Global Recession Probability Dashboard
12-month-ahead recession probability — NY Fed Estrella-Mishkin probit applied to nine sovereign yield curves.
Data as of 2026-05-17T01:30:39ZWhat does "12-month recession probability" actually mean?
The number you see in each tile is an estimate of the probability that a country enters recession within the next twelve months. A reading of 5% means the model assigns roughly a one-in-twenty chance to a downturn starting in the year ahead; 40% means roughly four-in-ten, which historically has been a near-certain signal in the US. The number is not a forecast you would bet on dollar-for-dollar — it is a calibrated reading from a statistical model. The signal that matters most is the change over time and the level relative to historical thresholds.
The probability comes from one input only: the yield curve. Specifically, the model uses the gap between the 10-year government bond yield and the 3-month treasury bill — the so-called 10s2s or, more precisely here, the 3m10y spread. To make this concrete: if the 10-year US Treasury yields 4.10% and the 3-month bill yields 5.25%, the 3m10y spread is 4.10% − 5.25% = −115 basis points (a deep inversion). Feeding that into the formula gives a recession probability around 55%. By contrast a +200 bp upward-sloping curve (10-year at 4.50%, 3-month at 2.50%) maps to a probability near 4% — essentially the bond market saying everything looks fine.
Who watches this signal? Just about everyone: the New York Fed publishes the headline number monthly, bond traders use it to position duration, the Wall Street Journal and Financial Times quote it in recession debates, corporate treasurers consult it when sizing inventory or capex plans, and economists cite it because every US recession since 1969 has been preceded by yield curve inversion. It is not infallible, but it is one of the most consistently useful single indicators in macro.
Estrella-Mishkin probit: calibration, sample bias, and signal interpretation
Estrella and Mishkin (1998, REStat) estimate a probit of the form P(NBER recession in t+12) = Φ(α + β·spreadt) on monthly US data, 1959-1995, obtaining α = −0.5333, β = −0.6629 with pseudo-R² of roughly 0.30 and an AUC near 0.92 in-sample. The model is unconditionally simple by design — adding the Fed funds rate, lagged GDP, or stock returns improves in-sample fit only marginally and tends to overfit out-of-sample (Estrella-Trubin, 2006). The dependent variable's recession dating is NBER business-cycle committee output: peak-to-trough US calls, which embed America-specific labour-market and sectoral dynamics that do not translate one-for-one to other economies.
Sample-period bias is the most important caveat for current use. The 1959-1995 window is dominated by demand-driven cycles in a high-real-rate world (average 10Y real yield ~2.8%); the post-2008 sample features structurally lower real rates, persistent term-premium compression (Adrian-Crump-Moench), and central-bank balance-sheet expansion that mechanically depresses the long end without signalling weak nominal growth. Bauer-Rudebusch (2020) document that controlling for term premium roughly halves the marginal predictive value of the raw 3m10y spread post-2000. The NY Fed continues to publish the original calibration unchanged, which is why the headline reading should always be cross-checked against term-premium-adjusted variants.
For non-US economies the level is indicative, not literal. Local probit re-estimation (Moneta 2005 for the euro area; Chinn-Kucko 2015 for an OECD panel; Hasse-Lajaunie 2022 for emerging markets) consistently finds attenuated slope coefficients and lower AUC than the US benchmark, with the euro area's coefficient roughly two-thirds the US value. Applying the original US-calibrated coefficients to Japan, China, or India systematically over-reads the signal in administered or capital-controlled curves. Best practice is to use the US calibration consistently as a cross-country comparator (rank order) rather than as nine independent probability statements.
Alternative and complementary signals. The Sahm rule (Sahm 2019) triggers when the three-month moving average of US unemployment rises 0.5pp above its trailing 12-month minimum; it is faster than the yield curve but only fires at or after recession onset. The Conference Board LEI, initial jobless claims (four-week MA breakouts above ~310k historically precede US recessions by 2-4 months), and ISM Manufacturing New Orders < 45 also serve as cross-checks. Threshold interpretation: in the US post-war sample, a 3m10y inversion of any depth has preceded every recession with a 6-24 month lead; the NY Fed model's 30% threshold has been crossed before every recession since 1969 with no false positives in the post-1985 sample save 1998 (a near-miss).
Current ranking
12-month history — all 9 countries
Dashed reference line at 30% marks the level historically associated with elevated US recession risk in the NY Fed publication.Detail table
| Country | Currency | Central bank | 3m10y spread | Recession prob (12m) | Status |
|---|---|---|---|---|---|
| Canada | CAD | Bank of Canada | +0.33pp | 22.6% | Low |
| Australia | AUD | Reserve Bank of Australia | +0.38pp | 21.6% | Low |
| United Kingdom | GBP | Bank of England | +0.60pp | 17.6% | Low |
| China | CNY | People's Bank of China | +0.68pp | 16.3% | Low |
| Switzerland | CHF | Swiss National Bank | +0.87pp | 13.3% | Low |
| United States | USD | Federal Reserve | +0.90pp | 12.9% | Low |
| Eurozone | EUR | European Central Bank | +0.94pp | 12.4% | Low |
| India | INR | Reserve Bank of India | +1.13pp | 10.0% | Low |
| Japan | JPY | Bank of Japan | +1.66pp | 5.1% | Low |
About the model
The probit recession-probability model was introduced by Arturo Estrella and Frederic Mishkin in their 1998 paper "Predicting U.S. Recessions: Financial Variables as Leading Indicators" (Review of Economics and Statistics). The Federal Reserve Bank of New York publishes a monthly series based on this model — the version cited universally by economists, fund managers, and the financial press.
Functional form:
P(recession in 12m) = Φ( -0.5333 + -0.6629 × spread3m10y )
where Φ is the standard normal cumulative distribution function and the spread is in percentage points (e.g. -0.50 for a 50-bp inversion).
Why 3m10y? The 3-month bill closely tracks the current policy rate, so the 3m10y spread directly captures the gap between current policy and the long-run growth/inflation expectation embedded in the 10-year yield. When this spread inverts, markets are saying that current policy is sufficiently restrictive that long-run nominal returns are expected to fall — which historically is associated with central banks ultimately cutting in response to a downturn.
Cross-country caveats. The model coefficients are calibrated on US post-WWII data. Applying them to Japan (chronic low inflation, post-YCC dynamics), China (state-influenced bid for sovereign duration, capital controls), and India (higher inflation target, different demographic and growth regime) produces probabilities that should be read as shape signals rather than calibrated forecasts. For developed-market peers (Eurozone, UK, Australia, Canada, Switzerland) the cross-country mapping is more defensible.
Curve fitting: Each country's term structure is also fit to a Nelson-Siegel-Svensson model. See the NSS methodology page and the per-country yield-curve pages for the parametric fits.