The SEC publishes every insider filing as its own XML document. We turn all of them into three clean, joinable tables — and keep adding the new ones every month.
No account needed for the sample · CSV + SQLite · 3.1 MB
Insider filings are the most direct disclosure there is of what the people running a company are doing with their own money. Every one of them is free to download. That is where the convenience ends.
The SEC publishes them one XML document at a time — nested elements, tags that differ between filing agents, one HTTP request per filing, and about 1.5 million filings since 2019. Getting from that to “which officers bought their own stock on the open market last quarter” is a parsing project before it is an analysis.
We did the parsing project. You start at the question.
One row per SEC form. Company, CIK, ticker, form type, filing date, period of report. 1,533,585 rows.
One row per insider named on a filing. Officer, director and 10%-holder flags, plus job title. 1,669,356 rows.
One row per transaction. Date, code, shares, price, dollar value, holdings after. 2,317,601 rows.
All three join on accession_number, the SEC's unique filing identifier. Ships as gzipped CSV with a build script that produces an indexed SQLite database in about two minutes, plus full schema documentation and ten working starter queries.
Neither of these is in the SEC's documentation. You find them by getting a strange result and working backwards. They are the first two pages of ours.
Of 2,317,601 transactions, just 218,895 carry code P — an open-market purchase, someone choosing to buy shares with their own money. Codes A (grant), F (tax withholding) and M (option exercise) outnumber P by about five to one and run on vesting schedules set years earlier. Aggregate without filtering and you are mostly measuring the payroll calendar.
Affiliated funds and general partners co-file the same transaction under separate owner rows. A naive SUM(value) counts one purchase several times over — and the inflation is worst exactly where private-equity and venture holders are involved, so it is not uniform noise you can ignore.
An interactive view of the same data: pick an insider, see every filing they have made plotted against the share price on the days they filed. Buys and sells marked on the chart, with the full transaction list underneath.
Every plan carries the identical schema and the same documentation. The only difference is how much history you get and whether it keeps arriving.
Commercial use permitted on every plan. You may not resell the compiled files as a competing dataset.
The SEC's EDGAR full-text filing archive — the same public source the filings are submitted to. Nothing is scraped from third-party sites, and no figure is estimated or modelled. Every row traces back to a specific accession number you can look up yourself.
Yes. Filings submitted to the SEC are public records, and works of the US Government carry no copyright under 17 U.S.C. §105. Use it for research, internal tooling, client work or a commercial product. The one restriction is that you may not repackage and resell the compiled files themselves as a competing dataset.
The archive currently runs through 30 June 2026. Subscribers get a refreshed build each month containing every filing accepted by EDGAR since the previous one. Filings appear in a refresh once the SEC has published them — insiders have two business days to report most transactions, so there is an inherent lag in the source that no data provider can remove.
Gzipped CSV, plus a Python script that builds an indexed SQLite database from them in roughly two minutes. No dependencies beyond Python itself. If you would rather stay in pandas, read_csv handles the gzip directly.
No, and neither is anyone else's, because the errors are upstream. Filers occasionally mispunch share counts and prices, and those mistakes go into EDGAR unchallenged. We parse faithfully rather than silently "correcting" figures, so an outlier you find is a real outlier in the public record. The schema documentation flags where this shows up most.
Not yet. Today it is bulk files, which is what most analysis actually wants — you can run a seven-year query locally in seconds rather than paginating through HTTP requests. If you need an API for a production application, get in touch and tell us what you're building.
No, and we won't pretend otherwise. This is raw source data and tooling to explore it — a publication, not an advisory service. What you conclude from it is entirely your own work.