Treasury Daily Par Yield Curve Csv Data: Dataset Structure and Field Coverage

One publisher, five products, one headline file. U.S. Treasury Interest Rate Data (Daily Yield Curve) bundles Daily Treasury Par Yield Curve Rates, Daily Treasury Par Real Yield Curve Rates, Daily Treasury Bill Rates, Daily Treasury Long-Term Rates and Extrapolation Factors, and Daily Treasury Real Long-Term Rate Averages into a single catalog record. The par curve you load for pricing is built from closing bid quotations on the most recently auctioned Treasury securities, collected by the Federal Reserve Bank of New York at roughly 3:30 PM each business day.

That construction detail matters more than it looks. Because the prints are quotations rather than transactions, Treasury labels them indicative rates, not trade prices — a distinction that shows up in every compliance review of a valuation model's inputs.

Inside Datadory's investment banking brokerage slice this record is the anchor: twelve primary datasets cover the industry, ten are free, six refresh daily, and exactly one carries a perfect quality score — this one, at 10 out of 10 on a rubric where the all-catalog average across 1,744 datasets is 7.81.

Every row opens with a Date in MM/DD/YYYY form followed by one column per tenor. Current-year files carry fourteen constant-maturity points — 1 Mo, 1.5 Month, 2 Mo, 3 Mo, 4 Mo, 6 Mo, 1 Yr, 2 Yr, 3 Yr, 5 Yr, 7 Yr, 10 Yr, 20 Yr and 30 Yr — but the set grew over time: 1990 files hold only nine columns from 3 Mo to 30 Yr, while 2026 files add the 1.5 Month, 2 Mo and 4 Mo points. A calendar year holds roughly 250 business-day rows, and a 2024 par-curve file weighs about 19 KB, so the full 1990-2026 archive is a few megabytes.

Sample rows show how the shape moved. On 12/31/1990 the 3 Mo yielded 6.63 against a 30 Yr of 8.26. On 12/31/2024 the 10 Yr printed 4.58 and the 30 Yr 4.78. By 08/20/2026 they read 4.69 and 5.23 with the 20 Yr at 5.20. Field definitions in Datadory's record are marked verified against pages consulted August 21, 2026.

Par-curve CSV runs by year from 1990 to present, so a regime study can span inverted 1990s curves, the mid-2000s expansion and the post-2022 tightening without stitching publishers. Two boundaries matter when you design the schema. First, the tenor set is not stable: nine columns in 1990 versus fourteen today means any panel joining decades must union column names and tolerate blanks where a tenor did not exist. Second, the 30-year point disappears for 2002-2006 — Treasury issued no 30-year bonds then — which is precisely why the Long-Term Rates and Extrapolation Factors product publishes a 20-year-plus-adjustment series for that window.

For investors and quant researchers this file is the first pull, not a nice-to-have: the persona page ranks it relevance 3 of 3, and the workflow below is why. A tenor-by-tenor dated curve lets a model interpolate the exact maturity a cash-flow schedule needs instead of borrowing whatever single benchmark rate a vendor happens to quote:

1. Pull the current year's CSV plus as many prior years as your backtest needs; the whole 1990-2026 archive loads in a few megabytes.
2. Parse Date as MM/DD/YYYY and coerce every tenor column to numeric, keeping blanks where a maturity was not yet published.
3. Interpolate along the row — linearly between adjacent tenors, or with a bootstrap if you need zero rates rather than par yields — to hit your cash-flow dates.
4. Join the funding leg separately: SOFR is an overnight print, so a 10-year discount rate exists only in this Treasury product.
5. Store the full panel, not just the 10 Yr; curve regime features such as the 2s10s slope come free once every column is retained.

They answer different questions, and desks that price funding-sensitive books ingest both. The nearest neighbor in the slice, Federal Reserve Bank of New York Markets Datasets (quality score 9, also free and key-free), publishes EFFR, OBFR, SOFR, TGCR and BGCR — overnight and near-overnight reference rates. Its August 20, 2026 SOFR print read 3.63% on $2,922 billion of volume with percentile bounds of 3.58 to 3.71, alongside repo and reverse repo operation results, weekly SOMA holdings back to 2003 and hundreds of primary dealer timeseries.

Three failure modes recur when teams wire this feed into pricing infrastructure. The XML route fails silently: without Accept: application/xml, pages/xml returns HTTP 406, which naive clients log as an empty response rather than a header problem. The static mirrors mislead: yield.xml and siblings look like an archive but hold roughly the last two weeks, so teams that build on them lose history the first time they re-run a job. And the URL base moved: legacy www.treasury.gov addresses now 302 onto home.treasury.gov's interest-rate-statistics page.

Schema drift is the quieter one — the tenor set grew from nine columns in 1990 to fourteen by 2026, and the 1.5 Month, 2 Mo and 4 Mo points exist only in recent years. Pin your loader to named columns, not positions, and expect blanks in the early years. Datadory's verification notes flag all four issues explicitly, which is part of why this record scores 10 out of 10 while most feeds leave discovery to trial and error.