Reading the Yield Curve: What 2026's Inversion Tells Us

The Most Reliable Signal in Macroeconomics

Since 1955, every US recession has been preceded by an inversion of the yield curve — a condition where short-term Treasury yields exceed long-term yields. The track record is remarkable: seven correct signals in seven recessions, with only two brief false positives (1966 and 1998) that produced slowdowns rather than technical recessions.

The yield curve works as a recession predictor because it reflects the collective judgment of the bond market about future economic conditions. When investors expect growth to slow, they buy long-term bonds for safety, pushing long-term yields down. When the Fed raises short-term rates to fight inflation, short-term yields rise. The resulting inversion is the bond market's way of saying: "The Fed is tightening into a weakening economy."

The spread most commonly watched is the 10-year minus 3-month Treasury spread (T10Y3M), which the NY Fed uses in its official recession probability model. As of late March 2026, this spread stands at approximately -0.45%, having been negative since mid-2025.

The Estrella-Mishkin Probit Model

The NY Fed's recession probability model was developed by Arturo Estrella and Frederic Mishkin in their 1998 paper "Predicting U.S. Recessions: Financial Variables as Leading Indicators." The model uses a probit regression — a statistical technique that estimates the probability of a binary outcome (recession or no recession) based on a continuous input (the yield spread).

The model's specification is elegant in its simplicity. It takes a single input — the T10Y3M spread — and produces a probability using two coefficients: β₀ = -0.5333 (the intercept) and β₁ = -0.6330 (the slope on the spread). The probability is computed as Φ(β₀ + β₁ × spread), where Φ is the standard normal cumulative distribution function.

At the current T10Y3M spread of approximately -0.45%, the model outputs a recession probability of roughly 18%. This may seem surprisingly low given the geopolitical crisis, but the model is designed to capture the yield curve's signal about credit conditions and monetary policy — not supply-side shocks. The probit model is a monetary policy thermometer; it does not measure geopolitical fevers.

Historical Timing: Inversion to Recession

One of the most important features of the yield curve signal is its lead time. Historically, the gap between initial inversion and recession onset has ranged from 6 to 24 months, with a median of approximately 14 months.

The 2006-2007 inversion preceded the Great Recession by 23 months — long enough that many commentators had dismissed the signal as a false positive before the recession began. The 2019 inversion preceded the 2020 recession by roughly 8 months, though the proximate cause was the COVID pandemic rather than the credit cycle the curve was measuring.

The current inversion began in approximately September 2025, placing us roughly 6 months into the inversion window. If the historical median holds, the recession risk window peaks around November 2026 — but the Hormuz crisis could accelerate the timeline significantly.

An important nuance: the yield curve often un-inverts (returns to positive slope) shortly before the recession actually begins. This un-inversion occurs because the Fed begins cutting rates in response to deteriorating conditions. The un-inversion is not an "all clear" — it is typically the final warning.

What the Yield Curve Cannot Tell Us

For all its predictive power, the yield curve has significant blind spots. It measures expectations about monetary policy and credit conditions, not external shocks. The 1973 oil embargo, which triggered a severe recession, was not predicted by the yield curve — the inversion followed rather than preceded the shock.

The current situation presents a similar challenge. The Hormuz crisis is a supply-side shock that operates through a different transmission mechanism than the monetary tightening cycles the yield curve is designed to detect. The 18% probability from the Estrella-Mishkin model likely understates the true recession risk because the model's training data does not include scenarios where a geopolitical event removes 20% of global oil supply.

This is why GeoWire's composite model incorporates five additional indicators beyond the yield curve: the Sahm Rule (labor market deterioration), Hamilton's NOPI model (oil shock detection), the Philadelphia Fed Leading Index, the credit spread, and a geopolitical risk adjustment. The composite probability of 20.4% is higher than the yield curve alone would suggest, and could rise significantly if the crisis persists.

What to Watch: Key Thresholds Ahead

Several yield curve signals bear watching in the weeks ahead. First, the slope of the un-inversion: if the T10Y3M spread begins moving toward zero rapidly, it could signal that the Fed is preparing emergency rate action — historically a bearish signal despite the intuitive appeal of "normalization."

Second, the 10-year minus 2-year spread (T10Y2Y), which has been negative since mid-2025, is approaching the -0.50% level that has preceded the last three recessions with a 100% hit rate. A sustained move below -0.50% would push the NY Fed model's implied probability above 25%.

Third, credit spreads — specifically the ICE BofA High Yield spread — are widening. When this spread exceeds 500 basis points, it has historically coincided with or preceded recession in 85% of cases. The current spread is approximately 420 basis points, elevated but not yet at crisis levels.

The yield curve is not a crystal ball. It is a barometer — one that has proven remarkably reliable over seven decades. Its current reading is cautionary but not yet alarming in isolation. Combined with the oil shock transmission and labor market signals, however, the overall picture warrants significantly more concern than any single indicator suggests alone.