Key takeaways
- Nonfarm payrolls rose 178,000 in March 2026; BLS said the unemployment rate "changed little at 4.3 percent", down from 4.4%.
- The civilian labour force shrank by 396,000 in March, from 170,483,000 to 170,087,000 — and household employment fell 64,000.
- The participation rate fell to 61.9% and the employment-population ratio to 59.2% — both lower than February.
- Average hourly earnings rose 0.2% on the month and 3.5% over the year, the smallest annual increase since May 2021.
- Between January and June 2026 — all months on the same population controls — the labour force fell by 1,107,000.
The unemployment rate fell, and 396,000 people left the labour force
The BLS Employment Situation released on 3 April 2026 reported payroll growth of 178,000 for March and an unemployment rate of 4.3%, down from 4.4%. BLS itself was more careful than the headlines were: the release says the unemployment rate "changed little at 4.3 percent" and that the number of unemployed, at 7.2 million, also changed little. A one-tenth move is inside the noise, and the agency said so.
The household survey says something different. The civilian labour force fell from 170,483,000 in February to 170,087,000 in March — a decline of 396,000. The participation rate fell from 62.0% to 61.9%. The employment-population ratio fell from 59.3% to 59.2%. The number of unemployed fell by 332,000. And household employment fell too, by 64,000, from 162,912,000 to 162,848,000.
An unemployment rate is a ratio. It falls when the numerator shrinks — because people found work, or because they stopped looking. In March the labour force fell faster than the count of unemployed did — and employment fell as well. Nobody was absorbed into work. The denominator simply got smaller. That is a different economic event with a different signal, and a comfortable headline does not change that.
The December comparison everyone is making is invalid
There is a trap in this series. BLS introduced updated population controls with the February 2026 release and revised January 2026 to match. December 2025 and earlier were not revised. The agency's own note quantifies the effect: the adjustment "decreased both the total civilian labor force and the number of employed people by 1.4 million each" — 1,417,000 off the labour force line — and lowered the participation rate by 0.4 percentage point. That is a Census re-benchmark, not people leaving work.
So the roughly 1,500,000 decline you get by comparing December 2025 with the spring of 2026 is mostly an artifact, and BLS warns that "these annual population adjustments can affect the comparability of household data series over time." The comparison that survives is January to June 2026, six months all on the new controls. On that basis the labour force fell from 170,465,000 to 169,358,000 — down 1,107,000. Smaller than the artifact, and real.
The chart plots those six months. Participation across the same window never rose once: 62.1%, 62.0%, 61.9%, 61.8%, 61.8%, 61.5%. A 4.3% unemployment rate produced by 61.5% participation is not the same labour market as a 4.3% rate produced by 62.1% participation.
Wage growth at 3.5% is the confirming evidence, not the contradiction
Average hourly earnings for all private employees rose 0.2% in March and 3.5% over the 12 months — the lowest annual increase since May 2021. If the labour market were tightening, wage growth would be the first place it showed. It is doing the opposite.
The two facts fit one story: labour demand is soft enough that fewer people find it worth searching, and soft enough that employers need not bid up pay. They do not fit the story a 4.3% headline rate tells on its own.
What this is worth in a portfolio
The practical use is as a filter on rate expectations. A falling unemployment rate ordinarily argues against cuts. One driven by exit does not — it is a demand-weakness signal wearing a strength signal's clothes. Duration positioning that keys off the headline rate is keyed to the wrong variable.
A workable rule: treat any decline in the unemployment rate as uninformative unless the labour force level rose in the same month. March 2026 fails that test (-396,000). April 2026 fails it (-92,000).
May passed the test. June took it back.
May 2026 passed. The civilian labour force rose 83,000 to 170,078,000, household employment rose 149,000, participation held at 61.8%, and the unemployment rate held at 4.3%.
June reversed it. The labour force fell 720,000 to 169,358,000 — the steepest one-month drop since the re-benchmark — participation fell from 61.8% to 61.5%, household employment fell 507,000, and the unemployment rate fell again, to 4.2%. That is the March mechanism at nearly twice the scale. The lowest unemployment rate of 2026 sits on top of the smallest labour force of 2026, and the two facts are the same fact.
What would falsify this and what the data cannot say
Labour-force exit is not automatically bearish. If it is demographic — retirements from an ageing cohort — it is a supply-side fact with no cyclical content, and the unemployment rate is telling the truth. The Employment Situation does not decompose the decline by reason, so this reading cannot be confirmed from the release alone. The number that would settle it is the count of people not in the labour force who want a job, and the flow from employment into non-participation.
There is also a vintage problem. The 178,000 payroll figure and the 3.5% wage figure are as first published on 3 April 2026. FRED's current vintage of the same series shows a March payroll change of 214,000 and annual wage growth of 3.4%, after revisions. The direction of the argument does not change; the decimals do. A first print quoted as final is a number that no longer exists.
The series to read before the headline
The payroll number will keep being the headline, and it will keep being revised. The series that carries the signal — labour force level, participation rate, employment-population ratio — ships in the same release, on the same day, and is almost never quoted. May showed what a clean print looks like: the level up, the rate flat. It lasted one month. Whether July restores it or confirms June is the question the next release answers. and .
This content is for informational purposes only and does not constitute financial advice. Always do your own research or consult a qualified financial adviser before making investment decisions.