Strategy Parameters

Strategy parameters are discovered in gordon-lab (Python walk-forward research), then ported to gordon-strategy (Rust). The lab validates candidates; the Rust crate executes them live. These are two separate layers — never reimplement strategy logic in Python for production.

How parameters flow

gordon-lab (Python)
  1. Sweep parameter space
  2. Walk-forward validation across BTC/ETH/SOL minimum
  3. Ablation: every parameter must justify its value

gordon-strategy (Rust)
  4. Implement Strategy::evaluate
  5. Configure validated params
  6. Execute via gordon-bot (live) or gordon-manager (backtest)
     — same code path, enforced by BacktestExecution parity test

Validated strategies (gordon-strategy 4.x retrospective)

Supertrend 3.0

Best Sharpe: 0.90 | Timeframes: D1–8h

Parameter	Type	Range	Description
atr_period	int	10–20	Lookback for ATR calculation
multiplier	float	2.0–4.0	ATR multiplier for band distance

EWMAC Ensemble

Best Sharpe: 0.90 | Timeframes: 6h–1h

Parameter	Type	Default	Description
speed_pairs	list	[(8,32),(16,64),(32,128),(64,256)]	Fast/slow EMA combinations
forecast_cap	float	20.0	Maximum absolute forecast value
target_forecast_sd	float	10.0	Target standard deviation for forecast scaling
atr_period	int	20	ATR lookback for volatility normalization

PSAR Tight

Best Sharpe: 0.69–1.43 | Timeframes: W1–D1

Parameter	Type	Range	Description
af_start	float	0.02–0.03	Initial acceleration factor
af_step	float	0.02–0.03	Acceleration increment per new extreme
af_max	float	0.20–0.30	Maximum acceleration factor

Common parameters

These apply across all strategies:

Parameter	Type	Description
timeframe	string	Candle timeframe (1h, 4h, 1d, etc.)
symbol	string	Trading pair (BTCUSDT, ETHUSDT, etc.)

Active engines added in 4.6.x (validation in progress)

11 additional engines are implemented and active in gordon-strategy::engines. Walk-forward validation is in progress — none are production-deployed yet.

Engine	One-line description
BasisRevertStrategy	Mean-reversion on the perp–spot funding spread
BollingerBreakoutStrategy	Bollinger squeeze→expansion momentum breakout
FundingArbStrategy	Single-leg crowded-trade mean reversion on extreme funding rates
HurstRegimeSwitcherStrategy	Meta-strategy selecting a sub-strategy based on Hurst exponent (trending vs. mean-reverting vs. choppy)
OpeningRangeBreakoutStrategy	Intraday momentum entry on the UTC-day opening range high/low
PairsMrStrategy	Relative-value mean reversion on cointegrated crypto pairs (e.g. ETH/BTC)
RsiMeanReversionStrategy	Wilder RSI oversold/overbought entries gated on a 200-period SMA trend filter
TsMomStrategy	Time-series momentum based on Moskowitz, Ooi, Pedersen (2011)
VolumeSpikeBreakoutStrategy	Intraday entry on volume surge combined with price range break
VpinImpulseStrategy	Directional entries triggered by order-flow toxicity (VPIN) spikes
VwapStretchMrStrategy	Fade overextensions from the session VWAP (resets at UTC midnight)

Retired

• Donchian Ensemble — median Sharpe 0.08 in the v4 retrospective. Removed from active rotation. Not worth the complexity.

Validation gate (promotion from lab to production)

1. Walk-forward consistency > 50% on BTC/ETH/SOL minimum.
2. Net-of-fees Sharpe > 0.5 (Level 1).
3. Max drawdown < 30%, positive CAGR with full cost model (Level 2).
4. Port to Rust in gordon-strategy. Byte-parity test must pass.
5. 30+ paper trades on Binance testnet via gordon-bot.
6. Micro live allocation, then full allocation.

Avoiding overfitting

• Multi-asset validation: parameters must work on at least 3 assets.
• Walk-forward windows only — never random train/test splits on time series.
• Ablation: removing a parameter that does not degrade performance means it is noise.
• 1-bar delay enforced: signal on bar i, execution on bar i+1.