What an Inverted Yield Curve Says—and What It Cannot Predict

The definition, and why the pair matters

A yield curve plots yield against time to maturity for bonds of the same credit quality. It is normally upward-sloping: lending for longer carries more risk and usually more compensation. It is inverted when that relationship reverses at some part of the curve.

The most quoted measure is the 10-year minus 2-year spread, published by FRED as series T10Y2Y. A negative value means the 10-year yield is below the 2-year yield.

That definition already contains a warning. “The curve” is not one object. The 10-year minus 2-year spread and the 10-year minus 3-month spread compare different horizons, and the 3-month bill is tightly linked to the current policy rate while the 2-year responds to expectations of policy over the next two years. The two spreads can disagree, and when they do, the disagreement is information rather than noise.

The current reading

As of 2026-10-23, the 10-year minus 2-year spread stood at +0.31 percentage points, according to FRED series T10Y2Y retrieved 2026-10-24. The curve was not inverted on that date.

The components: on 2026-10-22 the 10-year constant maturity yield closed at 4.96% and the 2-year at 4.65%, per FRED series DGS10 and DGS2, both retrieved 2026-10-24.

The relevant history in the same series:

These are counts and dates from one published series, not interpretations. The interpretation starts when you ask what produced the negative numbers.

What a long yield is made of

A long-term yield is not a forecast of one number. It is approximately the average expected short-term rate over the bond’s life, plus a term premium — the extra return investors demand for committing capital for longer instead of rolling short securities.

That decomposition is the key to reading inversion correctly. If the 10-year falls below the 2-year, at least one of two things is happening:

1. Expectations component: the market expects short rates to be lower, on average, over years three to ten than over the next two years. That is the recession-story version: policy is high now, growth is expected to slow, and rates are expected to fall.
2. Term premium component: the compensation for holding long bonds has fallen, so long yields are lower than the expected path of short rates would imply. That is a demand-and-risk story rather than a growth story.

The same curve shape can come from either. The New York Fed publishes model-based term premium estimates precisely because this component is not directly observable.

A worked example: implied forwards and the premium

Take the 2026-10-22 yields: 2-year 4.65%, 10-year 4.96%. The implied 8-year rate, two years forward, is the rate that connects the two:

• (1.0496)^10 / (1.0465)^2 = (1 + f)^8
• f = 5.0376%

The market is pricing an 8-year rate starting in two years at about 5.04%, above both spot yields. That is the mechanical consequence of an upward-sloping segment, and it is a useful reference point.

Now add an assumption. Suppose an investor believes the average short rate over those eight years will be 3.60%. The difference between the implied forward and that expectation is a term premium of roughly:

• 5.0376% − 3.60% = 1.44%

That 1.44% is not a published number; it is what the arithmetic requires given the assumption. An investor who thinks the expected average is 4.50% gets a much smaller premium. The curve carries the combination, not the split.

The same arithmetic runs in reverse for inversion. With the 10-year at 4.96%, a 2-year yield of 5.20% would put the spread at −0.24. A 2-year of 5.46% would put it at −0.50. Whether that inversion meant recession would depend on whether the 5.46% reflected expected policy declines or a compressed term premium.

What the record supports

The historical association is real enough to be worth stating precisely: inversions of the 10-year minus 2-year spread have occurred before several U.S. recessions, and the 537-day run that ended in August 2024 was followed by a period in which the spread turned positive again.

What the record does not support is a schedule. The interval between the first inversion and the start of a downturn has varied across cycles, and the number of completed cycles in the sample is small. A base rate computed from a handful of observations has wide error bars, and the error bars are the honest part of the statistic.

There is also a selection problem. Analysts notice the inversions that were followed by recessions. An inversion followed by a soft landing is remembered as a false signal and quietly dropped from the tally. The 2,049 negative days in the series include runs that ended without a downturn.

What it cannot predict

Timing. The curve has no clock. A signal that appears twelve months before an event and a signal that appears two months before it look identical on the day it appears.

Depth. Nothing in a yield spread measures how far output will fall or how high unemployment will rise. The curve is a price, not a quantity.

Whether. The expectation embedded in the curve can be wrong. Markets priced a policy path in 2026 that may or may not arrive; the pricing is evidence about beliefs, not about outcomes.

The path of other assets. An inverted curve and an equity drawdown are not the same event. The curve says something about expected short rates and term premia; it says nothing direct about earnings, credit spreads or risk appetite.

The cause. This is the largest limitation. Inversion driven by a falling term premium is a different signal from inversion driven by expected rate cuts, and the spread alone does not distinguish them. The New York Fed’s term premium estimates exist because that component has to be modelled, not read off a quote.

The real rate. Nominal yields include inflation compensation. A curve that is inverted in nominal terms may look different in real terms if inflation expectations differ across maturities. Comparing nominal spreads without that adjustment attributes the whole shape to growth expectations.

How to use it without over-reading it

Treat the spread as one input with a known decomposition, and be explicit about which part you think is moving.

If you believe the inversion is expectations-driven, the trade is about the future path of policy, and the relevant question is whether the market’s expected path is too aggressive or too cautious. If you believe it is term-premium-driven, the relevant question is about duration demand, issuance and risk appetite. Those are different analyses with different failure modes, and the spread alone does not tell you which one you are in.

The second discipline is to state the horizon before the signal appears. “Inversion implies recession” is unfalsifiable without a date, which makes it useless for decisions. “Inversion implies a lower 2-year yield within twelve months” is testable, and the historical hit rate can be measured.

The third is to avoid translating a probability into a position size without a separate risk budget. The duration guide covers how much a given yield move matters to a bond position; the repo versus policy rates guide covers how the front end is anchored in practice.

How these sources were checked

All yield levels, spread values, observation counts and dates in this article come from three FRED series retrieved 2026-10-24: T10Y2Y for the 10-year minus 2-year spread, DGS10 for the 10-year constant maturity yield, and DGS2 for the 2-year. The counts were computed from the full downloaded series, not quoted from commentary: 12,596 valid observations from 1976-06-01, 2,049 with a negative spread, a longest consecutive negative run of 537 days from 2022-07-06 to 2024-08-26, and a minimum of −2.41 on 1980-03-20.

The definition of the term premium and the existence of published model-based estimates come from the Federal Reserve Bank of New York’s Term Premia page, retrieved 2026-10-24. The implied forward rate and the term premium decomposition are arithmetic on the stated yields; the 3.60% expected-short-rate assumption is illustrative and is labelled as an assumption, not a measurement. No recession dates are asserted in this article because none were verified against a primary source for this piece.