Not in a recession now, and the early-warning signs are calm.
The same odds every month since 1970. Grey bands are recessions.
The 2022-25 hump was a false alarm. The yield curve stayed inverted for a record two years, hiring cooled, factories slumped and the classic Sahm rule triggered, which is what every past recession looked like from the inside. The economy slowed and kept growing. A high reading means danger, not certainty.
How this is counted
Seven inputs (below) each get a 0-100 rank against their own history, and the overall score is their average: today 47/100 (right now 44, early warning 52). We replayed every month from 1970 to 2025 using only the data available at the time. Of the months that scored near today's, 5% had a recession underway or starting within 12 months (likely range 2-11%). That history holds 8 recessions, so treat the number as rough.
The yield curve alone, since 1960: when it was positive, 6%; positive within a year of an inversion ending, 19%; inverted, 47% of months were followed by a recession within 12 months. Today it is positive.
Recession dates are the official NBER ones. FRED serves revised data rather than what was known at the time, which flatters the replay a little.
Is a recession underway?
Jobs and output, as they stand today.
Sahm rule (job losers)+0.00pp · Sep 2026GREEN
Sahm-style rise in the unemployment rate counting only people who lost a job (FRED U2RATE): 3-mo avg 1.93% vs 12-mo low 1.93%. Job losers, not all unemployed, because new and returning job seekers can push the headline rate up without layoffs (that's what tripped the classic Sahm rule in mid-2024). Ranked better than the headline version before 2020 too. Warning at the 70th percentile, red at the 90th. Scores by rank: more stressed than 39% of readings since 1967. scored
Initial claims198k (+0%) · Oct 17GREEN
4-week average vs 52-week low of 198k. Warning at +10%, red at +20%. Scores by rank: more stressed than 7% of readings since 1967. scored
Payroll growth+51k/mo · Sep 2026GREEN
3-month average monthly change. Warning below +50k, red below zero. Scores by rank: more stressed than 76% of readings since 1946. scored
Industrial production+2.2% YoY · Sep 2026GREEN
Warning below 0%, red below -1%. Can false-positive in manufacturing-only slowdowns. Scores by rank: more stressed than 56% of readings since 1947. scored
Early-warning signs
Measures that tend to turn 6 to 18 months before a recession. The yield curve has the best record of the three.
Yield curve statenormal · +1.05% · Sep 2026GREEN
Positive in Sep 2026, with no sustained inversion ending in the past 12 months. The last sustained inversion ended in Nov 2024, 22 months ago. Since 1960, 6% [4-8%] of such months were followed by a recession start within 12 months. Scores by rank: more stressed than 59% of readings since 1953. Monthly 10-year minus 3-month bill (FRED GS10 minus TB3MS), the series the rates were counted on, complete months only, so one daily print can't move it. scored
Credit spread (Baa minus 10y)1.48% (29th pct) · Sep 2026GREEN
Moody's Baa corporate yield minus the 10-year Treasury, monthly, ranked against its history since 1953 (warning at the 70th percentile, red at the 90th). One blind spot to carry from the correction page: private credit can now hide stress that used to show up in public spreads. scored
Building permits-3.2% YoY · Sep 2026GREEN
Warning below -5%, red below -15%. Permits lead starts. Scores by rank: more stressed than 67% of readings since 1961. scored
Watched, not scored
Shown for context. Their history is too short to count odds on, or they carried no signal when we replayed the past.
Real retail sales+2.5% YoY · Sep 2026GREEN
Warning below 0%, red below -2%. Watched, not scored: its history only starts in 1992, too short to count recession odds on. Scores by rank: more stressed than 44% of readings since 1993. context
Indeed job postings+3.1% YoY · Oct 23GREEN
Daily labor-demand index, posted weekly. Warning below -10%, red below -20% (loose on purpose, prone to false positives). Watched, not scored: history only starts in 2020, and in the replay it carried no signal (it ranked recession months no better than a coin flip). Scores by rank: more stressed than 28% of readings since 2021. context
VIX (21-day avg)16.7 · Oct 27GREEN
Warning at 25+, red at 35+. Sustained, not single-day, levels. Watched, not scored: it starts in 1990 and in the replay it gave no early warning of a recession start. Scores by rank: more stressed than 43% of readings since 1990. context
Benchmark · Chauvet-Piger model0.6% as of 2026-08-01
Published Markov-switching benchmark. Lags ~3 months; >80% historically confirms recession. Not in the composite.
Can this be backtested? (read this before acting)
Honestly, not in the usual sense. There have been ~7 US recessions since 1970. Every recession model ever built, Sahm, Estrella, Chauvet-Piger, the Conference Board LEI, is fit to single-digit effective N. IC math doesn't apply and walk-forward barely does. So instead of pretending, this page does two honest things: it counts (the curve-state rates above are actual historical recession-within-12-months frequencies, with deliberately wide intervals), and it leans on components with published real-time records.
The Sahm rule has caught every recession since 1970 in real time (one false trigger, in mid-2024, from new job seekers rather than layoffs, which is why this page uses the job-loser version); initial claims and payroll momentum are the standard coincident core; the yield curve and credit spreads are the canonical forward stack. The composite inherits whatever validity those pieces have, no more, no less. It cannot see shocks (a COVID-style hit moves everything in weeks), and it shares the correction page's blind spot: private credit can now hide funding stress that used to show up in public credit spreads.
Read the cycle state and the two bucket scores, not the single overall number. The most useful configuration this page can flag is “forward elevated, nowcast still clean”: late-cycle deterioration. Today it reads expansion: neither bucket is flagging.
Each scored input ranks today against its own history: the share of past readings calmer than now, so the score moves a little with every print. The seven scored inputs count equally (equal weights tested slightly better than hand-picked ones). The green and red lines on each row set its colour, not its score. The yield-curve row ranks the spread and, after a sustained inversion ends, holds at least the counted post-inversion danger, fading month by month; it reads complete months only, so one daily print can't move it.
Track record
4 weekly readings logged since 2026-10-10. Graded against an NBER recession beginning within 12 months. A reading resolves 12 months on, and NBER dates recessions months after they start, so the first grade is in late 2027 at the earliest.