vs-benchmark-audit

Stars6

Forks0

UpdatedJuly 12, 2026 at 23:11

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.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Source

rgourley

rgourley/quant-garage

View GitHub Repository View Creator Repositories

Download

Related occupationsSOC

Based on SOC occupation classification

Financial and Investment AnalystsBusiness and Financial Operations Occupations·SOC 13-2051

File Explorer

5 files

SKILL.md

readonly

name

vs-benchmark-audit

description

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

Pull daily bars for each position and the benchmark.
Align to common dates.
Compute daily portfolio returns (weighted sum of position returns, renormalized to abs-weights = 1).
Run quant_garage.performance.tearsheet with benchmark kwarg populated so beta / alpha / tracking error come through.
Compute rolling ic_window-day Pearson IC of portfolio vs benchmark returns for a time-varying correlation lens.
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

More from this repository

same repository

commodity-cycle

rgourley/quant-garage

Single-commodity macro read. Answers "is this commodity in a winning or losing macro setup right now" and names the macro driver that dominates it. Pulls one commodity ETF (default GLD; accepts SLV, USO, DBC, or any commodity ETF) plus the macro context it needs (UUP for the dollar, TIP and IEF for a real-yield proxy, and GDX/SLV for gold's miner and silver co-move set) and computes the drivers that push the commodity around: rolling DXY correlation, real-yield correlation (vs the TIP-minus-IEF spread), miner divergence (gold only), silver co-movement (gold only), and a momentum quintile. The take reads constructive / neutral / headwind and names the dominant variable. It would have flagged a gold drawdown two weeks early via the dollar and real-yield readings. Use when the question is about one commodity specifically ("is gold in trouble", "what's driving crude", "is the dollar the problem for gold"). Runs on any stocks tier (Free Basic works with --sleep 13).

2026-07-186

single-name-vs-sector

rgourley/quant-garage

Split one name's relative strength into two clean legs, name vs its sector ETF and sector vs benchmark, so you can tell whether the name is strong because its sector is strong or because it is pulling away from its own sector. Maps the ticker to its SPDR sector ETF (XLK, XLF, XLE, XLV, XLI, XLY, XLP, XLU, XLB, XLRE, XLC) with a --sector override for any name. Computes name-vs-sector, sector-vs-benchmark, and name-vs-benchmark RS in basis points across 5/20/60/120-day windows, a divergence score, and a take that classifies the name as leading its sector, lagging its sector, or diverging. Use when a single name looks strong or weak vs SPY and you need to know if it is name-specific or sector-driven. Requires Stocks Starter (Free Basic works, only three series per run).

2026-07-186

hedge-suggester

rgourley/quant-garage

Proposes concrete, live-priced option hedges against a single long position and ranks them by cost per dollar of downside protected. Takes a ticker and a position size (shares or notional), pulls the underlying price and the options chain around the horizon expiry, and constructs five standard overlays (covered call, protective put, collar, put spread, ratio put spread) priced from chain mids with net cost, breakeven, max loss, max gain, and net delta/gamma/theta at open. The take line recommends the structure that fits the stated risk tolerance. Use when a concentrated long needs a hedge and you want the actual structures and their live cost, not just "you are over-exposed." Needs Stocks Starter plus the Options Developer add-on. Not advice.

2026-07-186

portfolio-macro-scenario

rgourley/quant-garage

Prescriptive macro scenario analysis on a current book. Given a position book (CSV of ticker,shares) and a scenario stated as flags (--rates-bp, --dxy-pct, --oil-pct, --gld-pct), it regresses each position's daily returns on four macro factor ETFs (TLT for rates, UUP for the dollar/DXY, USO for oil, GLD for gold), translates the scenario into factor return shocks, and reports the expected per-position and book-level P&L with a rough 90% band, plus the dominant position and factor contributors. Use when the question is forward-looking and conditional: "what happens to my book if rates keep rising / the dollar keeps rallying / oil spikes / gold falls." This is the prescriptive counterpart to risk-report and portfolio-review, which are descriptive of the past. Runs on any stocks tier (Free Basic works with --sleep 13).

2026-07-186

prediction-market-monitor

rgourley/quant-garage

Pull Kalshi prediction market prices for Fed decisions, CPI, GDP, NFP, and other macro / market events. Report implied probability per outcome, aggregate cross-strike distribution when the series is a laddered strike set (like KXFED-27APR-T4.25, T4.00, T3.75...), expected value, modal outcome, and open interest. Prediction markets now clear enough volume post-2024 to reflect a real market-implied policy path, often diverging from surveyed economist consensus. Uses Kalshi's public read-only API; no authentication required.

2026-07-126

pre-earnings-full-stack

rgourley/quant-garage

Workflow composite for a single ticker heading into an earnings print. Chains earnings-blackout (timing check) + event-study (prior print reaction distribution) + guidance-tracker (management raise/cut track record) + analyst-tracker (sell-side positioning) + mc-portfolio-simulator (P&L distribution at proposed weight over the horizon). Emits a posture verdict (constructive_setup / mixed_setup / avoid_or_hedge / no_imminent_print). Requires Stocks Basic; guidance-tracker and analyst-tracker sections skip gracefully without Benzinga entitlements.

2026-07-126

name

vs-benchmark-audit

description

vs-benchmark-audit

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

How it works

Pull daily bars for each position and the benchmark.
Align to common dates.
Compute daily portfolio returns (weighted sum of position returns, renormalized to abs-weights = 1).
Run quant_garage.performance.tearsheet with benchmark kwarg populated so beta / alpha / tracking error come through.
Compute rolling ic_window-day Pearson IC of portfolio vs benchmark returns for a time-varying correlation lens.
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