The yield curve's use as a recession signal rests on a single empirical regularity: when short-term Treasury yields rise above long-term yields — an inverted curve — a U.S. recession has reliably followed, typically within a year or so. The core tension is that this is among the most accurate macro forecasting tools ever documented, yet it is also a blunt one — it tells you a downturn is likely coming but not when, the lead time is long and variable, and the mechanism that makes it work can be distorted by central-bank policy. It is a probability shifter, not a timer.

Why an inverted curve signals recession

Long yields can be decomposed into the average expected path of short rates over the bond's life plus a term premium. When markets expect the Fed to cut rates in the future — usually because they anticipate weakening growth — the expected-short-rate component of long yields falls below the current short rate, and the curve inverts. So an inversion is fundamentally a market forecast that policy is too tight and will have to be eased, which historically coincides with the onset of recession. The Federal Reserve Bank of Chicago frames it as the curve impounding the pessimistic expectations market participants have already formed — a form of reverse causality, where the curve reflects a downturn the market sees coming rather than causing it.

The most-studied spread is the 10-year minus 3-month (10y–3m). The 3-month bill tracks the current policy rate closely, so the spread directly measures the gap between current policy and long-run growth/inflation expectations. Estrella and Mishkin found this pairing the most reliable; it is the basis of the New York Fed's official model. The popular 10y–2y spread is more widely quoted in the press but is empirically weaker (see below).

The NY Fed probit model

The standard quantitative version is Estrella & Mishkin's probit model, published monthly by the NY Fed:

> P(recession in t+12) = Φ(−0.5333 − 0.6629 × spread₁₀y₋₃ₘ)

estimated on monthly U.S. data (roughly 1959–1995), where Φ is the standard normal CDF and the spread is in percentage points. The model forecasts the probability of an NBER recession 12 months ahead. It is deliberately simple — adding the fed funds rate, lagged GDP, or stock returns improves in-sample fit only marginally and tends to overfit out of sample. Reported in-sample fit is strong (commonly cited pseudo-R² around 0.3 and AUC near 0.92), but in-sample fit overstates real-time reliability.

How it's used in practice

Practitioners watch for three things, not just a single day's print:

• Sign: has the spread crossed below zero (inverted)?
• Persistence: Campbell Harvey's original work emphasized a sustained inversion — he uses a minimum of roughly a full quarter (three months) of inversion before treating it as a signal, to filter noise.
• Lead time: historically the gap from first inversion to recession onset has run roughly 6 to 24 months, with a commonly cited average around 12 months. The 2006–07 episode had the longest historical lag — about 22 months from the 10y–3m inversion to the December 2007 recession onset (a figure Harvey himself cites as the prior record).

Asset allocators use rising model-implied probabilities to trim cyclical/credit exposure and lengthen duration ahead of expected Fed easing. The signal's long lead is a feature for strategic positioning and a bug for tactical timing.

Adoption, debate & evidence

The signal is genuinely well-established, not folklore. The 10y–3m spread has inverted before every U.S. recession since the late 1960s, with one widely cited false positive (the mid-1960s). The original insight traces to Campbell Harvey's 1986 Chicago dissertation — though he used a 5-year minus 90-day spread, not the 2-10 the media favors. It is taken seriously by the Fed, academics, and institutional investors alike.

But there are real, current disputes:

1. Which spread. A 2018/2022 Fed study (Engstrom & Sharpe, "(Don't Fear) The Yield Curve") argues the popular 10y–2y spread adds no incremental information once you use a near-term forward spread — the difference between a ~6-quarters-ahead forward 1-quarter rate and the current 1-quarter rate. This short-maturity measure reflects expected Fed policy over ~18 months with cleaner interpretation and, they argue, better recession-forecasting power.

2. Term-premium distortion. Long-term term premia have been compressed since the 2008 crisis, partly by central-bank asset purchases (QE) and price-inelastic buyers (pensions, insurers). A lower term premium mechanically flattens or inverts the curve for reasons unrelated to growth expectations, so some Fed and BIS researchers argue term-premium-adjusted models implied lower recession odds than the raw spread.

3. The 2022–2024 episode. The 10y–3m curve inverted in October 2022 and stayed inverted until December 2024 — roughly 26 months, the longest inversion duration on record (surpassing the late-1970s ~19-month mark) — yet by the time it un-inverted no NBER recession had materialized, well past the prior ~22-month inversion-to-recession lag of 2007. (Note these are two different metrics: how long the curve stays inverted versus how long after inversion the recession arrives.) Harvey himself, the indicator's pioneer, publicly cautioned it "may be a false signal" this time, citing the unusual post-pandemic and post-zero-rate environment. He stresses that a simple model will eventually produce a false signal; the open question is whether this was it. Whether the cycle ultimately turns into a delayed recession remains undetermined as of this writing.

Strengths & limitations

Strengths: Long, robust track record across many decades and regimes; a clear economic mechanism; simple, transparent, hard to overfit; available in real time and revision-free (unlike GDP).

Limitations: Very long and variable lead time — useless as a precise timer. Small sample of actual recessions (only about nine NBER recessions since 1959) means apparent accuracy rests on very few events. Vulnerable to term-premium distortion from QE. The #1 misuse is treating a single brief inversion — or any inversion regardless of which spread or how long — as a deterministic "recession is here now" trigger, and trading short-term off it. It raises a 12-month probability; it does not date the peak. Pair it with corroborating data (employment, credit spreads, leading indicators) rather than acting on it alone.

Sources

• New York Fed — The Yield Curve as a Predictor of U.S. Recessions (Estrella & Mishkin); monthly probability series.
• Federal Reserve — Engstrom & Sharpe, (Don't Fear) The Yield Curve, Reprise (2022), federalreserve.gov.
• Federal Reserve Bank of Chicago — Why Does the Yield-Curve Slope Predict Recessions? (2018).
• Duke Fuqua / CNBC / Fox Business — Campbell Harvey on his original research and the 2024 false-signal caution.
• BIS — Yield curve inversion and recession risk (Sept 2019); ECB Economic Bulletin (2020) on term-premium effects.
• St. Louis Fed FRED series T10Y3M; NPR The Indicator on the 2022–2024 inversion record.

Dispute flagged: genuine, unresolved disagreement over (a) best spread (10y–3m vs near-term forward vs 10y–2y) and (b) whether the 2022–2024 inversion is the model's first true false positive.