What problem does it solve?
Backtested or live strategy returns often look impressive until you ask whether a simple market, equal-weight, or momentum baseline explains them. This Skill runs returns-based attribution to test whether a strategy's performance contains statistically supported residual edge beyond predeclared baselines, producing an auditable verdict instead of a gut-feel judgment.
Core Features & Use Cases
- Returns-Based OLS Attribution: Fits an intercept OLS model of strategy returns on declared baseline return series, reporting annualized alpha, loadings, R-squared, and Newey-West (HAC) t-statistics.
- Robustness Gates: Runs rolling-window stability analysis, alternate-baseline sensitivity models, VIF multicollinearity checks, residual autocorrelation diagnostics, and predeclared regime breakdowns, then assigns one of four statuses: RESIDUAL_EDGE, BASELINE_EXPLAINED, RESIDUAL_FRAGILE, or INSUFFICIENT_EVIDENCE.
- Provenance Enforcement: Requires explicit declarations for baseline selection timing, gross/net cost basis, analysis scope, and universe data, downgrading results to REVIEW_REQUIRED when provenance is missing.
- Use Case: After backtesting a swing-trading strategy, feed a CSV of dated strategy, market, and equal-weight returns plus a JSON config to learn whether the apparent alpha survives comparison with simple baselines, or whether a drawdown came from baseline exposure rather than strategy-specific behavior.
Quick Start
Ask the assistant to run the residual edge analyzer on your dated strategy-and-baseline returns CSV with a JSON config declaring the primary and sensitivity baseline models, then review the generated JSON and Markdown attribution report.
