tcontorno/edgar-ma-deals

EDGAR M&A Deal Events (2010–2024)

4,156 U.S. public-company acquisition events, discovered directly from SEC EDGAR's own quarterly filing indexes — not scraped from a vendor list or a blog. Each row is a company that filed a merger proxy or tender-offer response between 2010 and 2024, with the earliest such filing's date as an announcement-date proxy.

Built as a byproduct of an M&A target-prediction research project. Free, reproducible, and honestly caveated — an open alternative to paywalled deal databases (SDC Platinum, Refinitiv) for event studies and takeover-prediction research.

Why EDGAR indexes are the ground truth

When a U.S. public company agrees to be acquired, it is legally required to put the deal to shareholders, which produces a dated paper trail:

Walking the SEC's quarterly master indexes (full-index/{year}/QTR{n}/master.idx) 2010–2024 and collecting every filer of these forms yields the candidate list, keyed by the earliest merger-related filing per company.

Schema

4,156 rows total: 3,596 with `passed_review=True`, 560 flagged out. Filter to passed_review=True for the cleaned deal set; keep all rows to audit the discovery-plus-review process yourself. Distribution by first form: PREM14A 1,907 · DEFM14A 1,201 · SC 14D9 879 · DEFM14C 169. Spread fairly evenly across 2010–2024 (212–338 deals/year).

The passed_review flag

Automated discovery catches everything that filed these forms, including entities that aren't operating-company acquisitions: limited partnerships, tax-credit funds, REITs winding down, SPAC combinations, and unit trusts. A manual pass flagged 560 such rows False (e.g. Cedar Fair L P, Atlas Acquisition Holdings Corp., Boston Capital Tax Credit Fund II LP, Golf Trust of America Inc). The flag is provided so you can use the cleaned set or apply your own inclusion rules.

Caveats — read before using as labels

• —The date is a proxy, not the true announcement date. It is the earliest merger-related filing date, which typically lands 2–8 weeks after the actual deal announcement (the 8-K / press release). For precise event windows, refine each row against the target's 8-K.
• —These are "announced," not "completed," deals. A proxy proves a deal was put to shareholders, not that it closed — some votes failed or deals were withdrawn. Deal outcome (completed vs. terminated) is not enriched here.
• —Acquirer and deal value are not included. Discovery reads the index, not the proxy body; acquirer identity and terms require parsing each filing.
• —Some non-M&A extraordinary-matter proxies can survive even the manual pass; treat passed_review=True as "very likely a deal," not "certified."

Reproduce it

The discovery walk is open source: github.com/tcontorno6/ma-target-screener (mascreener backfill-discover --start-year 2010 --end-year 2024). It downloads ~60 quarterly index files and aggregates one candidate per CIK by earliest merger filing.

Citation

@misc{contorno2026edgarmadeals,
  author = {Tyler Contorno},
  title  = {EDGAR M&A Deal Events (2010–2024)},
  year   = {2026},
  howpublished = {Hugging Face Datasets},
}

Data derived entirely from public U.S. SEC EDGAR filings. MIT licensed.

Part of the edgar-tok project

• —Tokenizers: edgar-bpe-65536 · edgar-bpe-65536-digits · edgar-bpe-131072-line
• —Dataset: EDGAR M&A Deal Events (2010–2024)
• —Code & benchmark: github.com/tcontorno6/edgar-tok
• —Write-up: Half the tokens: training a BPE tokenizer on SEC filings