The yield curve as a recession signal

The 10-year Treasury yield minus the 2-year yield is one number, and in monthly data it went negative before five of the six US recessions since 1976. Negative means the curve is inverted: lending for two years pays more than lending for ten. Gray bands are NBER recessions.

What the spread measures

• A Treasury yield is what the government pays to borrow for a given term. Lending for longer normally pays more, so the spread is usually positive.
• Investors accept less for ten years than for two only when they expect short rates to fall. Short rates fall when the Fed cuts, and the Fed cuts when the economy weakens.
• So an inversion is the bond market pricing a slowdown before GDP or unemployment shows it. It does not cause the recession.

How far ahead it moved

• The lead ran from 10 to 22 months. That range is the main reason to treat the curve as a warning, not a timer: "inverted, so recession now" is the common misreading.
• The 2020 recession had no monthly inversion before it. A pandemic shutdown is not the kind of slowdown a rate market prices a year ahead.
• The longest inversion in the record began in July 2022 and ran 26 months. As of the latest data, NBER has dated no recession after it. NBER announces turning points months after the fact, so the record can still change.

From a sign to a probability

A sign change is a blunt signal. A probit model turns the spread into a probability of recession twelve months ahead. Here it is refit every month from 1990 on, using only the outcomes known at the time, so each reading is one a reader could have seen then.

• Ahead of the 2001 and 2008 recessions the readings peaked at 32 percent (April 2000) and 23 percent (November 2006). Modest numbers, eleven and thirteen months before the recessions began.
• The highest reading came during the 2022 to 2024 inversion, and no recession followed within the twelve months it pointed at. A model fit on a handful of recessions carries that much uncertainty.
• The New York Fed's published version uses the 10-year minus 3-month spread. This one uses the same 10-year minus 2-year series as the chart above, so the two posts on this site never disagree.

Reading it without overreading it

• Watch the crossing, not the wiggle. Small moves around a positive spread are noise. The signal is the line crossing zero, and how deep and how long it stays below.
• Check the probability against the range. A reading of 10 percent and one of 50 percent mean different things, and neither is a forecast of a month.
• Use it with other data. The curve belongs next to labor, credit and consumer series. The Data Lab puts it beside unemployment, inflation and the fed funds rate, with twelve-month forecasts.

How the charts are built

• A script pulls the 10-year and 2-year constant-maturity yields from FRED's public endpoint and writes a JSON file. It is refreshed by hand; the as-of month is on every card.
• The forecast pipeline reads the same file and writes the probability series, with each month's model fit only on data available that month.
• Every chart here is server-rendered SVG from those files. The recession bands are the official NBER dates.

The spread with a time-range toggle is in the interactive version.