Skip to content

Marketing disclosure data

This project builds an auditable, filing-level dataset for studying firms' disclosure of marketing, advertising, selling, and related expenses. Its purpose is to preserve what a firm reported, when it reported it, and the XBRL concept and presentation context used to report it.

The current release covers SEC annual and quarterly filings and provides two analysis datasets:

  • annual_marketing_disclosures.parquet for Forms 10-K and 10-K/A;
  • quarterly_marketing_disclosures.parquet for Forms 10-Q and 10-Q/A.

Both datasets are intentionally long and retain raw facts rather than constructing firm-year aggregates. A filing with no selected marketing-related fact is still present as a single spine row with missing fact fields. Thus, an absent concept for that filing is interpretable as non-disclosure within the selected concept registry.

The Data section describes the external source inputs, the concept registry, and the final analysis files.

Scope and design principles

The project distinguishes the date a reported amount describes (ddate) from the date the SEC received the filing (filed) and the precise filing identifier (adsh). This makes it possible to study later disclosure about an existing reporting date, including amendments and later filings.

Selected facts retain their original XBRL tag names, values, units, duration, dimensions, and segment context. No values are summed, netted, or otherwise aggregated by the pipeline. Standard tags that serve as useful accounting anchors—such as revenue, assets, liabilities, income, and SG&A—are retained alongside marketing-related candidates.

Building the documentation

The documentation source is in docs/ and is versioned with the code. It uses MkDocs with the Material theme and builds to a static site/ directory:

python -m pip install -r requirements-docs.txt
mkdocs build

Opening site/index.html is sufficient for basic offline reading. Some browser security settings can limit client-side navigation when HTML files are opened directly; if that occurs, use a simple local static server or publish the same static directory.

The generated site/ directory is deliberately excluded from Git.