US Insider Trading Filings Database — 1.5M SEC Forms 3, 4 & 5 (2019–2026)

Try it before you buy — the free one-quarter sample is here: https://payhip.com/b/7n6AG

Every SEC insider-trading filing from January 2019 through June 2026 — pulled from EDGAR's raw XML and normalised into three clean, joinable tables. Skip the pipeline. Start with the question.

WHAT YOU GET

1,533,585 filings — Forms 3, 4, 5 and their amendments

2,317,601 individual transaction lines

132,229 distinct insiders, with officer / director / 10%-owner flags

218,895 open-market purchases (transaction code P)

10,583 companies · 11,974 tickers

Continuous coverage: 2019-01-02 through 2026-06-30

THE DATA IS FREE. MAKING IT USABLE IS THE WORK.

EDGAR publishes every insider filing — as individual XML documents, one at a time, rate-limited to ten requests a second. Getting from there to something you can query means writing parsers for every form variant, reconciling amendments against originals, linking owners to issuers across name changes, and normalising seven years of inconsistent formatting from thousands of different filers. It takes weeks, and most of those weeks go on edge cases you cannot see until you hit them.

This is that work, finished.

THREE TABLES, ONE JOIN KEY

Everything joins on accession_number, the SEC's unique identifier for a filing.

filings — form type, filing date, issuer, ticker, CIK

owners — who filed, their role, their job title, their CIK

trans — date, transaction code, shares, price, dollar value, holdings after

IN THE DOWNLOAD

csv/ — all three tables, gzipped. Works with pandas, R, DuckDB, Excel, anything.

build_db.py — builds an indexed SQLite database in about two minutes. Python only, no dependencies.

SCHEMA.md — every column documented, plus the two mistakes that produce wrong answers.

QUERIES.sql — ten working queries: cluster buying, CEO purchases, new insiders who then bought, late filings, monthly buy/sell ratios.

sample/ — one full quarter, uncompressed, so you can look before you rely on it.

WHAT MAKES IT USABLE RATHER THAN JUST LARGE

The documentation tells you the things that take weeks to discover on your own. That transaction codes A, F and M outnumber genuine purchases roughly five to one and are almost entirely automatic — vesting schedules, tax withholding, option mechanics. That affiliated funds co-file the same transaction, so a naive SUM of value multiplies single purchases severalfold. That a handful of source filings contain mispunched values in the billions. Every starter query already accounts for these.

WHO THIS IS FOR

Quant researchers testing insider-activity signals. Developers building screeners and dashboards. Finance students and academics who need a real dataset rather than a toy one. Analysts who want to answer a question this week instead of building a pipeline for a month.

LICENSING

SEC filings are public records, and US Government works carry no copyright under 17 U.S.C. §105. Use this commercially, in research, in internal tooling, in a product you sell. The only restriction is that you may not resell the compiled files as a competing dataset product.

Provided as-is, without warranty as to accuracy or completeness. Source filings occasionally contain errors made by the original filer. Nothing here is investment advice. Not affiliated with or endorsed by the US Securities and Exchange Commission.

UPDATES

Coverage runs through 30 June 2026, refreshed quarterly.