The rule

The curve is FRED's daily 10-year minus 3-month Treasury spread (T10Y3M), from Jan 1982. Each month is read at its last close; a month below zero is an inverted month, and consecutive inverted months are one inversion. The rule is the documented one — inversion — from the research that first tied it to recessions (Estrella and Mishkin, 1996); it was fixed before any measurement and has not been tuned.

The basis

Treasury yields are market prices: recorded as they trade, and never revised. Each month's reading is its last close, known that day, so the record is first-published by construction — nothing in it depends on data published later. The NBER's peak and trough dates are set after the fact, often a year or more later; they are used only to score what followed, never to decide when the curve fired.

The test, and its five parts

Each inversion is scored on one question: did an NBER business-cycle peak follow within 18 months of the month it began? Inversions that begin inside a recession are recognition, not warning, and are not scored. An inversion is scored once 12 months have passed after its 18-month window closes — time for the NBER, whose longest announcement lag in this record is 12 months, to declare any recession inside it; until then it reads “No recession declared yet”, and after that, with none declared, it counts as a miss. The base rate is the same question, under the same rule, asked of every month in the record.

The curve scored 7 of 9 against a base rate of 16.5% — 4.71 times as often. Random placements of the same number of inversions scored a median of 1 and a 95th percentile of 3.

Removing one episode at a time.

The 2-year to 10-year curve

The same rule and the same test on FRED's 10-year minus 2-year spread (T10Y2Y), from Jun 1976: 8 of 11 scored inversions were followed by a recession within 18 months, against a base rate of 20.7% — 3.51 times as often; 3 were not. It passes every test. Read at month-ends it did not invert before the recession that began in Feb 2020. Its series has its own page.

Against this site's own signals

For each inversion that a recession followed: did it begin before the stress signal opened, and before the Slowdown or Contraction stage began, in the 24 months before the NBER peak? Both come from the site's own record as first published, which starts in Aug 1981 — nearly all first-published from Sep 2016; earlier years partly use today's revised data (how much). The curve came first 2 times of 7 against the stress signal, and 3 against the stage.

Seven relationships that did not hold

Until September 2026 this page followed eight cross-asset relationships, each documented by practitioners or researchers before it was tested here, each with a rule fixed before measurement. In September 2026 every one was tested against the outcome its page claimed, by the test above. The yield curve held. The records below are why the others are no longer followed: each is kept in full, firing by firing.

The S&P 500 outcomes are measured on total return · Shiller monthly convention, at month-end closes (see "The total-return basis" on the Confluence methodology).

These records are as measured in September 2026: a firing whose window ran past the NBER dates then known (through July 2026) reads “not yet scored”. The curves above are scored by the rule under “The test”, and follow on their own.

One more was retired in July 2026, before the others were tested, when it failed the test it was built on. Its record on the same test: