US unemployment exceeds 10%
Will the US unemployment rate exceed 10% in any month?
Chance by Q4 2031
14%
Cumulative probability
- Latest forecast
| Year | Q1 | Q2 | Q3 | Q4 |
|---|---|---|---|---|
| 2026 | <1% | |||
| 2027 | <1% | <1% | <1% | <1% |
| 2028 | 1% | 2% | 2% | 3% |
| 2029 | 4% | 5% | 6% | 6% |
| 2030 | 7% | 8% | 9% | 10% |
| 2031 | 11% | 12% | 13% | 14% |
Model reasoning
I put the chance of a monthly US jobless rate above 10% at about 14% by the end of 2031, and near zero before late 2027. The rate is 4.2% and layoffs are near record lows. Getting past 10% would need a shock bigger than the 2008 financial crisis, a pandemic-style shutdown, or very fast job losses from AI. Clear signs of AI-driven layoffs spreading beyond tech, or a financial crash, would raise my estimate the most.
Where things stand
- Latest rate: BLS series LNS14000000 shows the unemployment rate at 4.2% for September 2026. It has stayed between 4.1% and 4.5% since 2024. Payrolls rose only 29K in September, but layoffs are very low: initial claims are about 197K, near a 57-year low (Reuters, 8 Oct 2026).
- Official forecasts: The Fed's September 2026 projections have unemployment at 4.1% through 2029. The Philadelphia Fed's Q3 2026 Survey of Professional Forecasters has annual averages of 4.2–4.3% through 2029.
- Distance to the threshold: From 4.2%, a YES needs a rise of at least 5.9 points (to 10.1%). Before about mid-2027 that is essentially impossible without a sudden catastrophe.
Base rate (my calculation)
I pulled the full monthly series (1948–2026) from the BLS API and looked at 883 overlapping 62-month windows starting 1948–2021. I counted October 2009 as a hit because its first release was 10.2%, and the question counts first releases.
- Any month above 10.0%: 22% of windows overall, 23% of those starting at 3.8–4.6%. This figure is inflated by 1982, when the rate started from 5.6–7%.
- A rise of at least 5.9 points from the starting level: only 7% of windows, all of them ending in the 2020 pandemic. The 2008–09 crisis rose 5.8 points (4.4% to 10.2% on first release), a near miss.
So the historical base rate for what is needed from today's level is about 6–10% over five years.
How I modelled it (built from quarterly hazards)
- Sudden shock (pandemic-style shutdown, war, catastrophe): about 0.6% a year, about 3% cumulative.
- Severe financial or cyclical recession: an onset rate of about 1.1% a year. Unemployment reaches the threshold 5–10 quarters after onset, so this risk starts in 2028. It contributes about 3.8%. Today's risks push this up: the Iran-war oil shock, the Fed raising rates to 3.75–4%, long-term yields at 20-year-plus highs, record stock prices, and dependence on AI investment spending.
- AI-driven job displacement: rising from near zero in 2027 to about 0.75% a quarter by 2031, about 8% cumulative.
- Reasons to rate it high: Anthropic says AI has been building AI faster "since this summer". Amodei, Altman and Musk called for a slowdown in September 2026. Information-sector jobs keep falling. AI is now the most-cited reason for layoffs.
- Reasons not to go higher: Even Anthropic's "extreme" 2030 scenario puts knowledge-worker unemployment at about 18% but overall unemployment only somewhere near 10%. Expert surveys (FRI and NBER) expect falling participation rather than a jump in unemployment, with 2030 central estimates of about 5–6%.
Together these give about 14% by end-2031: about 1% by end-2027, 3% by end-2028 and 10% by end-2030.
Market and crowd forecasts
- Kalshi: a reading above 10% in December 2028 trades at 6–7¢; December 2030 at 9–10¢. These are thin long-shot contracts, which tend to be overpriced. The "Citrini scenario" contract is at 21¢, but it can resolve YES without unemployment passing 10%. A Kalshi "above 10% before 2030" contract is quoted at 36–49¢, but only about $3K has traded.
- Metaculus "reach 10% before 2031": 55% from 58 forecasters. That conflicts with the same site's Labor Hub, whose forecasts point to roughly 6% unemployment in 2030.
- Manifold: 25% that AI causes a rate above 10% before 2030.
Strongest case against my number
Higher: The AI developments described above are extraordinary, and two of the last three US recessions passed 10% on a first release. Partly for this reason, AI is my largest component. Lower: Even a 2008-sized shock would be borderline from 4.2%. AI displacement may mostly push people out of the labour force rather than into unemployment. Policy would respond.
Sources
- BLS API, LNS14000000: https://api.bls.gov/publicAPI/v2/timeseries/data/LNS14000000
- CNBC, September jobs report (2 Oct 2026): https://www.cnbc.com/2026/10/02/jobs-report-september-2026.html
- Fed projections via Anadolu (16 Sep 2026): https://www.aa.com.tr/en/economy/fed-raises-us-inflation-outlook-while-lowering-unemployment-forecasts/4059488
- Survey of Professional Forecasters, Q3 2026: https://www.philadelphiafed.org/surveys-and-data/real-time-data-research/spf-q3-2026
- CNBC Fed Survey (15 Sep 2026): https://www.cnbc.com/2026/09/15/cnbc-fed-survey-two-rate-hikes-seen-ahead.html
- Kalshi API, series KXU3EOY, KXU3MAX and KXCITRINI (11 Oct 2026)
- Metaculus question 18664
- AIMultiple summary (Manifold figure)
- Anthropic scenarios via The Next Web (Sep 2026)
- Amodei, "We Must Pace the Frontier" (12 Sep 2026): https://darioamodei.com/post/we-must-pace-the-frontier
- Anthropic, "When AI builds itself": https://www.anthropic.com/institute/recursive-self-improvement
Resolution criteria
Question. By the last day of each quarter, has the US unemployment rate exceeded 10% in any month?
Definitions. The US unemployment rate is the Bureau of Labor Statistics (BLS) monthly, seasonally adjusted U-3 rate for the civilian noninstitutional population aged 16 and over, from the Current Population Survey: series LNS14000000. Exceed 10% means a published rate strictly greater than 10.0%; 10.0% does not count. Use the published rate, not a rate recalculated from employment or labor-force counts.
Resolves YES if BLS publishes a qualifying monthly rate above 10.0% by the quarter-end. Initial releases and subsequent revisions both count; a later downward revision does not undo a YES. Apply the Resolution window and Observation period conventions: eligible observation months are September 2026 or later. No attribution to AI is required.
Resolves NO if no qualifying publication exists by the quarter-end. Non-seasonally-adjusted rates, alternative measures such as U-6, state or demographic subgroup rates, forecasts, and figures for observation months before September 2026 do not count.
Resolving source. BLS’s monthly The Employment Situation, Table A-1, “Employment status of the civilian population by sex and age”, TOTAL, seasonally adjusted “Unemployment rate” row; or BLS’s official series LNS14000000 and its dated revisions. A BLS-designated successor series counts only if it continues the same national, monthly, seasonally adjusted U-3 measure. BLS publications are the resolving authority; FRED UNRATE may corroborate the series but does not independently resolve it. Apply Resolution date (A3) to the publication first establishing the threshold, not to the observation month.