What a Yield Curve Is Trying to Tell You

A yield curve plots the interest rates on US Treasury securities from the shortest maturity to the longest on a single day, and its slope summarizes what the bond market expects for interest rates, inflation and growth. When long-term yields sit below short-term yields, the curve is "inverted," and since 1969 every US recession has followed an inversion of the 10-year minus 3-month spread by 5 to 16 months. The 2022–2024 inversion was the longest in monthly records going back to 1959, yet as of October 2026 no recession had been dated after it.

Key takeaways

The short end follows the Federal Reserve's target range for the federal funds rate; the long end reflects the expected path of short rates plus a term premium that models estimate rather than observe.
Monthly averages of the 10-year minus 3-month spread turned negative 5 to 16 months before each of the eight NBER business cycle peaks from December 1969 to February 2020, with a median lead of 10.5 months.
In 2022–2024 the monthly spread stayed negative for 25 consecutive months (November 2022 to November 2024), dipped slightly below zero again in March and April 2025, and the NBER had announced no new peak as of October 13, 2026.
A compressed term premium and expected disinflation are the main explanations offered for the miss; neither fits the whole episode, and Fed research disagrees on how much they matter.

What the Treasury curve plots

The US Treasury publishes the Daily Treasury Par Yield Curve Rates for maturities from 1 month to 30 years. A par yield is the coupon rate at which a security of that maturity would trade at face value, which makes the points comparable. Per Treasury's methodology, the curve is fitted daily from quotes on the most recently auctioned securities, collected near 3:30 p.m.

The two most quoted spreads are 10-year minus 3-month and 10-year minus 2-year; the St. Louis Fed's FRED database publishes them daily as T10Y3M and T10Y2Y, using Treasury's own data since June 2019. The yields and spreads in this article come from Treasury's daily file.

Four shapes and what each means

Normal (upward sloping): longer maturities pay more, the usual state. It reflects expectations that short rates will hold or rise, a positive term premium, or both.
Flat: yields are similar across maturities. At the end of 2019 the segment from 3 months to 2 years was nearly flat, at 1.55% and 1.58%.
Inverted: short yields exceed long yields, which usually means markets expect short rates to be lower in the future than they are now.
Humped: middle maturities yield more than both ends. Curves in transition also take mixed shapes: in May 2023 the curve fell from 3 months to 5 years and then rose again through 20 years.

Date	3-month	2-year	5-year	10-year	30-year	10y − 3m	10y − 2y	Fed target range
Dec 31, 2019	1.55%	1.58%	1.69%	1.92%	2.39%	+0.37	+0.34	1.50–1.75%
Dec 31, 2021	0.06%	0.73%	1.26%	1.52%	1.90%	+1.46	+0.79	0–0.25%
May 4, 2023	5.26%	3.75%	3.29%	3.37%	3.73%	−1.89	−0.38	5.00–5.25%
Oct 19, 2023	5.60%	5.14%	4.95%	4.98%	5.11%	−0.62	−0.16	5.25–5.50%
Oct 9, 2026	4.25%	4.80%	5.02%	5.24%	5.60%	+0.99	+0.44	3.75–4.00%

Spreads in percentage points. May 4, 2023 was the cycle's deepest daily 10-year minus 3-month reading; October 19, 2023 was that year's 10-year high. Target ranges are from the Federal Reserve; the 3.75–4.00% range followed a 0.25-point increase effective September 17, 2026. The latest row is a snapshot, not a forecast.

Short end, long end, term premium

The Federal Open Market Committee sets the target range for the federal funds rate; its rate history page lists each change. Because investors can choose between a 3-month bill and overnight lending at rates the Fed steers, bill yields usually stay close to that range, though they can sit above or below it: 0.06% under a 0–0.25% range and 5.60% under 5.25–5.50%. On October 9, 2026, the 3-month yield was 4.25%, above the 3.75–4.00% range.

A 10-year yield is closer to an average: Fed Board staff describe it as the expected path of short-term rates over the bond's life plus a term premium, the extra yield investors demand, or give up, for holding a long bond instead of rolling over short ones. The premium can be negative.

10-year yield ≈ expected average short rate over 10 years + 10-year term premium

Neither part is visible in market prices. The New York Fed publishes the Adrian-Crump-Moench (ACM) estimates on its Treasury term premia page, based on the method in Staff Report 340; the Fed Board publishes the Kim-Wright estimates. In ACM figures, a 3.40% 10-year zero-coupon yield on May 4, 2023 split into a 3.99% expected path and a −0.60 point premium; on October 9, 2026, 5.23% split into 4.33% and +0.90.

Date	ACM 10-year term premium	Kim-Wright 10-year term premium
Dec 31, 2019	−0.79	−0.19
Dec 31, 2021	−0.68	−0.06
May 4, 2023	−0.60	−0.16
Oct 19, 2023	+0.45	+0.66
Oct 2, 2026	+0.91	+1.08

Percentage points, from the two models' published daily files. The levels differ by up to about 0.6 points, but the direction agrees.

The recession record since 1969

The research base starts with Arturo Estrella and Frederic Mishkin. Their 1996 New York Fed article, The Yield Curve as a Predictor of U.S. Recessions, found that the 10-year minus 3-month spread outperformed stock prices and indexes of leading indicators at predicting recessions two to six quarters ahead. Their 1998 journal paper concluded that beyond two quarters the slope of the yield curve "emerges as the clear choice."

The New York Fed turned that work into a recession probability model: it converts the monthly average 10-year minus 3-month spread into the probability of a recession twelve months later. In its data file, updated in September 2026, the August 2026 spread of 0.87 points gives 13.9% for August 2027; the cycle's peak reading was 70.9%, for May 2024.

The table below uses that file's monthly spreads and the NBER's business cycle chronology. "First inversion" is the first month with a negative monthly average spread in the 24 months before the peak, counting only months after the previous recession ended.

NBER peak	First inversion month	Lead (months)	Lowest monthly spread in the 24 months before the peak
Dec 1969	Dec 1968	12	−0.51
Nov 1973	Jun 1973	5	−1.59
Jan 1980	Nov 1978	14	−2.20
Jul 1981	Oct 1980	9	−3.51
Jul 1990	Jun 1989	13	−0.16
Mar 2001	Jul 2000	8	−0.70
Dec 2007	Aug 2006	16	−0.51
Feb 2020	Jun 2019	8	−0.36

Leads range from 5 to 16 months, with a median of 10.5 and a mean of 10.6. The 2020 entry counts as a hit, although that recession was set off by the COVID-19 pandemic. The New York Fed's 2005 FAQ by Estrella notes one earlier false signal: the 1966 inversion preceded a 1967 credit crunch and slowdown in production that the NBER did not classify as a recession.

The Fed staff alternative: a near-term spread

Fed Board economists Eric Engstrom and Steven Sharpe argued in a 2018 FEDS Note, (Don't Fear) The Yield Curve, for a "near-term forward spread": the implied forward rate on a 3-month bill six quarters ahead minus today's 3-month yield, a gauge of whether markets expect Fed cuts within about 18 months. In the note and a later working paper, it statistically dominated the 10-year minus 2-year spread in recession models.

The 2022–2024 inversion

From Treasury's daily data and the New York Fed's monthly file:

10-year minus 2-year: negative every day from July 6, 2022 to August 26, 2024 (537 trading days); deepest −1.08 on July 3, 2023.
10-year minus 3-month: first negative close October 18, 2022; negative every day from October 25, 2022 to December 12, 2024 (534 trading days); deepest −1.89 on May 4, 2023.
Monthly averages: negative for 25 consecutive months, November 2022 to November 2024, the longest run since the series begins in 1959; the previous longest was 18 months, from November 1978 to April 1980. The May 2023 low of −1.71 was the deepest monthly reading since 1981.
The 2025 echo: the daily 10-year minus 3-month spread closed below zero on 86 of 249 trading days in 2025, in runs of up to 18 days; the monthly average was −0.02 in March and −0.03 in April. It has closed positive every day since October 17, 2025.

By October 2026, 47 months had passed since the first monthly inversion, about three times the longest lead in the table. The NBER's list of announcements ends with the July 19, 2021 determination of the April 2020 trough. The NBER dates turning points after the fact (the February 2020 peak was announced on June 8, 2020), so a recent peak could still be identified, but the record so far is a miss. The Cleveland Fed's yield curve page lists the inversion from late 2022 through late 2024 as a false positive, alongside the late-1966 inversion and a very flat curve in late 1998.

Explanations offered, not settled

A compressed term premium. Both models were negative at the deepest point of the inversion, so long yields sat below the estimated expected path of short rates. One suspected cause is the Fed's bond holdings: a 2017 FEDS Note by Bonis, Ihrig and Wei estimated that they had lowered the 10-year term premium by about 1 percentage point at the end of 2016. The Fed began shrinking its holdings on June 1, 2022. The explanation does not cover the whole episode, though: on October 19, 2023 both premium estimates were positive (+0.45 and +0.66), yet the 10-year minus 3-month spread was still −0.62 and stayed negative for another 14 months.

The evidence cuts both ways. A 2008 New York Fed study by Rosenberg and Maurer found that expected rates carry the predictive power and the term premium alone predicts poorly; a recession followed the 2006 inversion that some had blamed on a low premium. A 2018 San Francisco Fed letter by Bauer and Mertens found both components carried the same recession signal and that the term spread's signal "is not diminished."

Expected disinflation rather than expected recession. A July 2022 FEDS Note by Ajello, Benzoni, Schwinn, Timmer and Vazquez-Grande split the near-term spread into policy gaps, the slope of expected inflation and a term premium. In June 2022 policy was still accommodative while inflation was expected to fall, a mix not seen before any recession in their sample since 1962, so recession odds were low. But the note predates the 10-year minus 3-month inversion and projected odds rising to about 35% by the end of 2023, or near 60% under tighter policy. It shows a curve can price cuts after disinflation, not that this explains 2023–2024.

What the curve cannot tell you

Timing. A 5- to 16-month range is wide, and the start date depends on how inversion is measured. Daily data put the first 10-year minus 3-month inversion before the 2007 peak on January 17, 2006, seven months before the monthly average turned; before the 2020 peak, March 22, 2019 versus June 2019. In 2022 the 10-year minus 2-year spread first dipped below zero on April 1 and stayed negative from July 6, more than three months before the 10-year minus 3-month run began on October 25; Estrella's FAQ warns that a benchmark built on one spread may not work for another.

Severity. Depth of inversion has not mapped to depth of recession. The 2007–2009 recession lasted 18 months, the longest since World War II, after a monthly spread that never went below −0.51 points. The 1980 recession lasted 6 months after a spread as low as −2.20 points.

What to do with your portfolio. The research measures recession probabilities, not the timing of asset prices, and none of it supports buying or selling on an inversion or a re-steepening.

Frequently asked questions

Why use the 3-month yield rather than the 2-year?

Estrella's New York Fed FAQ says background research for the 1998 paper found the 3-month rate, paired with the 10-year, gave a reasonable combination of accuracy and robustness over long periods. Most term spreads are highly correlated, so the choice is mainly fine tuning; the 2-year may be popular because those notes are highly liquid.

How should a recession probability from the curve be read?

It is a model output, not a call. In Estrella and Mishkin's 1996 table, a quarterly average spread of +0.76 points corresponded to a 10% probability of recession four quarters later, −0.82 points to 50% and −2.40 points to 90%. Even at 30%, a well-calibrated model implies that most periods with that reading are not followed by a recession.

Does it matter whether short rates rose or long rates fell?

According to the New York Fed FAQ, no: it is the level of the spread that helps forecast recessions, not the change in the spread or the source of that change. Because the model is not linear, a 50-basis-point drop hardly moves the probability when the spread is above 3 points, but from 0.5 points it can raise it by 10 percentage points or more.

Does the yield curve signal work outside the United States?

The New York Fed FAQ cites evidence of similar predictive relationships in Germany, Canada and the United Kingdom, from Estrella and Mishkin (1997) and Bernard and Gerlach (1998). US studies use Treasury yields because consistent data run from the 1950s and Treasuries carry no significant credit risk premium that could vary with maturity.

When is new yield curve data published?

Treasury posts each trading day's curve usually by 6:00 p.m. Eastern Time, though delays can occur. FRED's spreads draw on the same data; the New York Fed's probability model is updated monthly.

Reading the curve yourself

Open Treasury's daily par yield curve table and note the date of the latest row.
Compute 10-year minus 3-month and 10-year minus 2-year, and check them against FRED.
Look for persistence: Estrella's FAQ says a one-day signal may be dismissed but one lasting a month or more deserves attention. Near zero even monthly averages flip, as the March–April 2025 readings of −0.02 and −0.03 did.
Check the New York Fed's probability and the data month behind it.
Compare the ACM and Kim-Wright estimates to see whether a move in long yields came from rate expectations or the premium.
Treat the NBER's announcements as the official dating of US recessions, and expect them to lag.