The headline number is not the labour market: how the monthly payrolls print is built, and what it leaves out

For roughly ninety minutes on the first Friday of every month, the price of nearly everything with a tape pauses and listens to one number. The Bureau of Labor Statistics publishes its Employment Situation summary at 8:30 a.m. Eastern; within four minutes, the S&P 500 has usually moved by its annualised volatility budget for the day, the two-year Treasury yield has repriced, and a Federal Reserve watcher's spreadsheet has updated its probability of a cut in three months. The number in question is nonfarm payrolls, and it is treated, in practice, as the labour market.

It is not. The monthly payroll print is a survey estimate, benchmarked to a much smaller universe than the headlines suggest, revised twice in the two months after publication and again in annual benchmarks, and structurally blind to several of the most consequential shifts underway in how Americans actually work. A worker who switches from a salaried job at a hospital to a contract role delivering meals for a platform will, in most months, show up as zero change. A small-business owner who adds a second part-time helper at minimum wage counts identically to a hedge fund that hires a senior quantitative at seven figures. The headline is what markets trade, and the headline is a coarse instrument applied to a fine-grained reality.

The shape of the number

Nonfarm payrolls come from the Current Employment Statistics survey, a panel of roughly 119,000 businesses and government agencies that collectively employ about 629,000 workers. The BLS extrapolates from that panel to a national count of around 159 million jobs. The establishment survey, as it is known, is large, but it is not designed to detect the kind of compositional shift that has defined the post-2023 labour market: the migration of work from full-time salaried employment into gig, contract, and platform-mediated arrangements that fall outside its sampling frame in different ways.

The companion instrument, the Current Population Survey, is what produces the unemployment rate. It is a household survey, smaller, more volatile, and frequently revised. The two surveys are released together on the same morning but tell structurally different stories. A month in which establishments report robust hiring can coincide with a household survey in which labour force participation falls and the unemployment rate ticks up because the household survey counts someone who has stopped looking as out of the labour force entirely, whereas the establishment survey never sees them at all. The single-number framing flattens this distinction.

What the revisions erase

Initial payroll prints are revised twice. The first revision arrives with the following month's release; the second arrives with the month after that. Annual benchmark revisions, which reconcile the survey estimates with the universe of unemployment-insurance tax records, can and routinely do move the level of payrolls by hundreds of thousands of jobs in either direction. The August 2024 benchmark revision, for instance, knocked 818,000 jobs off the level previously reported for the year ending March 2024, an adjustment larger than the entire payroll gain reported in any single month of 2023 outside the post-pandemic reopening. The initial prints had been treated as fact by markets and policymakers for the better part of a year.

This is not a peripheral technical detail. If the Federal Reserve is calibrating policy to a series that will be revised by half a million or more in either direction in the following year, the precision implied by the headline number is fictional. The Fed funds futures curve trades on the basis of the first print; the underlying reality of the labour market is the second, third, and fourth derivative of that print as it is gradually corrected.

What the survey cannot see

Three shifts in particular fall outside the establishment survey's effective aperture. First, the rise of platform-mediated work. A driver who signs on with a delivery platform in addition to a part-time retail job will appear in both surveys; a driver who leaves a traditional job and works entirely through a platform may appear in neither, depending on how the platform classifies them and whether they file a tax form the BLS can match. Second, the steady increase in remote and hybrid arrangements, which has changed where work happens but not the count of jobs. Third, the growth of small-business and self-employment in sectors that are systematically under-represented in the establishment panel, including creator-economy, freelance knowledge work, and contract professional services.

The household survey picks up some of this, which is why participation, hours worked, and the multiple-jobholder series often diverge from the establishment headline in informative ways. A labour market in which the establishment count is flat, but multiple jobholding is rising and average weekly hours in the household survey are climbing, is not the same labour market as one in which hours and second jobs are stable and the headline is rising. The single number cannot tell which story is the real one; the package, taken together, can.

Why markets trade it anyway

The establishment of the payroll print as the central market event of each month is partly an artefact of history and partly a coordination problem. It is the first major macro release of the month, it has the longest unbroken series, and it is dense enough in sub-components (average hourly earnings, the workweek, the unemployment rate, the labour force participation rate) that a careful reader can build a fairly complete picture from the package. The market does not trade the headline because the headline is the best statistic available; it trades the headline because the headline is the synchronisation point, the moment at which a large enough fraction of participants update their priors to make a price move.

This is a structural vulnerability. A series that is the centre of gravity for global asset allocation, and that is known in advance to be revised in ways that can move its level by hundreds of thousands of jobs, is a series whose signal is permanently degraded by its own prominence. The August 2024 benchmark was not an anomaly; it was the visible instance of a routine that has repeated itself in every recent revision cycle.

The deeper risk is that policy follows the signal. If the Federal Reserve responds to a headline that is subsequently revised down by several hundred thousand, it has, in effect, tightened into a labour market weaker than it believed. If it responds to a print that is revised up, it has eased into a market tighter than it believed. Either error has real distributional consequences: in the first case, into a slowdown; in the second, into the inflation that follows. The composition of the print, the share of full-time versus part-time work, the diffusion across industries and wage tiers, and the revision path of the prior two months are the data a serious reader of the labour market ought to be tracking. The first derivative on release morning is the data they will get.

The other story in the same package

It is worth stating plainly what the release actually contains, because the headlines obscure it. Average hourly earnings, the workweek, the diffusion index of industries adding payrolls, the unemployment rate, the labour force participation rate, the breakdown by job tenure, the share of employment that is part-time for economic reasons, the multiple-jobholder count, and the revisions to the prior two months are all published in the same 8:30 release. A reader who treats the package as a package, rather than as a single number with footnotes, sees a labour market whose internal composition is shifting in ways the headline cannot summarise. A reader who treats the package as a single number will be right, on average, about the direction of the next quarter; they will be wrong, often and persistently, about its texture.

The texture is what policy is supposed to respond to. A labour market in which hiring is concentrated in healthcare and leisure and hospitality, with manufacturing and information contracting, is not the same animal as one in which the reverse is true, even if the headline net print is identical. The release has the data to make that distinction. The wire copy usually does not. The price action on the first four minutes of the first Friday of the month is the price action of a market that has agreed to forget this for thirty days, and then remember it again, and then forget it again.

Desk note: Monexus treats the monthly Employment Situation as a package, not a headline, and notes that the revisions path is part of the signal, not noise around it. The Bureau of Labor Statistics publishes the underlying series on its site on the morning of release; wire copy usually foregrounds the headline and buries the revision. Our coverage prioritises the revision.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://t.me/AngelList/0
• https://t.me/producthunt/0
• https://www.bls.gov/ces/
• https://www.bls.gov/cps/