agentsclimarketplace

Vs benchmark audit

Skill rgourley/quant-garage/skills/vs-benchmark-audit

Analyst workflows as Claude skills. 62+ tools and 8 workflows spanning earnings, comps, valuation, options flow, factor research, sizing, risk, TCA, and ops. Built in the garage, not the trading floor.

Install
npx -y skills add rgourley/quant-garage --skill vs-benchmark-audit

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 6 stars6 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Take a book (weights per ticker), compute the daily portfolio return series, and run the full tearsheet with deflated Sharpe correction (Bailey & Lopez de Prado) plus rolling IC vs benchmark. Emits a verdict (real_alpha / possibly_alpha / essentially_beta / underperforming / no_edge_evident) based on DSR significance, alpha annualized, and beta. Answers "is this book actually alpha, honestly?" Requires Stocks Basic.

SKILL.md

2.7 KB, as published. Nobody here has run it

vs-benchmark-audit

You hand over a book (weights per ticker) and a benchmark (default SPY). The skill pulls daily bars, computes the portfolio return series, and runs the full performance tearsheet with the deflated Sharpe correction, plus a rolling 63-day IC vs the benchmark.

Answers "is this book actually alpha, honestly?" — with a verdict that separates real alpha from beta from noise.

When to invoke

  • Post-quarter review: did my strategy add anything above beta?
  • Investment committee prep on a candidate manager or strategy
  • Auditing a historical backtest with proper DSR correction
  • The user says "vs benchmark", "alpha vs beta", "is this real"

What you need

  • Positions (--positions T=w,T=w,...)
  • MASSIVE_API_KEY exported
  • Stocks Basic minimum

Optional:

  • --benchmark (default SPY)
  • --lookback-days (default 504, 2 years)
  • --ic-window (default 63, one quarter)
  • --n-trials-dsr (default 1): multiple-testing correction for Deflated Sharpe. Pass N if this book was picked from N candidates during search.

What you get back

Layer 1: JSON. Full tearsheet (CAGR, Sharpe, DSR, Sortino, Calmar, max DD, ulcer, tail ratio, profit factor, hit rate, beta, alpha, tracking error) plus rolling IC mean and std vs benchmark. Top-level verdict.

Layer 2: rendered note. Header verdict + return stats block + vs-benchmark block + Take.

How it works

  1. Pull daily bars for each position and the benchmark.
  2. Align to common dates.
  3. Compute daily portfolio returns (weighted sum of position returns, renormalized to abs-weights = 1).
  4. Run quant_garage.performance.tearsheet with benchmark kwarg populated so beta / alpha / tracking error come through.
  5. Compute rolling ic_window-day Pearson IC of portfolio vs benchmark returns for a time-varying correlation lens.
  6. Emit verdict:
    • real_alpha: DSR significant at 5% AND alpha > 2% annualized
    • possibly_alpha: alpha > 0 AND Sharpe > 0.5 but DSR not sig
    • essentially_beta: beta > 0.8 AND |alpha| < 2%
    • underperforming: annualized return < 0
    • no_edge_evident: everything else

Foundations used

  • quant_garage.performance.tearsheet
  • quant_garage.backtest.rolling_ic_series
  • massive-api-patterns

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.