Appendix V — Data Card — SEC EDGAR 10-K / 10-Q (Annual & Quarterly Reports)

Provider & what it is

The U.S. Securities and Exchange Commission (SEC) runs EDGAR (Electronic Data Gathering, Analysis, and Retrieval), the public archive where every domestic public company files its mandatory disclosures. The 10-K is the comprehensive annual report: audited financial statements plus the long prose sections — Item 1 (Business), Item 1A (Risk Factors), Item 7 (Management’s Discussion and Analysis, “MD&A”), and Item 7A (market risk). The 10-Q is the lighter quarterly report filed for Q1–Q3, with unaudited condensed financials and a shorter narrative. Together they are the workhorse of text-based and fundamentals-based finance research: one document carries both the numbers (revenue, assets, debt) and the words (how management frames risk and performance). This is the source behind Priya’s “climate-risk language in insurer 10-Ks” idea and Week 6’s reader’s guides on 10-K text (Hoberg–Phillips TNIC, Loughran–McDonald sentiment).

Coverage

EDGAR holds electronic filings from 1993–1994 to the present; mandatory electronic filing phased in over 1993–1996, so coverage before ~1996 is partial. Roughly every U.S.-listed operating company files a 10-K annually and three 10-Qs per year. Foreign private issuers file a 20-F (annual) and 6-K (interim) instead — a trap if you assume everyone files 10-Ks. Smaller reporting companies get scaled disclosure but still file the same form types. [CHECK] the exact 1993-vs-1994 cutoff for the earliest fully-electronic filings if your sample reaches that far back.

Key identifiers (CIK / accession)

  • CIK (Central Index Key): EDGAR’s permanent company identifier, an integer zero-padded to ten digits in URLs (0000320193 = Apple). Unlike a ticker, it never changes on rename or re-listing.
  • Accession number: identifies one specific filing, formatted 0000320193-23-000106. This is the pin that makes a citation exact — see Gotchas on amendments.
  • A ticker maps to a CIK through EDGAR’s company_tickers.json. Mapping CIK onward to CRSP’s PERMNO or Compustat’s GVKEY is a separate crosswalk (Chapter 7.4), not something EDGAR gives you.

Access path

EDGAR is free and public — no API key, no registration. What it does require is the SEC fair-access rule: every HTTP request must carry a descriptive User-Agent header with your name and email, e.g. "Priya Nair priya@gmu.edu". A missing or generic User-Agent earns a 403 Forbidden. The documented rate ceiling is at most 10 requests/second, aggregated across all your scripts; exceed it and your IP is throttled.

High-level route — the edgartools package wraps EDGAR’s messy URLs into objects:

from edgar import Company, set_identity
set_identity("Priya Nair priya@gmu.edu")     # the fair-access identity; read from env in real code

co = Company("AAPL")
tenk = co.get_filings(form="10-K").latest()   # most recent annual report
mdna = tenk.markdown()                         # full text, including MD&A and risk factors
fin  = tenk.financials                         # structured XBRL financials (see the XBRL card)

Low-level route — plain requests against EDGAR’s submissions endpoint when you want one JSON file and no dependency:

import os, requests, time
HEADERS = {"User-Agent": os.environ["SEC_USER_AGENT"]}   # e.g. "Priya Nair priya@gmu.edu"
cik = "0000320193"
url = f"https://data.sec.gov/submissions/CIK{cik}.json"   # filing index for the company
resp = requests.get(url, headers=HEADERS, timeout=30); resp.raise_for_status()
filings = resp.json()                                      # form types, accession numbers, dates
time.sleep(0.15)                                           # ~7 req/s, comfortably under 10/s

The trick edgartools hides: a filing is a folder of documents (the primary HTML, exhibits, the XBRL instance), addressed by CIK and accession number at https://www.sec.gov/Archives/edgar/data/<cik>/<accession-no-dashes>/. For large loops, prefer edgartools — it builds in the politeness throttling for you.

License & note

FREE / public domain. SEC filings carry no copyright and no license restriction — unlike CRSP/Compustat, this data may be cached, committed, and shared. The only obligation is operational: identify yourself (User-Agent) and stay under 10 req/s. There is no GMU-infrastructure rule here. The practical limit is size, not law: full-text caches grow fast, so store raw HTML/text under a data/raw/edgar/ directory and commit only if size permits.

Gotchas

  • Amendments. A 10-K can be amended as a 10-K/A, and a 10-Q as 10-Q/A. “Apple’s FY2023 10-K” is ambiguous; accession 0000320193-23-000106 is exact. Always pin and record the accession number, and decide explicitly whether you use the original or the amendment.
  • Boilerplate and incorporation by reference. Large stretches of a 10-K — especially Item 1A risk factors — are repeated near-verbatim year to year, and items can be incorporated by reference from the proxy (DEF 14A) rather than printed, so the section you want may not be in the document you pulled.
  • Parsing. Modern filings are HTML with inline XBRL tags woven in; older ones are plain text or crude HTML. Naive BeautifulSoup stripping leaves table gibberish and XBRL artifacts. Extract section-by-section (by Item heading), not by blind tag-stripping, and expect to clean.
  • 10-Q financials are unaudited and use condensed statements; do not treat a 10-Q balance sheet as comparable in rigor to the audited 10-K.

Structure sketch (illustrative — not real values)

A “filings index” table for one company, the shape you iterate over:

cik accession_no form filing_date period_of_report primary_doc
0000320193 0000320193-23-000106 10-K 2023-11-03 2023-09-30 aapl-20230930.htm
0000320193 0000320193-23-000077 10-Q 2023-08-04 2023-07-01 aapl-20230701.htm
0000320193 0000320193-23-000064 10-Q 2023-05-05 2023-04-01 aapl-20230401.htm
0000320193 0000320193-23-000006 10-Q 2023-02-03 2022-12-31 aapl-20221231.htm
0000320193 0000320193-22-000108 10-K 2022-10-28 2022-09-24 aapl-20220924.htm

(Illustrative. Real accession numbers, dates, and document names differ; verify against the live filing index.)

Where this is used

  • Week 6 opening narrative (“When the data is words”) and the text reader’s guides — Hoberg–Phillips (Ch 6.2, 10-K cosine similarity) and Loughran–McDonald (Ch 6.3, finance sentiment dictionaries).
  • Chapter 7.2 (Data Acquisition in Practice) §7.2.3 — the narrated EDGAR tour this card is the spec sheet for.
  • Priya’s running example (climate-risk 10-K language linked to returns) and any capstone using disclosure text or fundamentals.