US Recession Risk Composite
Four signals, one per channel (labor, rates, credit, activity), each on or off against a published threshold. The score is the count that are on. The table below shows each signal's own track record against NBER recession dates.
Composite History
Count of triggered signals over time. Orange = 2 signals (elevated), red = 3+ (high). Grey bars mark NBER-dated recessions. Three of the four signals are coincident, so the count mostly confirms recessions as they start; only the yield curve has fired well ahead of them.
Track record
Methodology
This is one of two recession reads on the site. This page counts threshold triggers. The Cycle Risk Composite scores five signals as percentiles of their own history and averages them. A signal just short of its trigger counts for nothing here and for a lot there, so the two can disagree near thresholds.
- Labor: Sahm rule (Sahm, 2019; FRED SAHMREALTIME)
- 3-month average unemployment rate minus its low over the prior 12 months, using first-release data. Trigger at 0.50pp. It replaces three separate labor triggers (Sahm, 6-month unemployment change, payroll pace) that moved together and made one labor wobble look like three signals.
- Rates: 10Y−3M curve (Estrella & Mishkin 1996)
- Monthly average spread; trigger when negative. It inverted before every recession since 1968, with lags of roughly 6 to 18 months, and it also inverted from late 2022 into 2025 without a recession.
- Credit: Baa−10Y spread (Moody's via FRED BAA10Y)
- Trigger at 3.00%. Used instead of the ICE BofA high-yield OAS because FRED now publishes only about three years of the ICE series.
- Activity: CFNAI 3-month average (Chicago Fed)
- Weighted average of 85 monthly indicators, 0 = trend growth. Trigger below −0.70, the Chicago Fed's own rule of thumb for an increasing likelihood that a recession has begun.
What this is not: a recession forecast. It's a scoreboard of widely-watched signals. When multiple trip simultaneously, the historical record says recession probability rises — but the composite is not calibrated as a probability model. Read it as "how many canaries are coughing," not "P(recession) = X%."