Federal Reserve Yield Curve (FEDS) Daily CSV

Federal Reserve Yield Curve (FEDS) Daily CSV is the Diversified Financial Services shelf's parameterized view of the risk-free term structure: not a scatter of individual bond quotes but a fitted curve, re-estimated for every business day since June 14, 1961. Board staff published the method as working paper FEDS 2006-28 - Gurkaynak, Sack and Wright's estimation of the Svensson six-parameter specification - and the panel has been extended continuously ever since.

Each of the 17,013 observed rows is one trading day carrying exactly 100 fields: the four Svensson betas, then four rate families at annual maturities from 1 to 30 years. Estimation draws on off-the-run coupon securities only - Treasury notes and bonds; bills, floating-rate notes, and on-the-run or first-off-the-run issues are excluded - so liquidity effects do not bend the curve and any two dates are directly comparable.

Get a sample of this dataset and Datadory returns the date ranges and maturity points you name as populated tables rather than a raw file you have to decode.

Verified rows from opposite ends of the panel, exactly as they arrive:

date   : 2026-08-14
beta_0 : 0.00183632505069158
beta_1 : 3.9585608330016
beta_2 : -497.92511719727
beta_3 : 510.080611942214
sveny01: 4.223   (1-year zero-coupon yield)
svenf01: 4.1009  (1-year instantaneous forward)

date   : 1961-06-14
beta_0 : 3.91760612208239
beta_1 : -1.2779552459795
beta_2 : -1.94939709504669
beta_3 : 0       (Nelson-Siegel era: no second curvature)
sveny01: 2.9825
svenf01: 3.5492
svenpy01: 3.0025

Read together they show the panel's two structural facts. First, the regime boundary is visible in the numbers: the 2026 row uses all six parameters while the 1961 row pins BETA3 at zero because the fuller Svensson form only took over in 1980. Second, the betas are genuine parameters, not yields - BETA2 of roughly minus 498 against BETA3 near 510 looks alarming until you notice the tau-scaled exponentials cancel it into a well-behaved hump, which is precisely why practitioners want parameters rather than interpolated points.

Nine dictionary entries cover all 100 columns of the verified spine. One identity key - Date, one business day per row. Four model parameters - BETA0 through BETA3, the level, slope and two curvature components of the Svensson form. Then four rate families whose member columns step annually from 1 to 30 years:

1. Zero-coupon yields (SVENY01 through SVENY30), continuously compounded - the pure discounting inputs.
2. Par yields (SVENPY01 through SVENPY30), coupon-equivalent - the coupon rates that price bonds back to par.
3. Instantaneous forward rates (SVENF01 through SVENF30), continuously compounded - the marginal borrowing rates the curve implies.
4. One-year forwards (SVEN1F01, SVEN1F04, SVEN1F09), coupon-equivalent - forward agreements starting 1, 4 and 9 years out.

All rates publish in percent and missing cells carry an explicit marker instead of an implied zero. Three further pieces fold under additional fields on request: the real-time vintage variant from the academic literature, per-column precision handling, and the H.15 constant-maturity companion series. Name the ones your models touch when you request a sample and they arrive as populated columns.

Geography - the United States Treasury market, specifically the nominal off-the-run curve the Board staff estimate from outstanding notes and bonds. One panel prices a 2036 cash flow and a 1991 cash flow on identical terms.

Temporal - every business day from June 14, 1961 through the present: 17,013 observed rows through August 14, 2026, unbroken across six decades of rate regimes - the 1960s peg, the 1970s inflation, Volcker's dislocations, the secular decline, the zero bound, the rapid repricing of the 2020s. No stitched-together vendor history; one estimation method documented end to end.

Granularity - one row per business day by 100 columns, with rate families at annual maturity steps from 1 to 30 years. Parameters-plus-grid means you can evaluate the curve at intermediate maturities yourself rather than accepting whatever tenor grid someone else chose.

Set against the wider Datadory catalog - where the average quality score across all 1,744 datasets is 7.81 - this record scores 9/10, carried by sixty-five years of named methodology and complete field documentation.

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

Channel and cadence are settings, not projects. Files suit a quant pulling the full 65-year panel once and calibrating against it offline; structured feeds suit dashboards that surface a discount rate or curve slope inside a product; warehouse delivery suits risk teams running scenario and shock screens in SQL against the rest of their books. Every delivery travels with the field dictionary above, the sample rows and the coverage profile mapped to the maturities and dates you named - so the schema you see in the sample is the schema you ship against.

• Investors & quant researchers - calibrate term-structure models, build curve-slope and momentum factors, and price fixed-income relative-value trades against a common risk-free backbone (investors & quants use cases).
• Journalists, academics & students - chart six decades of rate history from the same documented series the literature cites, with the working paper behind every number (journalists academics use cases).
• Data scientists & ML engineers - engineer macro and rate features from a fixed 100-column schema that never surprises: level, slope, curvature, forward differentials (data scientists use cases).
• Market researchers & consultants - benchmark financing-cost environments and rate regimes for sector briefs and valuation packs (market researchers use cases).
• Risk and finance teams - mark liability streams and run asset-liability scenarios against a discount surface that is smooth by construction rather than by interpolation.

All five work from the same surface: one row per business day, 100 typed columns, a methodology paper anyone can read.

Investors and quant researchers come first - a fitted curve with published parameters is the calibration target for an entire branch of fixed-income modeling, and having it daily since 1961 turns backtesting from an archaeology project into a join. Risk and finance teams inherit a discount surface that is smooth by construction, which matters when small curve kinks would otherwise swing liability marks. Data scientists and ML engineers get a hundred-column schema frozen in place for decades - feature pipelines built once keep running. Journalists, academics and students get the methodology paper and the estimates side by side, citable without vendor interpretation.

Within the diversified financial services shelf, this record owns the estimated term structure: curve parameters, not quotes or forms. Neighbors own different jobs. Treasury Resource Center Data Center carries the official Treasury rate tables - observed constants of maturity rather than fitted parameters. Fed MDRM Data Dictionary documents regulatory report items, not market rates. BIS Data Portal (Global Banking and Financial Statistics) aggregates cross-border banking claims at national scale.

The closest relatives sit one shelf over. US Treasury Daily Yield Curve Rates publishes observed par yields straight from the issuer - read them when you want the market's own quotes; read the FEDS panel when you want a smooth, bill-free, comparable-anywhere curve with parameters you can integrate. The Federal Reserve's H.15 selected interest rates release adds constant-maturity series down to one month and sits under additional fields on request as a companion. The FEDS CSV vs OpenSanctions comparison shows what depth-in-time buys against a present-state entity graph.

When the question is what the whole curve looked like on any business day since 1961, nothing else answers it in one shape. When the question is today's quoted coupon rates, the neighbors are faster.

Three things worth having in hand.

First, values arrive exactly as the Board staff publish them: beta parameters at full floating-point precision, rate families in percent at four decimals, missing cells marked rather than guessed. Typed versions carry the family labels decoded, so a column named for a one-year forward never masquerades as a spot yield.

Second, the 1980 boundary is part of the data. Rows before 1980 are Nelson-Siegel fits with BETA3 fixed at zero; rows after carry the full Svensson set. Backtests spanning the boundary should expect it - treating BETA3 as a free signal across the break will manufacture an artifact.

Third, verification went deepest on the standing spine. The 100-column dictionary on this page was confirmed against the file during the August 2026 pass; the real-time vintage variant known in the academic literature is not part of the standing panel, which is why it settles empirically under additional fields on request. Name the variant you need and the sample resolves the rest.