synthetic_backtesting

Subject

Sampling, Validation and ML Diagnostics

Why This Module Exists

AFML Chapter 13 shows that selecting PT/SL rules on a single historical path is prone to overfitting; synthetic path ensembles let us evaluate rule robustness under calibrated process dynamics.

Mathematical Foundations

Discrete O-U (AR(1))

$P_t=\alpha+\phi P_{t-1}+\sigma\epsilon_t,\quad \epsilon_t\sim\mathcal N(0,1)$

Equilibrium Level

$\bar P=\frac{\alpha}{1-\phi}$

OTR Objective over Rule Mesh

$R^*=\arg\max_{R\in\Omega}\frac{\mathbb E[\pi\mid R]}{\sigma[\pi\mid R]}$

Usage Examples

Rust

End-to-end synthetic OTR workflow

use openquant::synthetic_backtesting::{run_synthetic_otr_workflow, StabilityCriteria, SyntheticBacktestConfig};

let cfg = SyntheticBacktestConfig {
  initial_price: historical_prices[historical_prices.len() - 1],
  n_paths: 10_000,
  horizon: 128,
  seed: 42,
  profit_taking_grid: vec![0.5, 1.0, 1.5, 2.0, 3.0],
  stop_loss_grid: vec![0.5, 1.0, 1.5, 2.0, 3.0],
  max_holding_steps: 64,
  annualization_factor: 1.0,
  stability_criteria: StabilityCriteria::default(),
};

let out = run_synthetic_otr_workflow(&historical_prices, &cfg)?;
if out.diagnostics.no_stable_optimum {
  println!("Skip OTR optimization: {}", out.diagnostics.reason);
} else {
  println!("Best PT/SL: {:?}", out.best_rule);
}

API Reference

Rust API

• calibrate_ou_params
• generate_ou_paths
• evaluate_rule_on_paths
• search_optimal_trading_rule
• detect_no_stable_optimum
• run_synthetic_otr_workflow

Implementation Notes

• Near-random-walk estimates (|phi| close to 1) often produce flat Sharpe heatmaps where any selected rule is unstable out-of-sample.
• Calibrating to process parameters and evaluating many synthetic paths reduces single-path lucky-fit risk compared to brute-force historical optimization.