AhaSignals/us-treasury-yield-history

U.S. Treasury Yield History

Version 0.1.0 - official nominal par yields, reconstructed from a later source capture. Coverage: 2020-01-01 through 2026-10-16. Maturities: 2, 5, 10 and 30 years.

The calendar contains 2,481 days and the long table has 9,924 date-tenor rows, including 6,796 observed yields. Weekend and absent weekday rows remain null. No value is forward filled. These are Treasury nominal par-curve observations, not auction yields, real yields, intraday bond quotes, market probabilities or contract settlements.

Files and fields

observations.csv / observations.jsonl: date, maturity in years, exact source yield as a decimal percentage string, missingness status, source field, URL, captured-body SHA-256 and actual retrieval time. A value of 4.50 means 4.50 percent, not 0.045 percent. calendar distinguishes observed rows, absent weekends and absent weekdays without inventing a holiday/outage cause. sources preserves seven yearly response versions. summary.json gives counts and coverage; schemas/ defines each table.

Versions and timing

All initial source responses were captured on 2026-10-18. This release does not reconstruct first-published values or historical intraday availability. Treasury may revise historical values. A source hash identifies retrieved bytes, not first-publication time or an independently certified vintage. Future observations are append-only; differences become explicit old/new values. Raw feed bytes are retained privately for source checking. No historical value may be overwritten under an existing release path.

Reproduce offline

Download and unpack research.zip:

python3 verify.py .
python3 example.py treasury . --start 2024-01-01 --end 2024-12-31 --tenor 10 --threshold 4.5

The example asks whether any observed official daily value met or exceeded the threshold. Missing days are reported and excluded; a window containing no observations returns null. A query outside the retained calendar fails. It does not decide a trading platform's settlement. Python 3.9+; no paid service or credentials. Pin the manifest hash and preserve the selected source-version IDs.

Rights and sources

The Treasury Data.gov catalog identifies these official observations as CC0-1.0: https://catalog.data.gov/dataset/interest-rate-statistics-daily-treasury-yield-curve-rates . Feed documentation: https://home.treasury.gov/treasury-daily-interest-rate-xml-feed . Original AhaSignals documentation and research annotations are CC BY 4.0; code is MIT. Our license does not restrict the source's CC0 observations. No third-party FRED series, commercial dataset or prediction-platform data are included.

This compact companion adds explicit missingness, source versions and reproducible daily-threshold queries. It does not claim a novel yield dataset, a complete historical-vintage archive or a forecasting advantage. AhaSignals is independent of and not endorsed by the U.S. Treasury or other named agencies. Research only; not investment or trading advice.

The Hub reader uses train as a table split label. It is not an independently held-out benchmark partition. Original Treasury observations remain CC0; see the license boundary above.