The Yield Curve as a Recession Signal: What History Tells Us

August 2006. The 2-year Treasury yielded more than the 10-year for the first time since 2001, after the Federal Reserve had raised its target rate 17 consecutive times, from 1.00% in June 2004 to 5.25% in June 2006. The yield curve had inverted. And the debate that followed — whether the old signal still applied, or whether structural changes to global bond markets had rendered it obsolete — was not a fringe conversation. It came from credible economists and, in some form, from the Fed itself. The argument that global demand for long-duration US Treasuries from Asian central banks and pension funds was compressing long yields for structural reasons, not recession-anticipating ones, had real analytical weight behind it.

The recession began in December 2007. Sixteen months after the inversion appeared. The signal worked. Those arguing against it were not wrong about the mechanisms they named — foreign buying of Treasuries was genuinely compressing long yields — but those mechanisms proved insufficient to override the three channels through which inversions historically feed into economic downturns. The historical analogue asserted itself, as it tends to do, despite the contemporary case for why it wouldn't.

That episode sits near the center of how I think about the yield curve as a tool. Not as a forecast — it doesn't tell you when or how badly — but as a probability shifter with a track record demanding serious attention, even when the contemporary arguments against it sound compelling. The full historical record is worth examining carefully, including the cases where the signal didn't work.

Every Inversion Since 1970: The Record

The cleanest way to examine the yield curve's predictive value is to define inversion consistently and map it against NBER recession dates. For readers less familiar with how the curve works mechanically, our yield curve primer covers the fundamentals. The 2-year minus 10-year Treasury spread will serve as the primary reference here — it's the most widely followed version — with the qualification that the 3-month versus 10-year spread is what the New York Fed uses in its formal recession-probability model, and it has a somewhat cleaner historical record. Both are worth tracking. Since 1970, the 2s10s has inverted ahead of every NBER-dated recession:

• Early 1973: Inversion preceded the November 1973 recession onset by roughly eight months. The first oil shock supplied the proximate trigger, but the credit tightening embedded in the inverted curve was already building.
• September 1978: Inversion appeared as the Fed began its pre-Volcker tightening cycle. The recession that followed started January 1980 — a sixteen-month lag, the longest in the sample to that point.
• September 1980: A second inversion in the Volcker cycle preceded the July 1981 recession onset by roughly ten months, at one of the deepest inversions on record as short rates reached historically extreme levels.
• Late 1988 through 1989: The curve inverted through most of 1989 under Greenspan's post-Black Monday tightening. Recession began July 1990, a lag of approximately twelve months.
• March–April 2000: A brief but sustained inversion followed 175 basis points of rate increases between June 1999 and May 2000, with the federal funds rate reaching 6.5%. The 2001 recession began in March — eleven months later.
• August 2006: As described above. Recession began December 2007, sixteen months after the inversion first appeared — the longest lag in the modern dataset.
• March–August 2019: The 3m-10y spread inverted in March 2019; the 2s10s followed briefly in August. The recession began February 2020, though the COVID shock makes the causal chain harder to read than any prior episode.

The average lag across these seven pre-recession inversions runs roughly twelve to thirteen months. But the range — eight to sixteen months — is wide enough that the signal functions as probability context, not a timer. Knowing an inversion has occurred tells you a recession is more likely. It doesn't tell you when to expect it.

The False Signals: 1966, 1998, and 2022

Two inversions in the postwar record did not produce NBER recessions, and a third case remains unresolved. The 1966 inversion was brief and shallow. Real GDP growth decelerated sharply in 1966 and 1967 — some economists call this period a 'mini-recession' — but it never reached the NBER's threshold for a broad-based contraction across multiple sectors. The signal identified a genuine deceleration, just not a severe enough one to cross the formal line. A near-miss in the right direction rather than a wrong call.

The 1998 false positive came amid the Russian debt crisis and the collapse of Long-Term Capital Management. The 2s10s dipped briefly into negative territory that fall, but for a specific reason: foreign capital flooding into US Treasuries as a safe haven drove long yields sharply lower, compressing the spread from the long end rather than from Fed-driven short-rate increases. The Fed cut rates three times between September and November 1998, the crisis resolved, and the US economy continued growing. This is part of why the 3m-10y version generates fewer false positives: it's harder for cross-border safe-haven flows to flatten the very front end of the curve while leaving short-term policy rates unchanged.

Then there is 2022. The 2s10s inverted in early April 2022 and deepened to roughly -109 basis points by March 2023 — the most severe inversion since the early Volcker years. It began normalizing in late 2023 and fully un-inverted in 2024. The 3m-10y has since flipped negative again in recent weeks after a brief positive excursion in early 2026. As of mid-2026, no NBER recession has been dated following the original inversion — already a longer lag than any prior precedent by a significant margin. I'm genuinely less certain about what to make of this case than I'd like to be. It may resolve as a false signal, a historically anomalous long lag, or a recession the NBER will eventually date from 2025 or 2026. That uncertainty is real, and it won't resolve here. What it does illustrate is that the yield curve is a probability tool — one with an imperfect record — not a guarantee.

Why the Signal Has Worked: Three Mechanisms

Understanding why inversions precede recessions matters as much as knowing that they do — it helps calibrate how heavily to weight the signal when contemporary arguments against it sound credible.

The first mechanism is bank profitability compression. Commercial banks borrow short and lend long: they fund themselves with deposits and short-term liabilities, then extend mortgages and business loans at longer maturities. An inverted yield curve collapses the net interest margin — the spread between what banks earn on assets and what they pay on liabilities. When that margin compresses, banks tighten underwriting standards and pull back on credit extension. JPMorgan Chase's 2006 annual report disclosed that rising short-term funding costs were already compressing net interest income before business conditions had meaningfully deteriorated. The credit-tightening effect builds slowly — it takes months to show up in GDP data — which is one reason the lag between inversion and recession is measured in quarters rather than weeks.

The second mechanism is the expectations channel. An inverted curve implies that bond markets collectively expect the Federal Reserve to cut short-term rates in the future — because lower anticipated future rates are what pulls long yields below current short yields in a rational market. But an expectation of future Fed cuts is itself a forecast of weaker economic conditions. The curve doesn't directly predict a recession; it reflects a collective probability-weighted assessment of the future growth environment, and that assessment has historically been more accurate than contemporaneous economic consensus forecasts.

The third mechanism is the borrowing-cost squeeze. When short-term funding costs exceed the return on long-duration investment, marginal capital projects don't pencil out. Business investment slows. Consumer borrowing becomes more expensive in real terms. These effects compound over the same 6-12 month window as bank margin compression. The regime shift is gradual — which is why inversions lead recessions by a year or more, and why they're easy to dismiss in the moment before their full effect emerges.

A Framework for Reading the Signal

Treating the yield curve as a probability shifter rather than a forecast means being precise about what the signal should change in your thinking. Three steps, in order.

First, watch the right spread. The 2s10s is the most quoted and works well as a general reference. But the New York Fed's model uses the 3m-10y for the empirical reason that it has fewer false positives and a more stable lead time. The two spreads have diverged in recent weeks: as of late May 2026, the 2s10s remains positive (the 10-year at 4.56%, the 2-year at 4.13%), while the 3m-10y has re-crossed into negative territory. That divergence — the bond market is pricing Fed cuts ahead even as the 2-10 belly has normalized — is itself worth noting. Not every part of the curve is sending the same message right now.

Second, require persistence before weighting the signal heavily. A one-week inversion is noise. The inversion worth paying attention to has lasted a month or more and has deepened beyond the shallow range. In the 2006 case, the curve stayed inverted for eleven months and reached -55 basis points at its deepest — a committed signal. A shallow dip that un-inverts within weeks, particularly one driven by Treasury supply dynamics or cross-border capital flows, warrants considerably less weight than one driven by Fed tightening compressing short yields against anchored long yields.

Third, check the New York Fed's recession-probability estimate directly. The model translates the 3m-10y spread into a twelve-month-ahead recession probability using the Estrella-Mishkin probit framework. The current reading, given the recent re-inversion of the 3m-10y, puts the twelve-month probability above 30% — elevated relative to most non-recessionary years, but below the 50% threshold that preceded every NBER-dated recession with at least twelve months of lead time in the model's historical calibration. That range — elevated but not definitive — describes the appropriate posture: heightened analytical attention, not a portfolio overhaul.

What It Means for Stock Portfolios

The yield curve is not a market-timing device. An investor who sold equities every time the 2s10s inverted would have missed substantial returns during the typical 12-16 month window between inversion and recession — a period that often includes some of the strongest final innings of a bull market. The S&P 500 gained roughly 18% between the August 2006 inversion and its October 2007 peak. The relationship between interest rates and equity multiples runs through the discount-rate channel on a lag, not on the day the curve crosses zero.

But the signal does identify a regime worth preparing for. Two patterns hold across cycles. Financials tend to underperform from the date of inversion through the subsequent recession — the net interest margin compression hits banks directly and visibly, and the market prices it before the income statement confirms it. Defensive sectors tend to outperform on a relative basis as investors rotate toward earnings stability. These rotations are more reliable than trying to call the broad market's direction off the yield curve signal.

The second pattern is the quality dimension. Wide-moat businesses with genuine pricing power, conservative balance sheets, and non-discretionary demand have historically drawn down less in yield-curve-flagged recessions and recovered more quickly afterward. When the equity risk premium is already compressed — as it is in a period of elevated starting multiples — an inversion-flagged credit tightening does considerably more multiple damage to high-duration, cyclically exposed businesses than to stable compounders with durable earnings. The inversion isn't a signal to rotate into quality. It's a signal not to trade out of it.

Key Takeaways

The yield curve is a probability tool, not a forecast. Here is the framework for using it without overreading it:

• The 2s10s has inverted ahead of every NBER-dated recession since 1970, with average lags of 12–13 months and a range of 8–16. The 3m-10y, used in the New York Fed's formal model, has generated fewer false positives and is the preferred spread for probability-model purposes.
• Two confirmed false signals exist in the postwar record: 1966 (brief, shallow, coincided with a real deceleration that didn't reach the NBER recession threshold) and 1998 (driven by foreign safe-haven flows into Treasuries, not domestic credit tightening). The 2022 inversion remains unresolved as of mid-2026.
• The mechanism runs through three reinforcing channels: bank net interest margin compression, bond-market expectations of future rate cuts, and the borrowing-cost squeeze on business investment. All three build slowly — which is why the lag is measured in months, not in days from first inversion.
• Use the signal to shift probability assessments and to review portfolio quality — not to time the market. A persistent 3m-10y inversion lasting more than one month is worth weighting seriously. A brief dip driven by external capital flows is not. The difference requires understanding what is causing the inversion, not merely that it exists.