brishen/fomc-meeting-transcripts

FOMC Meeting Transcripts (1976–2020)

Full-text transcripts of 373 Federal Open Market Committee (FOMC) meetings, from March 1976 through December 2020, converted from the official PDF transcripts published by the Federal Reserve Board.

The FOMC is the body of the U.S. Federal Reserve System that sets monetary policy (the federal funds rate target, balance-sheet policy, etc.). Verbatim meeting transcripts are released to the public with a roughly five-year lag, which is why coverage currently ends in 2020. The corpus spans the Burns, Volcker, Greenspan, Bernanke, Yellen, and Powell chairmanships — including the Volcker disinflation, the 1987 crash, the 2008 financial crisis, and the COVID-19 emergency meetings of 2020.

• 373 meetings · ~15.8 million words of deliberation
• One row per meeting, with the complete transcript text in markdown
• Each row links back to the official source PDF on federalreserve.gov

Schema

Field	Type	Description
meeting_id	string	Fed identifier, e.g. FOMC20201216 (date is the meeting's final day, per Fed file naming)
date	string	ISO date YYYY-MM-DD of the meeting's final day
year	int32	Meeting year, convenient for filtering
title	string	Document title, e.g. "FOMC Meeting Transcript — December 15–16, 2020"
text	string	Full transcript in markdown. Speaker turns are marked **SPEAKER TAG.** at the start of a paragraph, one paragraph per turn
num_words	int64	Whitespace-tokenized word count of text
num_chars	int64	Character count of text
source_pdf_url	string	Official source PDF on federalreserve.gov

Example usage

from datasets import load_dataset

ds = load_dataset("brishen/fomc-meeting-transcripts", split="train")

# The Lehman-week meeting, September 16, 2008
lehman = ds.filter(lambda r: r["meeting_id"] == "FOMC20080916")[0]
print(lehman["text"][:2000])

How it was built

1. Transcript PDFs were downloaded from the Federal Reserve's historical materials pages.
2. PDFs were converted to markdown (text extraction, speaker turns preserved as **NAME.**-prefixed paragraphs).
3. Converted files were packed into a single parquet file, one row per meeting, sorted by date.

Known limitations

• PDF-extraction artifacts. Text was machine-extracted from PDFs; occasional spurious spaces inside words (e.g. "Meet ing", "PR ESENT") and page-break hyphenation artifacts remain, especially in headers and attendance lists. The deliberation body text is largely clean.
• Conference calls are not included — only regularly scheduled (and some emergency) meetings whose transcript files follow the Fed's FOMC<date>meeting.pdf naming.
• Five-year release lag. Meetings after 2020 are not yet public.
• Attendance preambles, footnotes, and end-of-meeting votes are embedded in text rather than parsed into structured fields.

Provenance and license

The underlying transcripts are works of the United States federal government published by the Board of Governors of the Federal Reserve System, and as such are in the public domain in the United States. This dataset is a format conversion of those public documents; no restrictions are added. Please cite the Federal Reserve as the original source and link the relevant source_pdf_url when quoting.