U.S. Treasury Interest Rate Data (Daily Yield Curve)

One number for every length of time the US government borrows money, repriced each business day - and the reference line the rest of dollar-denominated finance hangs off. This record holds all five products in Treasury's Interest Rate Statistics program, not just the headline curve: Daily Par Yield Curve Rates, Daily Par Real Yield Curve Rates (the TIPS-based version, published since January 2, 2004), Daily Treasury Bill Rates, Daily Long-Term Rates and Extrapolation Factors, and Daily Real Long-Term Rate Averages back to 2000.

The par curve is built from closing bid quotations on the most recently auctioned Treasury securities, collected by the Federal Reserve Bank of New York, and spans fourteen constant maturities per business day from 1 Mo out to 30 Yr. Files come down by year back to 1990. Datadory scores it 10 out of 10 on its quality rubric - above the 7.81 average across the 1,744-dataset catalog - with field definitions marked verified against live captures. Get a sample of this dataset: name your date range and tenors and the extract comes back shaped around them.

One captured day beats a paragraph of specification. These are real rows read off the files during the August 2026 verification pass:

# one row per business day x 14 par tenors; values are percent yields

date : 08/20/2026
1 Mo 3.80   1.5 Month 3.77   2 Mo 3.79   3 Mo 3.87   4 Mo 3.88
6 Mo 3.94   1 Yr 3.99       2 Yr 4.19   3 Yr 4.26   5 Yr 4.39
7 Yr 4.53   10 Yr 4.69      20 Yr 5.20  30 Yr 5.23

date : 12/31/1990            # the file's oldest verified vintage
3 Mo 6.63   6 Mo 6.73   1 Yr 6.82   2 Yr 7.15   3 Yr 7.40
5 Yr 7.68   7 Yr 8.00   10 Yr 8.08              30 Yr 8.26

date : 08/20/2026  real (TIPS-based)  # five maturities only
5 YR 2.05   7 YR 2.18   10 YR 2.35   20 YR 2.73   30 YR 2.95

# machine-feed variant of the same day:
BID_CURVE_DATE 03-AUG-26  DAY_OF_WEEK MONDAY
BC_1MONTH 3.79   BC_10YEAR 4.70   BC_30YEAR 5.23

Three things ride in those lines. First, completeness: the 08/20/2026 row carries all fourteen tenors, so a 4.53 seven-year and a 5.20 twenty-year exist as quotes rather than interpolations. Second, depth - the 1990 row sits in the same format thirty-six years earlier, its nine columns marking how far the panel had grown by then (3.87 borrows cheaply where 6.63 once did; 5.23 costs far less at thirty years than 8.26). Third, the real curve prints separately at five maturities - 2.35 real against 4.69 nominal at ten years on the same August day, which is the market's breakeven inflation rate stated as subtraction.

Seven exampled fields form the core grid, every definition verified against live captures during the August 2026 cataloging pass: the observation Date plus percent par yields at six representative maturities - 1 Mo, 1.5 Month, 3 Mo, 2 Yr, 10 Yr and 30 Yr - chosen as the corners desks touch most. Date is the whole identification layer, so joining the curve against positions, issuance calendars or macro releases resolves on one column.

What folds under additional fields on request: the remaining eight par tenors, the five-maturity real curve, bill rates, both long-term products including the 2002-2006 adjustment bridge, and the typed machine-feed variants (NEW_DATE timestamps, BC_<TENOR> bid-curve fields, and the rolling recent-observations snapshot keyed on BID_CURVE_DATE and DAY_OF_WEEK). Field-definition confidence across the catalog is 85.7% - 1,495 of 1,744 datasets verified; this one sits inside that set.

Geography - the United States, deliberately. This is the sovereign curve: the closest thing markets have to a risk-free benchmark, which is precisely why it prices everything from a corporate bond to a mortgage pool regardless of where the underlying assets sit.

Temporal - the deepest rate shelf in the catalog. Par-curve files run by year from 1990 to present; the real curve starts January 2, 2004; real long-term averages reach back to 2000; and the long-term product carries its extrapolation bridge across 2002-2006 so nothing drops out of a long study. Freshness matches the cadence - one row lands per business day, putting this among the 395 of 1,744 cataloged datasets (22.6%) that update daily.

Granularity - one row per business day crossed with fourteen par tenors, roughly 250 rows per calendar year per product. A 2024 par-curve cut runs about 19 KB, so the entire 1990-2026 daily history weighs a few megabytes - complete enough to interpolate anything between quoted points, small enough never to think about storage.

API, files, or your warehouse. Daily, weekly, or hourly.

You pick the channel and the cadence; the dictionary above travels unchanged through all three. Rows arrive flattened - one date, one column per maturity, percent values ready to divide into discount factors - so joining the curve against holdings, trade blotters or macro panels is a join statement rather than a parsing project. Cadence changes are a settings conversation, not a re-integration, and a sample cut to your named dates comes first either way.

• Debt capital markets and valuation desks price bonds, swaps and DCF models off the official curve - fourteen points a day make interpolation a computation instead of an assumption.
• Quant researchers compute level, slope, curvature and inversion flags nightly as regime features; one of the cataloged datasets serving quant backtesting.
• Relative-value traders subtract the par curve from credit or municipal yields so the spread, not the rate level, becomes the tradable object.
• Inflation analysts pair nominal with TIPS-based real yields for direct breakeven series - 234 basis points at ten years on 08/20/2026.
• Researchers and writers anchor charts to the government's own publication; see citation grade research and ML model training for the adjacent workflows.

Ranked by Datadory's persona tagging - five of eight personas score this record at maximum relevance, which almost never happens:

1. Investors & Quant Researchers (relevance 3) - the discount rate, dated daily since 1990, behind valuation and backtests; see investors quants use cases.
2. Data Scientists & ML Engineers (relevance 3) - clean exogenous features at daily grain; see data scientists use cases.
3. Market Researchers & Consultants (relevance 3) - curve charts nobody can argue with; see market researchers use cases.
4. Developers & Data-Product Builders (relevance 3) - a tiny stable schema any rate-aware app consumes; see developers builders use cases.
5. Journalists, Academics & Students (relevance 3) - citable by construction; see journalists academics use cases.
6. Competitive Intelligence & Product Teams (relevance 1) - the financing-cost backdrop framing rate-sensitive sectors; see competitive intel product teams use cases.

This is the rates spine of the industry's twelve cataloged datasets - everything else measures markets or participants around it. Federal Reserve Bank of New York Markets Datasets (scored 9) covers what the Fed does - SOFR and EFFR reference rates, repo operations, primary dealer statistics - while this record covers what the government pays to borrow; the head-to-head runs in vs Federal Reserve Bank of New York Markets Datasets. FINRA API Developer Center adds corporate bond trades via TRACE, MSRB EMMA covers municipal trades and disclosures, and SIFMA Capital Markets Fact Book sizes the industry annually. Rank the whole slice on best investment banking & brokerage datasets or browse the investment banking & brokerage data hub.

Three things, stated plainly.

First, the panel widens over time: 1990 files carry only 3 Mo through 30 Yr across nine columns, while recent files add the 1.5 Month, 2 Mo and 4 Mo points to reach fourteen. Model the widening explicitly - treat early vintages as narrower panels rather than expecting today's width backward.

Second, these are indicative quotations, not trade prints: par yields derived from closing bids on recently auctioned securities describe where the market closed, not executed volume. Pair them with a trade tape such as FINRA TRACE when transaction detail matters.

Third, the score reflects the ceiling: Datadory rates this record 10/10 against a cross-catalog mean of 7.81, one of just 145 datasets scoring perfect marks. Samples precede any commitment - name your start date, your tenors and your cadence, and the extract arrives cut to exactly that.