How 258,000 jobs disappeared

Two consecutive monthly revisions removed 258,000 jobs that had already been reported. The cuts came in the July employment report, published on 1 August, which lowered May's gain to 19,000 and June's to 14,000 and put July itself at 73,000. The episode shows how poorly modern economic surveys now work. Response rates are falling and companies open and close more often, so the first establishment figures have come loose from what is actually happening to payrolls.

The same release nudged the unemployment rate up to 4.2 per cent. Treasury yields fell sharply on the day, and futures traders moved to price an interest-rate cut at the Federal Reserve's September meeting. A labour market that had looked steady in the first prints of early summer now looks as though it stalled in the spring.

Missing replies, extra jobs

The Bureau of Labor Statistics establishment survey gathers payroll data from a shrinking pool of company human resources departments. When the economy slows, struggling firms and closed shops are the ones that fail to return the questionnaire on time. The BLS fills the gaps by assuming that firms which do not reply look like those that do, which inflates job creation in the first prints. The phantom jobs disappear only when quarterly unemployment insurance tax filings arrive.

The timetable makes the bias hard to spot. Each month's figure is published within weeks of the reference period and then revised twice as late questionnaires come in; only after that does an annual benchmark reconcile the survey with tax records. A second source of error is the birth-death model, which estimates hiring at newly opened firms the survey cannot yet see and nets it against closures. Because the model leans on past patterns, at a turning point it keeps crediting jobs to businesses that were never started.

Reading a noisy first print

For an investor, the practical response is to treat the first payroll figure as an estimate with a known bias rather than as a fact. Cross-checks help. The household survey, which counts people rather than jobs, weekly claims for unemployment insurance, private payroll counts from processors such as ADP and the job openings data all offer independent readings, and none of them depends on the same late responders. When the establishment survey runs well ahead of all of them for several months, the gap is more likely to be closed by a downward revision than by the others catching up.

Tax records arrive late

This has happened recently. In August 2024 the preliminary benchmark revision cut 818,000 jobs from the level of payrolls in the year to March 2024, a sign that the monthly survey had run well ahead of the tax data for a whole year. The next preliminary benchmark, covering the year to March 2025, is due in September. Another large downgrade would turn this summer's revisions into the visible part of a bigger overstatement.

Politics enters the data

President Donald Trump responded to the July report by dismissing the BLS commissioner, Erika McEntarfer, on the day it was published, and accused the agency of manipulating the figures. The charge misreads the mechanics. Revisions of this size come from late responders and model assumptions, and they are printed in the release for anyone to read. The greater risk is that the row weakens an agency that needs larger samples and faster access to tax data, while handing markets one more reason to discount its headline numbers.

The strongest objection to this argument is that revisions cut both ways. Over long periods the gap between first and final estimates averages out close to zero, and in the recovery after 2020 the early prints often understated hiring. The trouble is that the errors cluster at turning points, when the data matter most. In 2008 the first estimates understated the pace of job losses, and later revisions made the recession look markedly deeper than policymakers believed at the time.

The revisions matter well beyond the Fed. Payroll growth feeds into forecasts of income-tax receipts, state budget plans and company hiring decisions, and an overstated trend flatters all of them at once. A state government that sets spending on the assumption of steady job growth, or a retailer that staffs up for demand that is not there, makes the same error as the central bank, only with less capacity to reverse it.

A Fed misled

Because early data are persistently biased upwards, the central bank keeps policy tight after demand for labour has started to stall. The Federal Open Market Committee held its benchmark rate at 4.25 to 4.5 per cent on 30 July, two days before the revisions landed, with governors Christopher Waller and Michelle Bowman dissenting in favour of a cut. The 258,000 phantom jobs delayed easing by months and kept borrowing costs needlessly high for credit-sensitive industries. Homebuilders, small firms on floating-rate bank loans and indebted companies needing to refinance bore the cost, while savers in money-market funds collected the extra yield.

The market consequences follow a familiar chain. A payroll surprise moves expectations for the Fed's next decisions, which moves two-year Treasury yields, which feed through to the rates on car loans, adjustable mortgages and corporate credit lines. A statistical error at the start of that chain is therefore not an academic matter: for as long as the first prints overstated hiring, the pricing of short-term borrowing assumed an economy that was stronger than it really was.

Where the evidence comes next

Weekly jobless claims, the September benchmark and the quarterly census of employment built from unemployment insurance filings will show whether the slowdown is deeper than even the revised figures suggest. A benchmark revision close to zero would weaken the case for systematic bias and leave the summer's cuts looking like ordinary noise. A large negative one would confirm that the Fed has been steering by an overstated dashboard. The revisions show that the first payroll numbers owe more to statistical guesswork than to hiring, and investors will probably keep marking down headline releases for the foreseeable future.