Risk factor delta

risk-factor-delta

You hand over a ticker. The skill pulls the most recent 10-K risk-factor disclosures and the one before it, diffs the standardized category taxonomy, and reports what management added, dropped, and materially rewrote year-over-year.

This is the "what changed in Item 1A?" read that fundamental analysts do by hand. It works because Massive already parses and categorizes risk factors from every 10-K into a three-tier taxonomy (primary, secondary, tertiary). No NLP on our end. No EDGAR text scraping.

When to invoke

• A fundamental analyst asks "did AAPL add anything new to Item 1A?"
• A credit analyst wants a heads-up on new balance-sheet or liquidity risks flagged for the first time this cycle
• A macro-driven investor scanning for regulatory / tariff / geopolitical risk-factor additions across a basket
• The user says "risk factor delta", "10-K diff", "what's new in Item 1A", "compare risk factors YoY"

Not for: single-filing risk catalog (fine as a fallback but the primary value is the delta). Not for prose-level word-for-word diff (this is a category-level diff with supporting text quoted for confirmation).

What you need

• A ticker (--ticker, required)
• MASSIVE_API_KEY exported in the environment
• Stocks Basic plan minimum. The /stocks/filings/vX/risk-factors endpoint is included on every Stocks plan.

Optional:

• --current-filing-date (YYYY-MM-DD): pin a specific "current" filing. Defaults to the most recent on record.
• --prior-filing-date (YYYY-MM-DD): pin a specific "prior" filing. Defaults to the second-most-recent on record.

What you get back

Two output layers from one run.

Layer 1: canonical JSON matching output-schema.json. Top-level: filings.current, filings.prior, and summary counts (added, removed, materially changed, retained unchanged). changes.added[], changes.removed[], changes.materially_changed[] each carry per-entry {primary_category, secondary_category, tertiary_category, supporting_text}. Materially-changed entries also include prior_supporting_text, both lengths, and length_delta_pct.

Layer 2: rendered narrative. Header with the delta counts, three sections (NEW / DROPPED / MATERIALLY CHANGED) grouped by primary category, each with the supporting-text quote so the reader can confirm the taxonomy call, followed by a one-line Take. See references/rendering.md.

How it works

1. Pull risk factors for the ticker via GET /stocks/filings/vX/risk-factors?ticker={T}&limit=50000&sort=filing_date.desc. Massive returns one row per unique (primary, secondary, tertiary) category per filing, with a supporting-text snippet.
2. Group by filing_date. Each 10-K filing produces N rows all sharing the same filing_date. Sort dates descending; pick the two most recent as current and prior (or use the caller-supplied dates).
3. Diff by category tuple. For every (primary, secondary, tertiary):
   • In current only → added
   • In prior only → removed
   • In both → check supporting_text length delta. >= 25% flip → materially changed. Otherwise retained unchanged.
4. Group results by primary category. The primary axis is the headline shape ("all the new categories are financial_and_market"). Secondary/tertiary render as bullets under it.
5. Take. One sentence summarizing counts and the concentration of new categories.

Massive's taxonomy comes from a published research paper linked in the endpoint docs; see references/methodology.md.

Foundations used

• massive-api-patterns for REST auth, retry, and pagination on the filings endpoint.

Output mode: note

Narrative note. This is a category-level diff on a small number of rows (10-K risk factors typically 15-40 per filing); a wide table would waste space. The rendered format optimizes for a fundamental analyst reading the delta once, then quoting the supporting text into a note or a call.

Endpoints used

• GET /stocks/filings/vX/risk-factors?ticker={T}&limit=50000&sort=filing_date.desc All categorized risk factors for the ticker across every 10-K on record. One paginated call.

Doesn't handle (yet)

• Sentence-level text diff. The skill reports a length delta as a "materially changed" proxy and quotes the current supporting text. A proper word-level diff (highlighting added/removed phrases) would be a clean PR extension.
• Cross-ticker roll-ups. A watchlist mode ("scan my 30 names for new regulatory risk factors YoY") would compose this skill and aggregate by primary_category. Queued.
• 10-Q updates. Item 1A can be amended in a 10-Q. The endpoint covers annual 10-K disclosures only for the diff. 10-Q updates are a separate lane.
• Historical trends. Only diffs two filings. A "risk factor trajectory over N years" view would surface which categories are chronic vs newly-appearing; queued.
• Peer comparison. No "what risks does AAPL cite that MSFT doesn't?" yet. The taxonomy makes this trivially composable; queued as a separate peer-risk-comparison skill.

These are clean PR extensions. The output schema is forward-compatible.