Research Methodology

How strategies are validated before going live.

Validation Pipeline

  1. 6-fold Purged CV — chronological folds with 5-day gap to prevent label leakage.
  2. Pure OOS test — a held-out period never seen during hyperparameter search.
  3. Slippage stress — Sharpe must survive +2bp slippage on top of baseline.
  4. Deflated Sharpe Ratio (DSR) — corrects for multiple-testing bias from Optuna trials.
  5. CPCV (Combinatorial Purged CV) — finalists validated on 15+ OOS paths before promotion.
  6. Live drift monitoring — calibration, state occupancy, execution quality tracked in real time.

Research Ledger

Every experiment is recorded, including failed ones. The number of trials is used as the denominator in DSR calculation — honest bookkeeping is required for honest statistics.

Source: docs/experiments/research_ledger.json

Strategy promotion requires mean per-fold DSR ≥ 0.90 and passing all gates above. Drift flags automatically reduce risk if live execution diverges from backtest baseline.