takeoff.watch

US labour-force participation falls below 55%

Will BLS report a US labour-force participation rate below 55% for any month?

Chance by Q4 2031

2%

Cumulative probability

0%25%50%75%100%
  • Latest forecast

Forecast history

Forecast made on 13 Oct 2026

Latest of 5 forecasts · Methodology v2.1

Model reasoning

Forecast made 13 Oct 2026 · Methodology v2.1

I put the chance of YES at about 2% by the end of 2031. The September 2026 participation rate was 61.8%. Getting below 55% means a drop more than twice the largest five-year fall in 78 years of records, ending 3 points under the all-time low of 58.1%. Aging alone only gets the rate to about 60.5% by 2031. The development that would most raise this forecast is evidence that AI is pushing millions of people out of the workforce.

Where things stand

  • Latest reading. BLS's Employment Situation for September 2026 (released 2 October 2026) puts the seasonally adjusted participation rate (LNS11300000) at 61.8%. That is the first eligible month. The October figure comes out on 6 November 2026. To resolve YES, a published value would have to be 54.9% or lower, a drop of 6.9 points (about 19 million people leaving the labour force, roughly 11% of it).
  • 2026 so far. The rate fell from 62.4% in December 2025 to 61.4% in July 2026, then recovered to 61.8%. The St. Louis Fed (4 August 2026) attributes most of the first-half fall to two things. One is an unusually large January population-control revision, worth about −0.35 to −0.4 points. The other is aging, worth about 0.02 points a month. A one-month drop in June among people aged 25–54 largely undid an earlier rise in that group.
  • AI so far. The Yale Budget Lab's September 2026 update, which uses CPS data through August, still finds no clear AI-driven labour-market disruption. Sam Altman has publicly conceded that AI is spreading through the economy more slowly than he expected.

Reference class and base rate

I downloaded the FRED CIVPART series, which mirrors LNS11300000, for 1948–2026 (https://fred.stlouisfed.org/graph/fredgraph.csv?id=CIVPART) and computed:

  • The all-time low is 58.1% (December 1954). The COVID low was 60.1% in April 2020.
  • The largest fall within any window of up to 62 months is 3.2–3.3 points: February to April 2020, and 2008 to 2013.

So YES needs a fall about 2.1 times the worst on record, to a level 3 points below anything seen in 78 years. Out of about 900 historical windows, none comes close, which puts the pure base rate well under 1%.

Trend and shocks

  • Trend. BLS's 2025–35 projections (Table 3.3, updated 1 September 2026) have the rate going from 62.4% in 2025 to 61.1% in 2035. If aging continues at about 0.02 points a month, the rate would be about 60.6% by November 2031. That still leaves a gap of about 5.7 points to the threshold.
  • Recessions. A typical recession takes off another 1–2 points. Even a COVID-sized shock (−3.2 points) landing at the end of the window would leave the rate around 57–58%.
  • High unemployment is not the same thing. Laid-off people who keep looking for work still count as being in the labour force. In 2020, unemployment of 14.8% went with only a 3.2-point fall in participation.

Paths to YES (rough contributions by end-2031)

  1. Extreme AI displacement (about 1–1.5%). This needs very fast deployment that pushes tens of millions of people out of the labour force, through early retirement, discouragement, or income support like a universal basic income, by 2031. The current evidence and the diffusion lags make this a tail scenario.
  2. A pandemic far worse than COVID (about 0.3%).
  3. A depression-scale downturn on top of aging (about 0.2%).
  4. War or other catastrophe where BLS still publishes, a method change, or a publishing error (about 0.2%). Population-control revisions move the rate by tenths of a point, not whole points.

Cross-check against Metaculus

  • Metaculus gives about 5–6% that the employment-population ratio falls below its COVID low of 51.3% in 2030. That is an easier bar than this one, because high unemployment alone can push that ratio down while participation stays near 57–58%.
  • A Metaculus question on the 2030 participation rate shows a median of 58.7%. It has few forecasters and appears anchored to an old BLS projection, so I gave it little weight.

Quarterly risk and resulting series

The chance of YES first happening in a given quarter starts at about 0.01% and rises steadily, by about 0.01 points per quarter, to about 0.2%. It rises because the gap narrows with aging and because AI risk grows later in the window. Compounded, the cumulative probability is about 0.14% by end-2027, 0.9% by end-2029, and about 2.2% by end-2031. The near quarters are essentially zero: the October and November 2026 figures would need a single shock twice the size of COVID's.

Strongest case for a higher number: if a fast AI takeoff arrives around 2028–29 together with income support, millions could leave the labour force, as the pessimistic Metaculus figures hint. Strongest case for a lower number: the rate is slow to move and AI shows no measurable effect yet, so 1% would also be defensible. I settled at about 2%.

Sources

Resolution criteria

Opened 16 Sep 2026Not yet resolved

Question. By the last day of each quarter, has BLS reported a US labour-force participation rate below 55% for any month?

Definitions.

  • Labour-force participation rate: the nationwide, seasonally adjusted Current Population Survey (CPS) series LNS11300000, covering the civilian noninstitutional population aged 16 and over. It is the civilian labour force (employed plus unemployed) as a percentage of that population.
  • Below 55%: a published rate strictly less than 55.0%; at the current one-decimal precision, 54.9% or lower. Use the published rate, not a rate independently calculated from population and labour-force counts.
  • Eligible month: September 2026 or later, under the Observation period convention.

Resolves YES if BLS publishes a qualifying monthly value before or on the quarter-end. An initial estimate or a subsequently revised value counts. A revision first taking an eligible month's rate below the threshold qualifies on the revision's publication date. Later revisions do not undo a YES. No attribution to AI is required.

Resolves NO if no qualifying value has been published by the quarter-end. Unadjusted rates, subgroup or state rates, employment-population ratios, annual averages, forecasts, and revisions solely to months before September 2026 do not count. A value of exactly 55.0% does not count. Missing or delayed data do not establish a threshold crossing.

Resolving source. BLS publications only: the CPS series LNS11300000 database, including official BLS downloads or API releases, or The Employment Situation, Table A-1, “Employment status of the civilian population by sex and age”, TOTAL Participation rate, seasonally adjusted columns, including archived releases. A BLS-designated replacement series counts only if it retains the same national population coverage and seasonally adjusted monthly participation-rate concept. Apply the Resolution window and Observation period conventions, and publication-date rule A3; the resolving-source restriction is narrower than Public confirmation (A2).

Download all forecasts (JSON)