.9% | 19% | 97 |
| BTC holding | +1.1% | 53.0% | 100% | 0 |
| BTC 200-day filter | +30.0% | 20.6% | 43% | 6 |
| ETH ensemble | +7.9% | 15.8% | 12% | 98 |
| ETH holding | +47.2% | 67.6% | 100% | 0 |
| ETH 200-day filter | +42.1% | 26.6% | 38% | 5 |
The daily excess over holding was negative on both coins, with 95% lower bounds of -0.16% a day for BTC and -0.35% for ETH. Both fail the rule.
Since September 2018
| Net | A year | Sharpe | Worst drawdown | |
|---|---|---|---|---|
| BTC ensemble | +325% | 19.6% | 1.03 | 26.2% |
| BTC holding | +1,091% | 35.8% | 0.81 | 76.6% |
| BTC 200-day filter | +1,266% | 38.2% | 0.95 | 56.5% |
| ETH ensemble | +223% | 15.6% | 0.84 | 31.4% |
| ETH holding | +853% | 32.1% | 0.76 | 79.3% |
| ETH 200-day filter | +1,391% | 39.7% | 0.87 | 72.8% |
On the long window the ensemble does what the paper says: on BTC its Sharpe (1.03) beats holding (0.81) and the filter (0.95), with a third of the drawdown of holding. It would still fail our rule on both coins, because it compounded at about half the rate.
Why it trails
The volatility target does the damage. BTC and ETH ran at 50 to 70% realised volatility, so a 25% target puts each sub-strategy in with only 35 to 50% of its capital, and a spot account cannot borrow the rest. Averaged over time the ensemble held about 20% of capital in the market. The paper earns its Sharpe on a levered, rotating basket of twenty coins. The Sharpe survives the move to a spot wallet; the return shrinks by about half.
The fee bill is the second weight. The two shortest lookbacks traded 122 to 239 round trips each since 2018 and paid 33 to 84% of their starting capital in costs. The longer lookbacks (60 and 90 days) were the ones that paid on BTC, with 627% and 459%; alone, they still trail the one-line 200-day filter.
Limits
- Eighteen months is short for a daily trend rule. A fail on that window says the rule did not pay recently; the long window is shown for that reason, and it points the same way.
- No leverage and a fixed size at entry differ from the paper, so this is a test of the rule in a spot account, and says nothing against the paper’s own numbers.
- Two coins. The paper’s top-20 rotation is a different, larger portfolio.
Verdict
FAIL. In a spot account the nine-lookback Donchian ensemble trailed a plain 200-day trend filter on BTC and ETH, on the 18 months after the paper’s data and on the eight years before it. Not financial advice.