Run a walk-forward analysis to validate strategy robustness out-of-sample. Walk-forward divides a historical period into N sequential windows and backtests each to measure consistency across regimes.

Workflow

walk_forward(strategy_id, total_lookback="1yr", windows=5)   # → walk_forward_id
get_walk_forward(walk_forward_id)                            # poll until completed

Level 1 (consistency check) — no sweep; runs the strategy as-is on each window. Reports oos_sharpe_mean, oos_sharpe_std, pct_profitable_windows. High mean + low std = robust across regimes; pct_profitable_windows > 0.75 is strong, < 0.5 is concerning.

Level 2 (rolling re-optimization) — requires a sweep config. Each window splits into in-sample (IS) and out-of-sample (OOS): the sweep runs on IS, params are selected by select_by (default sharpe), then applied to the OOS period. Reports Walk-Forward Efficiency (WFE = OOS Sharpe / IS Sharpe: > 0.5 strong, 0.3–0.5 acceptable, < 0.3 suggests overfitting) plus parameter stability — params that hold steady across windows suggest real signal.

Portfolios — pass portfolio_id instead of strategy_id. Portfolio walk-forward is Level 1 only (one coupled backtest per window with a per-member breakdown); it does not support sweep / overrides / oos_pct / select_by.

Key params — total_lookback (e.g. "1yr", "2yr", "6mo"), windows (default 5), mode (rolling or anchored, default rolling), oos_pct (Level 2, default 0.3), initial_cash (default 100000), fee_pct / slippage_pct (default 0; fee auto-detected by asset class).