The Yield Curve Manual

Research note. Every figure computed at render from our daily Treasury-spread file (10Y−2Y from 1976, 10Y−3M from 1982) — the same data behind the live yield curve page. Recession dates are NBER's. Methods in How we checked it; editorial standards in How we use AI.

What is the yield curve?

Line up Treasury yields from short maturities to long and connect the dots — that line is the yield curve. Normally it slopes upward: locking money away longer commands more yield. The one-number summary everyone quotes is the 10-year minus 2-year spread, and its sign is the headline. Positive is normal. Negative — inversion — means the market expects the Fed to cut rates substantially in coming years, and for five decades the only thing that reliably forced such cuts was recession. That is the entire logic of the indicator: not that inversion causes downturns, but that it is the bond market putting a probability on one.

Is the curve inverted right now?

As of August 6, 2026: no. The 10Y−2Y spread is +0.44pp and the 10Y−3M is +0.79pp — both positive. The 2022–24 inversion ended August 2024, which puts today 24 months into the post-un-inversion window — the stretch of the cycle where, in 2000 and 2007, the recession actually began. The live page charts both spreads daily, and the curve is also cross-listed under Risk in our nav for exactly this reason.

The two spreads we track — and why they disagree

Both use the same 10-year far end; the near end is the choice. The 2-year is a forecast — it embeds where markets think the policy rate will average over the next two years — so 10Y−2Y inverts early, as soon as cuts are priced anywhere on the horizon. The 3-month bill sits on top of the current policy rate, so 10Y−3M inverts late, only when the stance is actually restrictive against long-run expectations. The disagreement is informative: 2019 saw the 2-year version close negative for just three sessions while the 3-month version spent five months inverted — and the academic literature (Estrella & Mishkin's probit work, the New York Fed's recession-probability model) has always preferred the 3-month version that headlines ignore. Our data runs daily from 1976 for 10Y−2Y and 1982 for 10Y−3M, from FRED's constant-maturity series.

Every inversion cycle since 1976, computed

Eight cycles. First inversion day, deepest reading, total inverted sessions, and the lead from first inversion to the NBER recession start — all computed from the data inside each window:

Notes the table can't hold: the 1998 row is the classic false alarm — a brief LTCM-panic inversion, no recession for another three years. The 2019 row is genuinely ambiguous — three sessions of 10Y−2Y inversion, months of 10Y−3M inversion, and then a recession that arrived via pandemic, which no bond market predicted; we count it with an asterisk the table states. And the leads that made this indicator famous — 10 to 24 months — are far too wide to time anything with. Inversion has been a probability statement, never a calendar.

2022–24 — the record breaker still on trial

The most recent episode broke every scale in the table: 537 trading days of continuous inversion (versus 423 in 1978–80, the previous longest), a deepest print of −1.08pp in July 2023 — the most negative since 1980 — and, 24 months after it ended, no recession. Explanations differ and matter: inflation, not recession, was what markets expected the Fed to cut against this time, so the inversion's meaning genuinely changed; and years of QE compressed the term premium, making the long end structurally lower than history's. Either the indicator failed, or its lead time is still running, or this inversion measured something different. We present all three readings because the data cannot yet distinguish them — that is the honest state of the most famous signal in macro.

The un-inversion trap

The least-known fact in this Manual: historically, the dangerous moment was not the inversion — it was the return to normal. The curve un-inverted in late December 2000; recession began in March 2001. It un-inverted through mid-2007; recession began that December. The mechanism is mechanical: curves re-steepen fastest when the Fed starts cutting short rates into visible weakness — which means rapid steepening after a long inversion has usually been the sound of the easing cycle starting too late. When headlines celebrated the 2024 un-inversion as the all-clear, they were celebrating the exact pattern that preceded the last two recessions. So far this cycle has defied that history too — 24 months and counting — which belongs in the same open verdict as the section above.

Where this gauge will mislead you

• The lead is unusable for timing. 10 to 23 months from first inversion to recession in our record — and equities often rallied hard inside that window. 2006–07 inverted with a year of gains still ahead.
• Un-inversion is not the all-clear. The section above. Steepening driven by Fed cuts into weakness is the late-cycle pattern, not the escape.
• Which spread you quote changes the verdict. 2019: three days inverted on 10Y−2Y, five months on 10Y−3M. Anyone arguing a "signal" without naming the spread is choosing the answer first.
• The 2022–24 episode weakened every base rate. One enormous counterexample now sits in an eight-cycle sample. Quoting the old "perfect record" without it is folklore.
• Structural shifts contaminate levels. QE-era term premiums make today's curve flatter than the 1980s at the same expectations. Cross-era depth comparisons (ours included) are directional, not precise.
• It says nothing about stocks on your horizon. This is a business-cycle gauge. For market-facing risk, our Cycle Score and credit spreads operate on the horizons where portfolios live.

The last 12 sessions

Full series as CSV: yield-curve.csv.

How we checked it

Spreads are FRED's daily constant-maturity series, differenced (DGS10 − DGS2, DGS10 − DGS3MO), maintained in our pipeline. Cycle windows are stated identities; inside each, the first-inversion date, deepest reading, inverted-session count and last-inversion date are computed from the closes, and leads are calendar months from first inversion to the NBER-dated recession start. The 2022–24 episode's "months since un-inversion" recomputes daily. Brief sub-10-session dips are reported inside their cycle rather than as separate episodes. Everything recomputes at render, so this Manual always matches the live page.

Frequently asked questions