162,000 Against 53,000 Expected (4 September 2026): August Payrolls Tripled the Consensus, the Labour Force Grew 683,000 — and the Two-Year Took Back Every Basis Point Waller Gave

Total nonfarm payrolls rose 162,000 in August against a Dow Jones consensus of 53,000, the unemployment rate held at 4.1%, and June and July were revised up by a combined 55,000 — turning July's reported loss of 23,000 into a gain of 21,000. The number that mattered nearly as much sat in the household survey, where the labour force grew 683,000 and the participation rate rose to 61.6%, reversing the shrinking-denominator effect that had been flattering the unemployment rate all year. The transmission was immediate and it ran exactly where this note said to watch: the two-year Treasury yield rose 7.8 basis points to 4.412% within three minutes, the thirty-year only 1.5, and the dollar gained against all seven G10 peers. Of the three scenarios mapped here before the print, the beat is the one that landed — and the condition Governor Waller named for holding rates was inflation, which does not report until next week.

What actually happened

At 8:30 a.m. Eastern on Friday 4 September 2026 the Bureau of Labor Statistics reported that total nonfarm payroll employment rose by 162,000 in August. The Dow Jones consensus was 53,000. CNBC published the figure at 8:31 a.m. under a headline noting payrolls rose much more than expected, reporting the unemployment rate at 4.1%.

Take the beat apart before reading anything into it. Total private payrolls added 127,000 and government 35,000, the latter after shedding 50,000 in July. Goods-producing industries added 41,000 — construction 22,000 and manufacturing 16,000, the latter notable in a year of tariff disruption — and private service-providing industries 86,000. The unemployment rate was unchanged at 4.1%, exactly as forecast. Average hourly earnings for all private employees rose ten cents to $37.75, a monthly gain of 0.27% that rounds to the 0.3% consensus, and an annual rate of 3.1% against 3.2% in July.

That annual figure is worth a moment because it was predictable and was predicted. August 2025 was itself a strong month for wages, so the base effect alone was going to pull the year-over-year rate down about a tenth even if the month came in on consensus. It did, and it did. The wage side of the inflation argument therefore gained nothing from this release in either direction — which leaves the payroll count and the household survey to carry all of the information.

Source: BLS Employment Situation, 4 September 2026; consensus per CNBC, 3 September 2026.

The revisions went the other way

This is the part of the release that changes the picture most, and it is the part almost nobody trades in the first minute.

June and July were both revised up. July's reported loss of 23,000 became a gain of 21,000, a revision of 44,000; June's 20,000 became 31,000, a revision of 11,000. Together the two months gained 55,000 jobs after publication. Recall what the pre-release arithmetic looked like: CNBC noted on 3 September that June and July combined showed a net loss of 3,000 jobs. Those same two months now show a combined gain of 52,000.

Run that through the three-month average, which is the statistic policymakers actually use to strip out monthly noise. Before this release the trailing three-month average payroll gain was roughly 20,000 a month against a prior-twelve-month average of 34,000 — a labour market that had, on the published numbers, essentially stopped adding jobs. After it, the June-July-August average is about 71,300 a month. That is not a rounding change. It is a different characterisation of labour demand, arrived at without anyone changing their model.

Two cautions, both real. First, this cuts against a well-documented pattern: CNBC observed before the print that the initial August figure had been revised lower in each of the previous four years, and this month's revisions to prior months ran the other way. Second, August's own 162,000 is a first print, and first prints are the least settled numbers in the document — June's original 57,000 became 20,000 and is now 31,000, a round trip covered here when it first landed. Anyone treating 162,000 as final is making the same mistake in the opposite direction to those who treated −23,000 as final a month ago. The fuller account of that release and its revisions is in the July note.

The denominator stopped shrinking

The household survey had been the most interesting document in American macro for seven months, for a reason that had nothing to do with jobs and everything to do with arithmetic. Between December 2025 and July 2026 the civilian labour force fell from 171.495 million to 169.094 million — 2.401 million people out of the measured workforce, about 343,000 a month — and the unemployment rate fell from 4.4% to 4.1% while that happened. Those two facts were the same fact. Unemployment is a ratio and the labour force is its denominator.

In August the denominator grew. The labour force rose 683,000 to 169.777 million, its first substantial monthly increase of the year, while household employment rose 569,000 to 162.746 million. Because the workforce grew slightly faster than employment, the number of unemployed people actually rose 115,000 to 7.031 million and the rate stayed at 4.1%. Participation went from 61.4% to 61.6%, the employment-population ratio from 58.9% to 59.1%, and the broader U-6 measure of underemployment fell from 7.9% to 7.7%.

One consequence deserves flagging because it points the other way. A shrinking workforce had pushed the breakeven payroll number — the monthly gain needed to hold the unemployment rate flat — down a long way, which is why weak prints kept failing to raise unemployment. If the workforce is now growing again, breakeven rises with it, and the same 162,000 that looks strong against a contracting labour force looks merely adequate against an expanding one. The August report is one month of data on that question, not an answer to it.

Note also what did not appear. The government cancelled Temporary Protected Status for thousands of Haitians in July, a move projected to affect around 350,000 people and widely expected to subtract from the employment rolls. The labour force grew by 683,000 regardless. Whatever the policy's eventual effect on the data, it did not show up as a labour-force contraction in August.

Why ADP said 38,000 and the BLS said 127,000

The private-payroll gap was 89,000, and the sector-level disagreement was sharper than the headline gap suggests.

Sources: BLS Employment Situation, 4 September 2026; ADP National Employment Report, 2 September 2026, as reported by CNBC.

Two of those rows have opposite signs, which is more informative than the size of the miss. ADP samples the payroll records of its own clients; the BLS establishment survey samples roughly 120,000 businesses and government agencies and is benchmarked annually against unemployment-insurance tax records. They are estimating the same quantity from different frames, and in a month when hiring is concentrated — food services and drinking places alone accounted for 68,000 of the leisure and hospitality gain, more than the whole of ADP's private total — frame differences get amplified.

The ISM services employment index at 47.8%, in contraction for a second month, pointed the same way ADP did and was wrong in the same direction. Initial jobless claims at 206,000 for the week ended 29 August, published by the Department of Labor, were the one high-frequency series consistent with what arrived: employers were not firing, and it turns out they were hiring rather more than the survey-based proxies believed.

What repriced, and where on the curve

The reaction was fast, and its shape is the whole mechanism.

The two-year moved five times as far as the thirty-year. That asymmetry is what distinguishes a repricing of the expected policy path from a repricing of the premium investors demand for duration risk — the distinction traced in detail through August's long-end selloff, where the move ran the opposite way and the dollar barely responded. Here it ran at the front, which is the leg of the curve on which carry is actually earned.

Note the level, not just the change. The two-year had fallen more than five basis points on Thursday when Governor Waller signalled he could support a hold; at 4.412% it now sits above the 4.39% at which it closed on 2 September. The entire Waller repricing was undone at the front end in roughly three minutes, on a curve that was already priced above the midpoint of the current target range.

The currency followed, as the mechanism requires. The ICE US Dollar Index rose 0.47% to 99.368. Against the G10: EUR/USD −0.28% to 1.1592, GBP/USD −0.19% to 1.3495, AUD/USD −0.17% to 0.7187, NZD/USD −0.26% to 0.5866, USD/CHF +0.54% to 0.8119, USD/CAD +0.52% to 1.3863 and USD/JPY +0.46% to 156.51 — the yen giving back part of Thursday's best day since July, a storyline followed in the intervention note. Thursday the dollar fell against all seven; Friday it rose against all seven. The factor-by-factor read on the currency itself sits on the USD page.

Two cross-asset confirmations round it out. WTI for October delivery fell 1.03% to $90.36 and COMEX December gold 0.62% to $4,511.70. A stronger dollar and higher real front-end yields raise the opportunity cost of holding a non-yielding asset, and gold behaved accordingly — reasonable evidence that the market read this as a rates event rather than a growth event. For context on the session before, Thursday had closed with the S&P 500 up 1.06% at 7,747.71, the Nasdaq Composite up 1.40% at 26,584.06 and the Dow up 1.18% at 53,686.11, all on the dovish reading that Friday partially reversed.

The scenario that landed

This note mapped three before the print, and the first one is what arrived. It read, in part, that a beat above 80,000 "corroborates the Chair's characterisation of the labour market as broadly consistent with full employment, removes the strongest argument for waiting, and would do so into a market that has just given up thirteen points of hike probability," with the work "done at the front of the curve." That is a fair description of what happened, and the reason to record it is not that the map was right but that the mechanism was legible in advance. The channel was knowable; the number was not.

What the beat did not do is settle the September question, and this is where the distinction between an argument and a test matters. Waller's stated condition was explicit and it was about prices: he said he would be inclined to support holding "if this continues in the data due over the next two weeks," meaning the disinflation he had identified in the three-month rate of the Fed's preferred gauge, which has fallen from 4.76% in February to 3.05%. The only major inflation releases in that window are next week's consumer and producer price indexes. A strong payroll print weakens the case for waiting on labour-market grounds — Waller had already called the jobs picture "satisfactory," and Governor Michael Barr "stable" — without touching the test he actually set.

And the argument pointing the other way did not go away either. The ISM services prices index registered 72.6% in August, its highest since August 2022, with the twelve-month average at 68.5%. Brent has been trading in the mid-90s. In its 29 July statement the Committee held the target range at 3-1/2 to 3-3/4 percent and wrote that "job gains have kept pace with the workforce, and the unemployment rate has changed little" — a benchmark that a 162,000 print alongside a 683,000 increase in the workforce clears, but not by the margin the headline implies. The hold passed 9-3, with Hammack, Kashkari and Logan preferring a quarter-point increase; Chair Kevin Warsh used Jackson Hole to say that the summer's better inflation readings did not tell him "underlying trends have meaningfully improved." A Committee that was already split now has a stronger labour market and an unresolved inflation question.

What would change the picture

Four things, in rough order of how much they would move the front end.

Next week's CPI and PPI, decisively first. Waller published his own reaction function and named the trigger, which is unusually clean information: the market can price two known releases rather than guess at a policymaker. Friday's payroll figure raised the bar those prints have to clear to produce a hold, but it did not clear it for them.

The revisions to July and August, second — and this is not a formality. The last four August first prints were revised lower, and the two months revised this Friday moved by 55,000 combined. A statistic that can move 44,000 after publication should not be carrying a policy decision on its first release, and yet the front end just repriced almost eight basis points on one.

Participation, third. One month at 61.6% is not a trend, and the question it raises cuts against the hawkish read of the headline: a genuinely recovering workforce raises the breakeven payroll number and rebuilds the slack that seven months of contraction had quietly erased from the statistics. Watch whether the labour force adds people again in September.

Energy, fourth, because it has been setting the inflation expectations that set the front end all summer. WTI fell about a dollar on Friday and remains near $90, with the ISM panel still citing the Middle East conflict among its most-reported supply-chain problems; the mechanics of that channel were traced through the collapse in Hormuz transits.

The Committee's decision on 15-16 September is not knowable from here and this note has not guessed it. What the August report established is narrower and more useful. The labour market the market had been pricing — roughly 20,000 jobs a month, a workforce shrinking by a third of a million people every month, an unemployment rate falling for the wrong reason — was substantially an artefact of unrevised data. On the numbers as they now stand, hiring is running near 71,000 a month, the workforce is growing again, and the unemployment rate is flat for the right reason. Whether that is enough to raise rates depends on two inflation prints that have not been published. What is knowable is which number carried the information, and where on the curve it landed. More about how this site reads data releases is on the about page.

Educational macro context only — not investment advice.