Manager Portfolio Diff

manager-portfolio-diff

You hand over a filer (an alias like "berkshire" or a raw CIK). The skill pulls the two most recent quarterly 13-F filings for that manager, aggregates holdings by CUSIP (handles multi-manager joint filings), and reports the quarter-over-quarter change: initiations, exits, adds, trims, and portfolio value delta.

The Massive endpoint doesn't support an issuer-oriented lookup ("who holds AAPL?"), so this is the correct shape for 13-F work: pick the filer whose decisions you want to see and diff their book.

When to invoke

• "What did Berkshire do this quarter?"
• "Is Baupost still in that name?"
• "Did Pershing Square initiate anything new?"
• Any smart-money-following workflow across a curated cohort of funds

Not for: "who owns AAPL?" (endpoint doesn't support issuer filter at scale). Not for real-time positioning (13-F is quarterly and lagged ~45 days after quarter-end).

What you need

• Either --filer (an alias) OR --filer-cik (10-digit CIK)
• MASSIVE_API_KEY exported
• Stocks Basic plan minimum

Known aliases:

• berkshire / buffett
• baupost / klarman
• renaissance / rentech
• bridgewater
• third point / loeb
• pershing / pershing square / ackman
• tiger global / coleman
• scion / burry
• appaloosa / tepper

For anyone else, look up their CIK at https://www.sec.gov/cgi-bin/browse-edgar.

What you get back

Two output layers from one run.

Layer 1: canonical JSON matching output-schema.json. Top-level filer_cik, filer_display_name, periods, summary (holding counts, portfolio value, activity counts), and changes with per-bucket entries (initiation, exit, add, trim, unchanged). Each entry carries issuer_name, cusip, prior_shares, current_shares, prior_market_value, current_market_value, delta_shares_pct.

Layer 2: rendered note. Header + activity counts, then four buckets (NEW POSITIONS / EXITED / ADDS / TRIMS) capped at 10 entries per bucket, sorted by market value. One-line Take highlighting the biggest new position and biggest exit. See references/rendering.md.

How it works

1. Pull all 13-F rows for the filer via GET /stocks/filings/vX/13-F?filer_cik={CIK}&limit=1000&sort=filing_date.desc.
2. Group by period (quarter-end YYYY-MM-DD). Take the two most recent periods as current and prior.
3. Aggregate by CUSIP within each period. Sum shares and market value across multiple manager-of-record rows (joint filings produce one row per manager).
4. Classify each CUSIP:
   • Only in current → initiation
   • Only in prior → exit
   • In both, share change >= +25% → add
   • In both, share change <= -25% → trim
   • Otherwise → unchanged
5. Sort each bucket by market value. New positions and adds by current value; exits and trims by prior value. Renders top 10 per bucket with overflow count.

Foundations used

• massive-api-patterns for REST auth, retry, pagination.

Output mode: note

Narrative note. A single fund's 13-F is 30-300 rows typically; a grouped-by-bucket note reads better than one wide table.

Endpoints used

• GET /stocks/filings/vX/13-F?filer_cik={CIK}&limit=1000&sort=filing_date.desc All 13-F rows for the filer. Paginated.

Doesn't handle (yet)

• Fund-of-funds cohort scan. No "top 20 hedge funds this quarter." A workflow composite could iterate over aliases.
• Multi-quarter trajectory. Only diffs two quarters. A rolling 4-quarter cluster (add-add-add-add vs churn) would be a great extension.
• Shorts / derivatives. 13-F is longs + long options + a few other instruments only. Shorts don't appear.
• Marked-to-market values. Uses filer-reported market_value at the period date. Not marked to today's price.
• Position-in-portfolio-percent. Adds are currently ranked on absolute market value, not share-of-portfolio. Both matter.

These are clean PR extensions. Output schema is forward-compatible.