The dataset · run dated 2026-08-14

Historical SEC 8-K Market-Reaction Dataset

The market's reaction to every kind of SEC 8-K news, measured properly. Every figure on this site is computed from it, so you can check the product against its own public output before deciding whether you want the file.

Start here — download the sample

Free. No email address, no account. The sample is small on purpose. It has the complete column schema and almost none of the rows. You can see exactly what the file looks like and whether it fits what you are building, but it will not stand in for the product.

Why pay when EDGAR is free

The filings are free and always will be. What is not free is the work between a filing and a number you can rely on: normalising every disclosure category, aligning each filing to the right trading day in the right timezone, fitting a market model per company to separate the reaction from the market's own movement, correcting for the fact that hundreds of cells were tested at once, and then checking every surviving result twice out-of-sample — on later filings, and on companies never used to find it.

You could rebuild that. It is roughly the work this project has spent its existence on. The failures are documented here too, because the results that did not replicate are part of what makes the ones that did believable. What you are buying is that layer, dated and versioned. Not the filings.

Two ways to get it

A forward-looking version — the same classification applied to filings as they arrive — does not exist yet and has no date. If that is the one you would actually use, say so here. What people ask for is how the order of work gets decided.

What the extra resolution actually buys you

The public pages show the official filing categories. The file goes below them. Here is a real row, so the difference is visible rather than asserted. It gives the coarse category, then what that category splits into. Each split is shown with the figure it was discovered at, and the same figure recomputed on independent holdouts it was never fitted to.

Discovery is where a pattern was found; the two holdout columns are later filings, and a separate set of companies. A finding only enters the file if it survived both. Findings that did not survive are in there too — the failures are part of what makes the survivors believable.

What is in the one-time file

• Every validated finding with its discovery figure and BOTH out-of-sample holdout columns, refutations included.
• Classification below the raw filing categories — the resolution that the public pages do not show.
• The full aggregate grid across size, filing time, disclosure lag and reaction window, including cells too sparse to publish here.
• A field-by-field data dictionary and the methodology document.
• A dated, versioned release, so a figure you cite stays checkable.

What stays free

The aggregate results, the full method, and every answered question stay public. The argument for the dataset is the dataset. A research artifact that hides its own numbers is asking to be taken on faith.

Public pages carry aggregate cells only — no issuer names, tickers, or individual filing records, by policy rather than by oversight.

Which licence applies to you

One licence, one price — whoever you are and whatever the use. There is no separate academic tier and no commercial upcharge; the table exists to say so explicitly.

One licence at one price: $99 one-time, covering commercial use as well as academic and personal research. The full licence text ships inside the download as LICENSE.md; ask if you want to read it before buying and it will be sent to you in writing.

Before you buy

Read the methodology first, especially the limitations. If the survivorship caveat or the multiple-testing correction is a problem for your use, this dataset will not solve it and you should not buy it.

Who made it

Mustafa. I designed the requirements and the honesty rules; AI implemented them; everything is reproducible from the methodology page.