Donchian Slow Turtle Backtest on BTC/USDT (2023, hourly, real fees)

Post 14 tested the "fast" Turtle — the 20-bar entry breakout. This is the original Turtle entry that Richard Dennis and William Eckhardt actually traded: buy a new 55-day high, sell on a 20-day low. On 2023 BTC it beat its own faster sibling in every cost scenario, which is exactly what the lab's frequency finding predicts. Strategy Lab #57.

The setup

Long when close closes above the highest high of the previous 55 bars; exit when close closes below the lowest low of the previous 20 bars. Long-only, same engine, same costs.

Results

The fee bill

| Scenario | Cost/leg | Total return | MaxDD | Sharpe | Trades | |---|---:|---:|---:|---:|---:|---:| | naive (zero cost) | 0.00% | +49.23% | -25.40% | 1.54 | 55 | | taker fee 0.05%/leg | 0.05% | +41.25% | -27.27% | 1.35 | 55 | | + funding 0.01%/8h | 0.05% | +36.25% | -28.14% | 1.22 | 55 | | + slippage 10bp/leg | 0.15% | +22.04% | -32.33% | 0.84 | 55 | | + slippage 25bp/leg | 0.30% | +3.44% | -38.16% | 0.26 | 55 |

+3.44% even at 25bp. The slow Turtle joins the short list of strategies (now six strong) that survive a 0.30% cost stack. Fifty-five trades is the sweet spot this lab has been triangulating all series: rare enough that 25bp is 16 points of annual cost, frequent enough to actually catch the year's trend. The 55-bar entry filters out the false breakouts that the fast Turtle's 20-bar entry loved to buy.

The Turtle sibling fight

| Version | Trades | Naive | Fees+funding | 25bp | MaxDD | |---|---:|---:|---:|---:|---:|---:| | Fast 20/20 (post 14) | 91 | +44.11% | +24.24% | -21.04% | -41.90% | | Slow 55/20 (this) | 55 | +49.23% | +36.25% | +3.44% | -28.14% |

The slow version trades 36 fewer times, makes more gross and more net, and halves the drawdown. The 55-bar breakout is not just cheaper — it's a better filter. This is the cleanest within-family control the breakout group has produced: same channel logic, slower entry, strictly better.

The breakout family, restated

The family now ranks almost perfectly by trade count. Donchian slow is the ranking's best evidence that deliberately slowing down an entry is a legitimate alpha source on its own.

What this does NOT prove

• The real Turtle system pyramided (added size every 0.5 ATR) and traded short side too; this is the plain long-only entry/exit skeleton.
• 55/20 is one point in a large parameter space; the fast version's worse result is consistent with the lab's pattern, not proof of a global optimum.
• One pair, one year, one regime.

Code

from backtest_base import fetch, backtest_signal, metrics
from strategy import sig_donchian_slow

df = fetch("BTCUSDT", "binance", "2023-01-01", "2023-12-31", "1h")
signal = sig_donchian_slow(df, entry_period=55, exit_period=20)

res = backtest_signal(df, signal, cost_per_leg=0.0005, funding_per_bar=0.0000125)
print(metrics(res, 8760))

Reproduce it

cd blog-drafts/scripts
python gen_post_assets.py --ids 57

Data: Binance public API, hourly OHLCV, 8,735 bars. Tables above reproduce exactly from this command.

This is a backtest on historical data, not investment advice. Past performance does not predict future results.