Datasets:

erlenbusch
/

sec-edgar

language:
  - en
license:
  - other
size_categories:
  - 100M<n<1B
task_categories:
  - tabular-classification
  - tabular-regression
tags:
  - finance
  - sec
  - edgar
  - duckdb
  - datapond

SEC EDGAR Financial Statement Data Sets

This dataset contains structured, numeric data extracted from corporate financial statements (10-K, 10-Q, 8-K, etc.) filed with the U.S. Securities and Exchange Commission (SEC).

It has been processed into a highly compressed, instantly queryable DuckDB database as part of the Datapond open-source registry.

Dataset Details

Source: U.S. Securities and Exchange Commission
Timeframe: 15 Years (2009 to 2026)
Format: DuckDB (sec_edgar.duckdb)
Row Count: ~231.5 Million rows
Size: ~5.27 GB
License: Public Domain

How to use with Datapond

Since this database is hosted on Hugging Face, you can query it remotely over HTTP without downloading the entire 1GB file!

First, install the Datapond Python client:

pip install datapond

Then, execute SQL directly against this repository:

import datapond

# Connect to the remote DuckDB file
con = datapond.connect("sec_edgar")

# Query the database
df = con.execute("""
    SELECT s.name, n.tag, n.value, n.ddate
    FROM numbers n
    JOIN submissions s ON n.adsh = s.adsh
    WHERE n.tag = 'NetIncomeLoss'
    LIMIT 10
""").df()

print(df)

Schema

For a complete list of columns, data types, and definitions, please refer to the DICTIONARY.md file.

The database contains the following core tables:

submissions (sub.txt): Index of all filings, company names, industry codes, and filing dates.
numbers (num.txt): The actual numeric financial data points (Assets, Liabilities, Net Income, etc.).
presentations (pre.txt): How line items are presented in the statements.
tags (tag.txt): Definitions and labels for the XBRL financial tags.
_metadata: Required Datapond registry metadata.

How to Build

If you wish to rebuild the database locally from the raw SEC files:

(Optional) If you want the script to automatically generate the Datapond DICTIONARY.md, ensure the Datapond registry repository is cloned in a sibling directory:

git clone https://github.com/datapond-db/registry.git ../registry

Ensure you have uv installed, then set your SEC user agent (required by the SEC EDGAR API to prevent rate-limiting/blocks):

# On Windows PowerShell
$env:SEC_USER_AGENT="Your Name your.email@example.com"

# On Mac/Linux
export SEC_USER_AGENT="Your Name your.email@example.com"

Run the orchestration script:

uv run python main.py

(Tip: You can run a quick test build of just the last 4 quarters instead of the full 15-year dataset by passing --quarters 4)

This script will:

Download the raw ZIP files from the SEC
Ingest them using explicit schema.sql strict typing
Apply data transformations (e.g., date parsing, zero-padding CIKs)
Generate Datapond _columns and DICTIONARY.md (if the registry repo exists)
Clean up all temporary raw/ files

Citation

If you use this dataset in your research or application, please cite the U.S. Securities and Exchange Commission (SEC) as the data source:

@misc{sec_edgar_financial_statement_data_sets,
  author = {{U.S. Securities and Exchange Commission}},
  title = {Financial Statement Data Sets},
  year = {2026},
  url = {https://www.sec.gov/dera/data/financial-statement-data-sets},
  note = {Data extracted from corporate financial statements filed with the SEC.}
}