US Payrolls Fell 23,000 (August 2026): 103K Cut From May and June — and Why the Dollar Broke 100 as Unemployment Fell to 4.1%
US payrolls fell 23,000 in July against +83,000 expected, and May and June were cut by a combined 103,000 — yet unemployment fell to 4.1%. The mechanism, factor by factor.
US Payrolls Fell 23,000 (August 2026): 103K Cut From May and June — and Why the Dollar Broke 100 as Unemployment Fell to 4.1%
The July employment report landed on Friday 7 August at 8:30 a.m. Eastern, and it delivered the bottom row of the scenario map. Total nonfarm payroll employment fell by 23,000 against a Dow Jones consensus of +83,000. The revisions were worse than the headline: May was cut by 66,000, from +129,000 to +63,000, and June was cut by 37,000, from +57,000 to +20,000 — a combined 103,000 removed from two months that had already been read as weak. The three-month average is now roughly 20,000 jobs a month. And the unemployment rate went down, to 4.1% from 4.2%, which is the part of this report most likely to be misread. The market did not misread it. The 2-year Treasury yield fell about 8 basis points to 4.166% while the 10-year fell only 5.5 to 4.615%, and the dollar index dropped 0.43% to 99.498, breaking the 100 level it had been holding. That is the configuration this preview described in advance as the dollar-negative version: the front end takes weight off a rate increase — no cut was ever priced — while the long end barely moves.
The mechanism worth carrying out of this release is the one that explains why a falling unemployment rate is the least reassuring number in it. The household survey's improvement came from the denominator, not the numerator, and the report contains the evidence for that in the row directly beneath it.
- Payrolls fell 23,000 in July against a Dow Jones consensus of +83,000 — a miss of more than 100,000 on the headline alone.
- The revisions were the bigger number. May was cut 66,000 (129,000 → 63,000) and June cut 37,000 (57,000 → 20,000), a combined −103,000. Three-month average hiring is now about 20,000 a month against a prior-12-month average of 34,000.
- Unemployment fell to 4.1% from 4.2%, with unemployed people down to 6.9 million — but participation slipped to 61.4% and the employment-population ratio to 58.9%. Both falling together means people left the labour force rather than found work.
- The first crack in the low-fire story: people on temporary layoff rose 153,000 to 921,000, while permanent job losers were little changed at 1.7 million.
- Wages decelerated. Average hourly earnings rose 2 cents to $37.62, up 3.2% year over year, down from 3.5% in June — against June headline CPI of 3.5%, still a negative real wage.
- The front end led. The 2-year fell to 4.166% (−8bp) against the 10-year at 4.615% (−5.5bp), so 2s10s widened slightly to about 44.9 basis points.
- The dollar index fell 0.43% to 99.498, low of 99.422, breaking below 100. USD/JPY −0.83% to 157.10 led the majors; USD/CAD −0.41% to 1.3954; EUR/USD +0.36% to 1.1565.
- Canada printed the mirror image in the same minute: employment +75,000 and unemployment down to 6.4%, a two-year low.
- Two dates now matter more than usual: the preliminary benchmark revision on 28 August, and the August payrolls report on 4 September — the last one before the FOMC meets on 16 September.
- Interest-rate expectations and growth are two of the five factors that move a currency, and this print moved both. See how they are scoring the dollar right now on the live meter.
What actually happened
One release, two surveys, and this month they disagreed in the direction that flatters the data. The establishment survey said employment fell. The household survey said unemployment fell. Both are accurate descriptions of what they measure.
| Measure | July 2026 | Expected / prior | What changed |
|---|---|---|---|
| Nonfarm payrolls | −23,000 | +83,000 consensus; June was +57,000 as first reported | A miss of more than 100,000 against consensus |
| Revisions, May + June | −103,000 | May 129,000 → 63,000; June 57,000 → 20,000 | Three-month average cut to roughly 20,000 |
| Unemployment rate | 4.1% | 4.2% expected and prior | Fell; unemployed down to 6.9 million from 7.1 million |
| Participation rate | 61.4% | 61.5% in June | Down 0.7 percentage point since January |
| Employment-population ratio | 58.9% | 59.0% in June | Down 0.5 percentage point since January |
| Average hourly earnings | $37.62, +3.2% y/y | 3.5% y/y in June | Wage growth decelerated 0.3 percentage point |
| Temporary layoffs | 921,000 | +153,000 on the month | Permanent job losers little changed at 1.7 million |
The revision line deserves more attention than it will get. This preview flagged in advance that a large downward cluster would be the more consequential outcome, because it does not change one month's number — it changes the level from which every subsequent month is measured. June, described at the time as the weakest month in four at 57,000, is now 20,000. May, which looked respectable at 129,000, is 63,000. The trend traders believed they were in during May and June simply was not the trend. (Primary release: BLS, Employment Situation — July 2026; coverage: CNBC.)
Why the falling unemployment rate is the worst part
The unemployment rate is a ratio, and ratios can improve for bad reasons. It measures unemployed people as a share of the labour force — those working plus those actively looking. Someone who stops looking is not counted as unemployed at all; they leave the numerator and the denominator, and the rate falls.
The way to tell which kind of improvement you are looking at is the employment-population ratio, which has no such escape hatch: it measures employed people against the whole civilian population. If the jobless rate falls because people found work, that ratio rises. In July it fell, to 58.9% from 59.0%, while participation fell to 61.4%. Both moving down together settles it. The 200,000 decline in unemployed people did not go into jobs.
Wages tell the same story from the price side. Average hourly earnings rose 2 cents to $37.62 and 3.2% over the year, down from 3.5% in June. That is genuine wage disinflation — and against June headline CPI of 3.5%, it still leaves the average worker's real wage slightly negative. A labour market that is cooling on quantity and on price is a different object from the one described a week ago, when ADP reported that pay for job-changers had accelerated to 7%.
Where the jobs actually went
The composition explains most of the headline, and one line of it deserves a caution.
Local government education shed 50,000 jobs, after showing little net change over the prior twelve months. Education payrolls are among the most heavily seasonally adjusted series in the report, because school staffing swings enormously between the academic year and the summer; when the timing of that swing differs from the seasonal pattern the adjustment expects, the adjusted figure moves sharply without much changing on the ground. A 50,000 move in a series that had been flat for a year is the kind of print worth watching for reversal next month rather than extrapolating.
The rest is harder to explain away. Retail trade lost 19,000, with warehouse clubs, supercenters and other general merchandise retailers down 21,000 and gasoline stations down 5,000. Financial activities continued a longer decline, down 14,000 on losses in credit intermediation (−9,000) and insurance carriers (−7,000); that sector is now down 121,000 from a peak in May 2025.
And the engine kept running, but slower. Health care added 22,000, against a prior-twelve-month average of 36,000. This preview flagged the specific risk that the one sector carrying net hiring was also the one whose forward demand was cooling fastest — JOLTS had shown health care and social assistance openings falling 147,000 in June, the largest sector decline in that report. July is the month it showed up in the payroll data. Every other major industry showed little change.
What it did to the September question
The reaction function described in this post before the release is the reason the market response looked modest relative to the size of the miss. When a central bank is debating cuts, a report like this produces a violent front-end rally. This Fed was not debating cuts. Policy has sat at 3.50%–3.75% since December 2025, the July decision was a 9–3 hold with three officials dissenting in favour of a hike, and September pricing had run in a 55–65% hike band since that meeting.
So the repricing had a ceiling built into it. A negative payroll print can take weight off an increase; it cannot add weight to a cut nobody was pricing. The observable result was exactly that shape: the 2-year fell about 8 basis points to 4.166%, the 10-year fell 5.5 to 4.615%, and because the front end moved more, the 2s10s spread widened from roughly 42.5 to 44.9 basis points. The expected path flattened rather than collapsing.
That is why a report this weak moved the dollar less than a percent. It is also why the move it did produce was clean: a currency losing front-end support while the long end holds is losing on the rate factor without gaining anything on the growth story, and the dollar index went through 100 — a level it had been defending since the last week of July — to a low of 99.422.
The scenario that landed
The map published before the release had four rows. July delivered the fourth, and delivered the third column of it too.
| Row from the pre-print map | What it said would happen | What actually happened |
|---|---|---|
| Below ~30K, or negative revisions | Prices the hike out, into a 3.5% headline inflation backdrop | Both at once: −23,000 headline and −103,000 of revisions |
| Growth factor down; long end the variable | Front end down with the long end sticky is the dollar-negative version | 2-year −8bp vs 10-year −5.5bp; 2s10s widened to ~44.9bp |
| The participation cross-check | A rate that falls on shrinking participation is not reassurance | Rate fell to 4.1%; participation 61.4%, employment-population 58.9% |
| The wage cross-check | A hot wage print was the only clean hawkish combination | Wages decelerated to 3.2% from 3.5% — the opposite |
What the map deliberately did not contain was a claim about where the dollar would go, and that restraint is worth keeping now. The dollar fell on the day. Whether it keeps falling depends on a question this report cannot answer: whether the Fed reads a −23,000 print with decelerating wages as a growth problem that outranks its inflation concern, or as one month of noise inside a low-hire equilibrium that has not yet produced rising unemployment. Nothing in the July data settles that, and the committee has six weeks and two more major releases before it votes.
Where it landed across the majors
One release, eight different counter-currencies, and the sizes of the moves ranked in the order the rate-differential channel predicts.
| Pair | Move | Level | Why this size |
|---|---|---|---|
| USD/JPY | −0.83% | 157.10, from a 158.42 close | Most mechanically tied to the US front end; BoJ at 1.00% makes the gap the dominant input |
| USD/CAD | −0.41% | 1.3954, from 1.4012 | Two events in one minute — Canadian employment rose 75,000 and unemployment fell to 6.4% |
| EUR/USD | +0.36% | 1.1565, from 1.1524 | Muted by the euro's own live rate question after July inflation at 2.9% |
The yen leading is the cleanest confirmation of the mechanism. This preview named it the pair most exposed to a flattening US path, because a rate gap is the difference between two numbers and only one of them needed to move — and it moved from 158.41 at the open, the level the yen had backslid to after July's intervention, which we traced in the intervention-rally post.
The loonie is the one to read carefully, because two prints hit in the same minute and both pushed the same way. Statistics Canada reported employment up 75,000 in July with the unemployment rate down to 6.4%, its lowest since July 2024 and a third consecutive monthly decline — against a consensus near 5,000 that we set out in the Canada jobs preview. Gains came in wholesale and retail trade (+21,000), finance and insurance (+18,000), professional and scientific services (+17,000) and construction (+16,000), while Canadian wage growth decelerated to 2.8% from 3.3%. A USD/CAD move of 0.41% therefore contains a weak-dollar leg and a strong-loonie leg, and separating them is why the pair's factor scores matter more than its chart. The euro's smaller move reflects the mirror problem: EUR/USD is a contest between two live rate debates rather than a referendum on the dollar, as the July eurozone CPI read set out.
What would change the picture
Three dates, and the first of them is unusual.
On 28 August the BLS publishes the preliminary estimate of the annual benchmark revision to establishment survey data, which rebases payroll estimates against the Quarterly Census of Employment and Wages — comprehensive counts drawn from state unemployment-insurance tax records rather than a survey sample. In a year when the sample-based estimates have already been revised down repeatedly, that carries more weight than usual, and it lands nineteen days before the FOMC meets. Official estimates are not updated on the preliminary figure; the final revision arrives with the January 2027 report.
Then 4 September brings the August employment report — the last payroll print before the committee votes on 16 September. July CPI arrives in between, and given that the hike case has rested on energy rather than core, the headline-versus-core split there may still matter more for September than a single negative payroll month does. (Schedule: Federal Reserve.)
The proposition this post set up before the release was that a labour market averaging 92,000 jobs a month could coexist with a central bank whose live debate was whether to tighten further. July did not just fail that test — it removed the premise. The revised data says the average was never 92,000, and the current run rate is nearer 20,000. What has not yet happened is the thing that would force the Fed's hand: unemployment has not risen, and on the published rate it fell. Read the methodology behind the five factors to see how growth and rate expectations are separated when they point in different directions, and the Q2 employment cost index read for the wage-cost side of the same question.
Educational macro context only — not investment advice.