Five families, fitted from real cashflow schedules on every trading day since September 2, 2008. Not interpolated from published constant maturities, and not quotes. Every fit publishes its own error.
Looking for today's numbers? Every family for the latest close, with the session's move at each maturity, is on one page. Treasury rates, September 9, 2026. Any earlier trading day is at /treasury/rates/YYYY-MM-DD.
Every family, one trading day
September 9, 2026 · last fitted close
One trading day, every family at once. The money market curve ends where the zero curve begins, the real curve runs below the nominal one by roughly what the market charges for inflation, and par sits close enough to zero to show that those two are one curve expressed two ways.
Money marketZeroParLSCForwardReal (TIPS)
All five families are drawn, “LSC” being the level-slope-curvature model described below. Par and the forward are not families of their own: both are implied by the same fit as the zero curve, which is why they move with it. The forward stops at 20 years because it is the derivative of that fit, and the error in a derivative grows fastest at the long end. The forward curve has its own page.
Which curve to use, and when not to
The families are not five renderings of one curve. Two of them cover maturities the others do not, one is in inflation-adjusted terms, and two are the same nominal curve expressed differently. Picking the wrong one is the most common way to get a right number that answers the wrong question.
Zero-coupon discount rates, with the par yield and instantaneous forward implied by the same fit.
Use it for:
Almost always. This is the family to discount a cashflow with. Measured like for like against the Federal Reserve's own SVENY zero curve across 4,498 overlapping days and ten tenors, it agrees to 3.06 basis points RMSE.
Where it does not apply:
Ten fitted tenors from one to thirty years. It is NOT fitted below one year, so read the front end off the money market curve instead. Its implied forward is not usable beyond twenty years: fit error there reaches 61 basis points against 7 at twenty, because a forward is the derivative of the fit and whatever is loose shows up there first.
The sub-one-year segment, fitted from Treasury bills alone on a bond-equivalent basis, from one week to one year.
Use it for:
Anything short-dated. The long-curve families are constrained by coupon securities and fit the very front end poorly, which is why the zero curve declines to publish it at all.
Where it does not apply:
Bills are quoted on a discount basis, so every rate here is a conversion to a bond-equivalent yield rather than the number Treasury quotes. Do not compare it to a discount rate without converting.
Yields above inflation, fitted from inflation-linked securities. Seven maturities from two to thirty years.
Use it for:
Anything in inflation-adjusted terms. The gap between this and the nominal curve at the same maturity is inflation compensation, which is a market price and NOT a forecast.
Where it does not apply:
There is no real curve inside two years: there are rarely enough short-dated linkers to fit one. It is fitted from 24 to 46 securities against the nominal curve's 300-odd, so its errors run wider by construction and it is judged against its own distribution rather than the nominal curve's.
Par yields from a Nelson-Siegel-Svensson fit: the coupon rate at which a security of that maturity would price at par.
Use it for:
Comparing against a published par curve, which is what Treasury and the Federal Reserve publish. Thirteen tenors from three months to thirty years, the widest span of any family here.
Where it does not apply:
It carries a coupon effect, which makes it the wrong choice for discounting a cashflow. Use the zero curve for that.
A three-factor model reducing each day's curve to a level, a slope and a curvature, fitted with a fixed decay parameter.
Use it for:
Regressions and time-series modelling. Three factors with stable meaning make a series you can model; ten correlated tenors do not.
Where it does not apply:
It fits less tightly than the Svensson forms, 9.6 basis points on average against their 3.8, because three factors cannot bend as many ways as six. That is the trade being made rather than a defect.
Each day's curve is fitted to the actual cashflow schedules of the securities outstanding that day, not interpolated between published constant maturities. Daily error averages 3.8 basis points across the history and reaches about 20 on the worst days, in December 2008, when the market was genuinely hard to fit one smooth curve to.
Every dated page publishes its own fit error and whether that day's fit converged, rather than burying it. Sixteen days in eighteen years did not converge, and each of those says so on its own page.
Index levels before September 30, 2026 are back-tested. They were computed after the fact by applying the rules to historical data, which benefits from hindsight in the choice of rules, and an index cannot be invested in directly. Methodology v1.0 takes effect at that rebalance, when levels begin to be struck on the day; the rulebook is identical either way, and the version is published on every row.
A fitted curve is a fit, not a quote. Daily error averages 3.8 basis points across the history and reaches about 20 on the worst days, in December 2008, when the market was genuinely hard to fit one smooth curve to. Every curve page publishes its own fit error rather than burying it.
Free to benchmark against. Paid only to track. Measuring anything against these curves and indices is free. No license, no fee, no registration, and no permission needed to say that you did, including in a prospectus. A fee applies to one thing: launching a product that tracks an index.
Safe Rate™, Safe Rate Indices™ and the Safe Rate US Treasury Index™ are trademarks of Safe Rate, claimed through use in commerce and not registered. Third-party marks are the property of their owners, which are not affiliated with and do not endorse this data.
Curves are fitted from public Treasury data and carry fit error; figures are not a record of trading, and an index cannot be invested in directly. No claim of compliance with the IOSCO Principles for Financial Benchmarks is made or implied. Not investment advice, not an offer, and not a recommendation to buy or sell any security.