Lesson 5 of 8

The yield curve

Step 3 of 6

Worked example

The 2022–24 inversion (FRED T10Y2Y and T10Y3M, NBER table, all pulled 20 Nov 2026).

T10Y2Y dipped below zero on 1 and 4 Apr 2022. It was then continuously inverted from 6 Jul 2022 to 26 Aug 2024, more than two years. The deepest point was −1.08 on 3 Jul 2023: the 2-year paid over a full percentage point more than the 10-year. Latest: +0.28 on 19 Nov 2026.

T10Y3M was continuously inverted from 25 Oct 2022 to 12 Dec 2024, with a low of −1.89 on 4 May 2023. Latest: +0.70 on 19 Nov 2026.

What followed: no recession. The NBER's latest peak is February 2020 and its latest trough April 2020. Nothing has been dated since, and FRED's recession flag (USREC) is 0 through October 2026.

Compare that with every recession since T10Y2Y begins (1976). Months from the start of each inversion to the next recession's start (the NBER peak month):

Inversion starts (T10Y2Y) NBER peak Gap
18 Aug 1978 Jan 1980 ~17 months
12 Sep 1980 Jul 1981 ~10 months
13 Dec 1988 Jul 1990 ~19 months
26 May 1998 (briefly), again 2 Feb 2000 Mar 2001 ~33 / ~13 months
27 Dec 2005 Dec 2007 ~23 months
27 Aug 2019 Feb 2020 ~5 months
1 Apr 2022 none (as of Nov 2026) 55+ months

Two lessons. An inversion came before each of these recessions. But the gap ran from 5 months to nearly 3 years, and the 2020 recession was caused by a pandemic no bond market foresaw. Then 2022 broke the pattern, so far.

The model's own number. The New York Fed turns the monthly 10-year minus 3-month spread into a recession probability 12 months ahead. Its data file (allmonth.xls, updated 5 Nov 2026) shows:

  • the model's peak in this episode: 70.85% chance of recession in May 2024, from the May 2023 spread of −1.71;
  • today: 12.4% for October 2027, from the October 2026 spread of 0.98.

A 71% probability that didn't happen isn't proof the model is broken; 29% outcomes happen. It does show you can't time anything with it.