Where the US cycle stands
Reading the data…
What changed in the last 3 months
What would change the call
Scorecard
Six composites, each scored 0–100 as a percentile of its own last 20 years (100 = most risk). Past readings use only data that had been published by that date.
Recession odds, three ways
A leading model (yield curve, 12 months ahead), a coincident rule (Sahm) and the composite. They disagree often; when all three agree, the signal has been reliable.
Methodology & sources
Regime. Growth = average robust z-score (median/MAD, trailing 10 years) of the 6-month annualized change in industrial production, payrolls and real retail sales. Inflation = the same for core CPI. The four quadrants are Goldilocks (growth up, inflation down), Reflation (both up), Stagflation (growth down, inflation up) and Disinflation (both down). The headline is a 3-month majority vote; conviction is the distance of the latest month to the nearest quadrant boundary.
Composites. Each signal is scored as its percentile within its own trailing 20 years and the composite is a weighted mean. Every page's "How this score is built" panel lists the signals, weights and transforms.
Dials. Growth uses the regime growth z and the Chicago Fed National Activity Index (3-month average; 0 = trend growth). Policy is the real fed funds rate (effective rate minus core PCE inflation, year over year) against its 20-year range. Financial conditions is the Chicago Fed NFCI (negative = looser than average).
Recession odds. Curve model: P = Φ(−0.533 − 0.633 × monthly-average 10Y−3M spread), the Estrella–Trubin specification behind the NY Fed's published probability. Sahm rule: FRED SAHMREALTIME.
Limits. FRED serves the latest revised data, so history looks cleaner than it did in real time. Weights are judgement, not fitted: four recessions are too few to fit weights without overfitting.