The yield curve and inversion

Plot yield against maturity and you get the yield curve — the single most-watched picture in macro, and the second of the roadmap's self-assessment targets.

   yield
     │                          ╭──────────  NORMAL (upward-sloping)
     │                     ╭────╯
     │                ╭────╯
     │           ╭────╯
     │      ╭────╯
     │ ╭────╯
     └─┬────┬────┬────┬────┬────┬────┬────▶  maturity
      3m   1y   2y   5y   10y  20y  30y

What the shape means

From M4.2, a long rate is roughly the average of expected future short rates. Add one more piece and the whole curve is explained:

   long yield  ≈  average expected future short rates  +  TERM PREMIUM
                  ───────────────────────────────────     ─────────────
                  what the market thinks the central       compensation
                  bank will do                             for the risk
                                                           of being wrong

Term premium exists because lending for thirty years is riskier than rolling overnight — you bear inflation risk, rate risk, and the duration exposure of M4.1. Normally you must be paid for that, which is why the curve usually slopes up even when no rate change is expected.

That decomposition matters because the two components have completely different meanings. A rise in long yields driven by expectations says the market expects tighter policy; one driven by term premium says the market wants more compensation to hold duration — a fiscal-supply or uncertainty story. QE (M4.2) works almost entirely on the second.

Inversion

An inverted curve — long yields below short — is strange on its face. Why accept less for locking money up longer?

Because you would only do it if you expected short rates to be much lower in future. Inversion is the market saying the policy rate is going to be cut, and the central bank cuts when the economy is deteriorating.

   INVERTED
   yield
     │ ╰────╮
     │      ╰────╮
     │           ╰────╮
     │                ╰────╮
     │                     ╰─────────────
     └─┬────┬────┬────┬────┬────┬────┬───▶  maturity
      3m   1y   2y   5y   10y  20y  30y

   short rates HIGH — the central bank is currently tight
   long rates LOWER — the market expects cuts, i.e. a downturn

So the curve is not a mystical indicator. It is an aggregated forecast, and the reason people respect it is that this particular forecast has an unusually good record.

The standard measure is the 2s10s — the 10-year yield minus the 2-year.

The record, honestly

Every US recession since 1976, checked against every sustained 2s10s inversion:

      inversion episode        days   next recession   lead time
   ─────────────────────────────────────────────────────────────
   1978-08-18 → 1980-05-01      423     1980-02-01     17 months
   1980-09-12 → 1982-07-16      400     1981-08-01     11 months
   1988-12-13 → 1989-11-06      174     1990-08-01     20 months
   1990-03-08 → 1990-03-29       16     1990-08-01      5 months
   1998-05-26 → 1998-07-27       27     2001-04-01     34 months
   2000-02-02 → 2000-12-28      227     2001-04-01     14 months
   2005-12-27 → 2007-06-05      238     2008-01-01     24 months
   2022-07-06 → 2024-09-05      539       none yet

Three readings, and you need all three.

The hit rate is remarkable. Every one of the six NBER recessions in this window was preceded by an inversion. Very few macro indicators can claim that.

The lead time is useless. Five months to thirty-four. An indicator that says "a recession will begin somewhere between next quarter and three years hence" cannot be traded, and the 1998 episode — 34 months to a recession that arrived for unrelated reasons — is arguably a false positive that history flattered.

The most recent one has not worked. The 2022–24 inversion ran 539 days, the longest in the whole sample, ended in September 2024, and no recession has followed. That is either the first clean failure, or a lead time longer than any on record. Anyone who tells you confidently which is being clever after the fact.

The honest summary: a genuinely good recession indicator, with a lead time so variable it cannot time anything, and one live counterexample.

Why it might work — and why it might stop

Two families of explanation, and they differ in whether the signal survives.

Causal. An inverted curve is bad for banks, which borrow short and lend long — it compresses their margin, so they lend less, and less credit slows the economy. Here the curve contributes to the recession, and the signal should keep working.

Informational. Inversion is only the bond market forecasting cuts, and the bond market happens to be good at that. Here the curve causes nothing and merely aggregates. The signal survives only while that forecast stays good.

There is also a specific reason for caution now: after a decade of QE, term premium was heavily compressed by central bank buying (M4.2). If the normal upward slope is partly term premium and term premium has been artificially squeezed, then the curve inverts on smaller expectation changes than it used to — the threshold moved, so the historical base rates may not transfer.

Which is a good general lesson about macro indicators, and the connection back to M3.7: a relationship estimated across regimes may not survive a change of regime, and the sample here is six recessions. Six. Every confident claim about the yield curve rests on a sample size you would laugh at in any other context.

Source: Campbell Harvey’s work is the origin of the curve-as-recession-predictor literature. The New York Fed publishes a recession-probability model built on the 3m10y spread and its methodology is public. Numbers above computed from FRED series DGS10, DGS2 and USREC — the NBER indicator — and reproducible in a page of code.