Phase 2b: Full walk-forward backtest, Monte Carlo, and Sharpe gate

[JLaborda]

Parent

#85

PRD: spec/prds/phase2/prd.md · Spec: spec/features/001-backtesting/spec.md

Spec

spec/features/001-backtesting/spec.md

What to build

Full backtest per architecture spec: 20+ years, walk-forward 3–5y windows, block-bootstrap Monte Carlo, crisis drawdown report, Sharpe compared to S&P 500 CW/EW, Russell 3000, and Magic Formula replica. pass flag only if strategy Sharpe beats all benchmarks on OOS window. Long runs on ECS/Batch, not PR CI.

Acceptance criteria

• Walk-forward windows roll annually with documented hold-out
• Monte Carlo block-bootstrap distribution logged to MLflow
• Crisis report HTML for dotcom, GFC, COVID, 2022 rate shock
• Sharpe gate vs four benchmarks; overfit_risk flag on IS/OOS gap
• Uses historical S&P 500 universe slice

Blocked by

• Phase 2b: Historical S&P 500 universe with delisted names #99
• Phase 2: Light backtest engine (5–10y QV) #96