Parameter Scan

Find the best strategy parameters automatically

hz.quick.scan() tests every combination of parameters you specify and ranks them by Sharpe.

Basic scan

python
import horizon as hz

scan = hz.quick.scan(
    "mean_reversion",
    params={
        "window": [10, 15, 20, 30],
        "entry_z": [1.5, 2.0, 2.5, 3.0],
    },
    tickers=["AAPL", "MSFT", "NVDA"],
)

print(scan.summary())
print(f"\nBest: {scan.best_params}")
print(f"Best Sharpe: {scan.best_result.sharpe:+.3f}")

This runs 16 backtests (4 windows x 4 entry thresholds) on the same data and shows the ranking.

Reading results

python
# Best parameters
scan.best_params         # {"window": 20, "entry_z": 2.0}

# Best backtest result
scan.best_result.sharpe
scan.best_result.max_drawdown
scan.best_result.n_trades

# All results (dict of label -> BacktestResult)
for label, result in scan.results.items():
    print(f"{label}: Sharpe={result.sharpe:+.3f}")

# Summary table (top 20 by Sharpe)
print(scan.summary())

Scanning different strategies

python
# Momentum lookback
scan = hz.quick.scan("momentum", params={"lookback": [5, 10, 20, 40, 60]})

# MA crossover windows
scan = hz.quick.scan("ma_cross", params={"fast": [5, 10, 15], "slow": [20, 30, 50]})

# RSI window and thresholds (pass any strategy-accepted params)
scan = hz.quick.scan("rsi", params={"window": [7, 14, 21]})

With regime data

Test parameter robustness across market regimes:

python
scan = hz.quick.scan(
    "mean_reversion",
    params={"window": [10, 20, 30], "entry_z": [1.5, 2.0, 2.5]},
    data="regimes",
    bars=500,
)

Parameters that work well on smooth data may fail under regime switches. Always validate with data="regimes".

Warning: overfitting

A parameter scan on synthetic data tells you which parameters work on that specific random path. It does not tell you which parameters will work in live trading.

To get honest estimates:

  1. Use hz.quick.scan() to narrow the search
  2. Take the best parameters and validate with Bootstrap confidence intervals
  3. Or use Walk-Forward to test out-of-sample

The scan is a research tool, not a deployment decision.

Next

Quick Mode overview All Quick Mode features. Full pipeline When you need more control.