Skip to content

Donchian Ensemble (Retired)

Retired from production

Donchian was retired from the active portfolio in the v4 retrospective. It is kept here for historical reference.

Retirement rationale

The Donchian ensemble produced a median Sharpe of 0.08 across the full walk-forward validation sweep — too thin an edge to justify the added complexity in the portfolio. The three validated strategies (Supertrend, EWMAC, PSAR) cover the portfolio's trend-following surface without it.

The strategy is not deleted from gordon-strategy — the implementation remains at src/engines/ (not present, as it was from the pre-split monolith era) and historical backtest runs are preserved in the database. It is not registered in the current StrategyRegistry and will not appear in the console's strategy selector.

What it was

Channel breakout strategy using 9 lookback periods (5, 10, 20, 30, 60, 90, 150, 250, 360). A combo score averaged the signals from all nine sub-strategies:

combo_score = mean(signal_5, signal_10, ..., signal_360)

Position changes triggered when the combo score crossed a threshold (±0.33). The ensemble was designed to require multi-lookback consensus before acting, reducing false breakouts.

It had academic backing — channel breakout systems are among the most studied trend-following approaches (Zarattini et al., 2025 provides recent crypto futures evidence). The strategy proved robust in aggregate validation but thin in portfolio context where the three validated strategies already captured the available trend-following edge.

Historical performance

MetricValue
Mean Sharpe (full sweep)0.08
Best Sharpe observed0.56
Timeframes testedAll
Assets validatedBTC, ETH, SOL

The 0.56 best-case Sharpe was the figure cited during v3 research. The median across the walk-forward sweep was 0.08, which drove the retirement decision.

Gordon — keep compounding without blowing up