NFP is no longer playing the same game: August 2026 jobs report preview
Friday 4 September, the last jobs report before the FOMC. Consensus is only +55k, against a break-even rate that has fallen to −10k/+30k. The 2 September ADP came in at +38k, with the details concentrated in a single sector, a single region and a single firm size. Under Warsh's Fed, a weak print no longer buys a rate cut: it only removes hike premium.
Friday 4 September, 2:30pm Paris time. The last jobs report before the FOMC. A weak print will not buy a rate cut: under Kevin Warsh's Fed, it will only remove some of the hike premium. The whole reading of the report rests on that asymmetry.
What has changed since the last report
For two years the reflex was simple: weak NFP → rate-cut expectations → weak dollar, strong gold, duration bought. That reflex is broken.
On 29 July the Fed held the policy rate at 3.50–3.75% by a 9-3 vote, with the three dissenters calling for a hike (Hammack, Kashkari, Logan). On 28 August at Jackson Hole, Kevin Warsh described the labour market as consistent with full employment and declined to call financial conditions restrictive. In mid-August the market was pricing almost no cut in September, but roughly a 30% probability of a hike.
The direct consequence for Friday: the distribution of reactions is truncated on one side. A strong print feeds a trajectory that is already priced and can therefore move markets sharply. A weak print must first erase about 30 points of hike premium before it can start pricing anything else, which makes the rates move potentially violent but mechanically bounded, and the equity move ambiguous.
The break-even rate has collapsed
A second shift, less discussed. With a shrinking labour force (immigration restrictions on one side, retirements outnumbering entrants on the other), the pace needed to hold the unemployment rate steady is now estimated at between −10k and +30k a month. An NFP of +50k, which would have signalled an imminent recession in 2019, now signals a labour market that is not deteriorating.
That is what makes this release's consensus remarkable. At +55k, it sits barely 25,000 jobs above the top of the break-even range. The market is not expecting a recovery in hiring: it expects near-stagnation, and regards that as normal.
With a break-even rate close to zero, it is no longer the NFP that measures labour demand. It is the unemployment rate.
Forecasters agree on nothing
| Metric | Value | What it says |
|---|---|---|
| Consensus | +55.0k | Median of August forecasts. |
| Forecast range | +12k → +100k | An 88k spread around a median of 55k: major disagreement. |
| Previous | −23k | July, published on 7 August. |
| 12-month average | +27.7k | The norm for this cycle, far below a healthy regime. |
| Gap of previous to average | −50.7k | July was not a simple slowdown. |
| 36-month extremes | −156k → +237k | The range of possible outcomes remains wide. |
| Historical rank of previous | 34% | July sits at the 34th percentile of the last 36 months. |
| 3-month trend | −132.3 | The recent dynamic is plainly negative. |
A range of +12k to +100k around a median of 55k is a dispersion of 88,000 jobs, or more than 1.6 times the consensus itself. Forecasters do not differ on the strength of the move; they differ on its direction. In practice, the probability of a large miss against consensus is much higher here than on an ordinary release, and the market reaction will be all the sharper because nobody has a solid consensus position to defend.
July: a report much worse than its headline
The July report, published on 7 August, missed consensus by more than 100k. But the real damage was in the revisions and the household survey, not in the headline.
| Line | Value | Reading |
|---|---|---|
| Nonfarm payrolls (NFP) | −23k | Consensus between +83k and +95k. First negative month of the cycle. |
| June revision | +57k → +20k | −37k. The reference month collapses after the fact. |
| May revision | −66k | Two consecutive months mangled: the initial signal was wrong. |
| 12-month average (as cited by the BLS) | +34k | The cruising pace, which is the current norm. |
| Private sector | +30k | Positive. All of the month's damage comes from the public sector. |
| Public sector | −53k | Of which local education −50k, probably a calendar effect. |
| Retail | −19k | Closures of unprofitable stores. |
| Unemployment rate (U3) | 4.1% | Down from 4.2%, but for the wrong reason. |
| Participation rate | 61.4% | Lowest in more than five years. It is what pulls U3 down. |
The unemployment rate fell while employment fell. That is not an improvement: it is the denominator shrinking. This configuration, U3 down, participation down, is exactly what lets Warsh's Fed talk about full employment while observing negative job creation.
The 28 August benchmark revision
The BLS published its preliminary benchmark revision for the year ended March 2026: −79k on the total (−0.1%), but −178k on the private sector alone, which implies an upward revision of about +99k for the public sector. It is the smallest benchmark revision since 2021, so, counter-intuitively, good news for data reliability. The final revision will arrive only in February 2027, with the January 2027 report.
What to take from it for Friday: the weakness is concentrated in the private sector (retail, education and health, manufacturing, business services) and masked by the public sector. The private/public breakdown therefore matters more than the headline.
Why the August figure should rebound without proving anything
Three purely technical effects push the August figure up, and none says anything about labour demand.
- The return of local education. July's −50k mostly reflects the school-year calendar and seasonal adjustment. Part of those jobs reappear mechanically in August.
- A late Labor Day (7 September). The timing distorts seasonal factors in education, leisure and hospitality. Workers who dropped out of payrolls in July can be counted back in August.
- The first-print bias of August. The August report often comes in below expectations at first release (response rates are degraded by holidays) and is then frequently revised up the following month. This bias gives no reliable direction: it mostly argues against over-reading the first number.
In other words: a rebound to +55k driven mostly by public education and hospitality would be a false positive. The clean signal is the private sector excluding education, the unemployment rate and hours worked. And given these three effects, simply hitting consensus would be no achievement: it would be the expected result of a calendar, not of labour demand.
The 2 September ADP: the headline is passable, the details are not
ADP reported +38k for private employment in August, against +48k expected, after July was revised to +46k. That is the weakest pace of job creation since January. But the headline is by far the least interesting part of this release.
| Item | Change | Reading |
|---|---|---|
| Total private | +38k | vs +48k expected. Weakest since January. |
| Goods-producing | −10k | Of which manufacturing −17k and mining −5k, construction +12k. |
| Services | +48k | But education and health alone contribute +45k. |
| Education & health | +45k | The least cyclical sector of the economy. |
| Leisure & hospitality | +16k | Probable calendar effect. |
| Professional services | −16k | The cyclical core is contracting. |
| Trade & transport | −5k | Confirms the retail weakness seen in July. |
| Information | −4k | Continued compression of tech headcount. |
Three concentrations that reveal fragility
The 38,000 figure is not a weak number spread across the economy. It is a number carried by three very narrow pockets, and nil everywhere else.
- A single sector. Education and health contribute +45k while all services together make +48k. Take that item out and US services create 3,000 jobs. Yet education and health is the segment least sensitive to the cycle and largely funded by public spending and insurance: it is not a measure of corporate labour demand.
- A single region. The Northeast contributes +38k, or the entire national gain. Midwest +5k, South +3k, West −8k. The rest of the country nets to zero. Even within regions the dispersion is brutal: East North Central −15k against West North Central +20k.
- A single company size. Firms with 500 or more employees contribute +34k, or nearly 90% of the net gain. Mid-sized firms are at zero. And among small firms the detail is telling: +20k for those with 1–19 employees but −17k for those with 20–49.
This article argued above that a rebound driven by education would be a false positive. The ADP has just documented it ahead of time: most of August's private job creation is already concentrated in education and health. If the official report comes in near consensus thanks to that same item, plus the return of local public education, the figure will be in line while the underlying labour market is nonetheless weaker than in July.
Wages, meanwhile, are decelerating
This is the point that cuts the other way, and it matters. Base pay is up +3.2% year on year for all employees, +3.0% for job-stayers and +4.7% for job-changers. On gross pay, job-changers are at +7.3% after +7.5%, a deceleration.
The premium for changing employer remains 1.7 points on base pay, which still reflects some tightness. But the direction is towards easing, not acceleration. That lowers the probability of the stagflation scenario, which assumes wages picking up while employment stalls. ADP measures are not the BLS average hourly earnings and do not substitute for it, but they point the opposite way from that scenario.
The standard commentary draws the usual conclusion from this ADP: slowing employment, so stronger rate-cut expectations and a pressured dollar. That is precisely the reading the current regime invalidates. With CPI at 3.4%, PCE at 3.7% and three dissenters who voted for a hike in July, weak employment does not create room to ease: it removes hike premium. The dollar may indeed give way, but because the market is erasing an expected tightening, not because it is pricing an easing. The move therefore stops where inflation blocks it, and is mechanically bounded.
Four outcomes, only one really dangerous
Subjective weightings, to be revised after ADP and jobless claims. The reactions shown are first-reaction directions, not targets.
Beyond the top of the forecast range. Local education comes back strongly and the seasonal catch-up plays out in full. The market concludes that July was calendar, not a break. The September hike premium climbs back towards 45–50%. Trap to avoid: this is the most hawkish scenario, not the most reassuring.
The central scenario, covering all the space between the break-even rate and the top of the range. Above −10k/+30k, this number validates Warsh's "full employment" reading. The report settles nothing and the event shifts to August CPI, published the following week, before the FOMC.
A weak number accompanied by rising unemployment: this time it is demand, not supply. The market first erases the ~30 points of hike premium. But it does not automatically flip to a September cut: with PCE inflation around 3.7%, Warsh has set his bar on inflation, not on employment.
Employment stalls, wages accelerate. Warsh's Fed then has no trade-off available. It is the only scenario where equities and bonds fall together. Low probability, maximum impact: the one to hedge against, not the one to bet on.
Reading order at 2:30pm
The headline arrives first and it is the one that matters least. Order of priority:
- The headline, for the record. Decide nothing on it alone. It will be revised, probably by several tens of thousands.
- Revisions to the two previous months. Print plus revisions equals "net payrolls", which is what the market actually trades. After a −37k and a −66k, another negative revision weighs more than a flattering headline.
- The unemployment rate and its direction. With a break-even rate near zero, U3 is the true barometer of labour demand. A rise to 4.3% changes the report; an NFP of +40k with U3 stable does not.
- The participation rate. 61.4% in July. If it rises and U3 rises: supply is coming back, benign. If it falls again: unemployment flatters reality.
- Average hourly earnings, month on month. This is the line that can flip the report by itself. Above +0.4% m/m, the "employment" reading moves to the back seat.
- Private payrolls excluding education and health. The benchmark showed that the weakness is private and masked by the public sector. In August, local education is calendar noise: neutralise it before concluding.
- Average weekly hours and the diffusion index. Firms cut hours before headcount. A drop in average hours with a decent NFP is a more reliable leading signal than the headline.
The week, and what follows
This report is the last employment data the Committee has before its decision. But in a regime where three members already voted for a hike in July, the following week's CPI is the arbiter. The NFP sets the frame; inflation sets the decision. Dates other than NFP and FOMC: usual sequence, to be confirmed against the official calendar.
What would invalidate this reading
- A reversal in Fed communication. If a Board governor (not a regional president) publicly opens the door to a cut before the FOMC, the asymmetry described here closes and the NFP becomes a two-way event again.
- A disinflation shock. An August CPI clearly below expectations would give the Fed back its room for manoeuvre: a bad jobs number would again be buyable as before.
- A rise in unemployment without a drop in participation. If U3 rises while participation rises, it is a supply shock, not a deterioration in demand, and the Fed will read it that way.
- A badly digested benchmark revision. The market treated the −79k as harmless. If the debate shifts to the −178k in the private sector, perception of the starting point changes without any new data being published.
The +55k consensus and the +12k/+100k range come from the PrismaQuant snapshot at 31 August. These values move until the eve of the release, in particular after Wednesday's ADP and Thursday's jobless claims: recalibrate them before using the scenario thresholds. Forecasters also diverge on unemployment, with some projections at 4.2% when July came out at 4.1%.
Sources
- BLS — Employment Situation, July 2026
- CNBC — July jobs report
- St. Louis Fed — July flash report
- Bloomberg — 28 August benchmark revision
- Fed — FOMC minutes of 28–29 July
- CBS News — the 9-3 vote and the three dissents
- Jackson Hole 2026 — Kevin Warsh's speech
- ADP National Employment Report — August 2026
- Consensus, range and historical statistics: PrismaQuant snapshot at 31 August 2026.
- CNBC — July 2026 CPI