911,000 jobs that never existed

The Bureau of Labor Statistics said in its preliminary annual benchmark revision that payrolls in the twelve months to March 2025 had been overstated by 911,000 jobs. That is the largest downward revision to employment in modern American history.

Spread across the year it covers, the cut amounts to about 76,000 fewer jobs a month than the monthly reports showed at the time. A labour market that was described as adding jobs at a healthy clip for most of that period was growing far more slowly than anyone at the time was told. The revision arrives after a summer of weak payroll reports, including August's gain of just 22,000, and so extends a slowdown that many had thought began only this year.

How the count goes wrong

The monthly payroll figure comes from a survey of employers, which is quick but incomplete. Once a year the statisticians check it against the Quarterly Census of Employment and Wages, which draws on the unemployment-insurance tax records filed by nearly every employer in the country. That census is slower but far more complete, and the difference between the two is the benchmark revision.

The weak link is the model that estimates jobs at firms too new to appear in the survey, minus jobs lost at firms that have closed. In a steady economy the births-and-deaths estimate is fairly accurate. When the economy turns, it tends to assume that new businesses keep opening at the old rate, and so it overstates hiring at exactly the moment hiring is slowing. The same pattern produced a large downward revision in 2009.

Nearly a million jobs that never were

The revision confirms that quirks in statistical modelling inflated the post-pandemic jobs boom by a wide margin. Nearly one million jobs that shaped companies' hiring plans and federal budget forecasts, and that justified a hawkish Federal Reserve, did not exist. The real economy was slowing much faster, and more dangerously, than policymakers realised.

This is the second year running that the benchmark has cut deep. Last August the preliminary estimate removed 818,000 jobs from the year to March 2024, and a month later the Fed opened its easing cycle with a half-point cut. Two consecutive revisions of this size suggest a persistent bias in the monthly estimates during a period of slowing growth, not a one-off error.

Those numbers fed straight into decisions. The Fed's staff projections, the Congressional Budget Office's revenue estimates and state budget offices all start from the official payroll series. Businesses use it to judge whether to expand, and bond investors use it to judge how long rates will stay high. An overstatement of nearly a million jobs tilted all of those judgements towards optimism during a year when caution would have been more appropriate.

The productivity objection

The strongest counterargument is that output has not been revised. Gross domestic product is measured separately, and if the economy produced the same goods and services with fewer workers, productivity was higher than thought. On that reading the revision describes a leaner, more efficient economy rather than a weaker one, and wages per worker were supported by real gains.

There is also a question about what the census misses. It counts only jobs covered by unemployment insurance, so work done off the books, including by some of the immigrants who arrived in large numbers in 2023 and 2024, may be under-recorded in the benchmark rather than over-recorded in the survey. Neither argument restores the missing 911,000 jobs to the payrolls that matter for tax receipts and household income, and neither changes the direction of the recent monthly data.

Short yields tumble

Bond markets repriced at once and sharply. Short-term yields fell steeply as traders bet on a larger and earlier run of Fed rate cuts.

The logic runs from the Fed's own framework. Policymakers judge how tight policy is by comparing the economy with its potential, and a labour market that was growing more slowly for longer implies more slack than they assumed. With the benchmark rate still at 4.25 to 4.5 per cent, well above officials' median estimate of the neutral rate, the case for keeping policy restrictive rested partly on job numbers that have now been cut. Two-year Treasury yields, which move with expected policy rates, adjusted accordingly.

The winners are borrowers whose loans reset with short rates, from companies with floating-rate bank debt to households with credit-card balances. The losers are savers in cash funds and, if the revision signals a sharper slowdown, lenders to weaker borrowers whose repayment depends on steady employment. For the dollar, lower expected rates narrow the yield advantage over other currencies.

The revision also changes the reading of the recent past. The Fed's decision to pause after its December cut rested on the view that the labour market was solid and that tariffs posed the bigger risk. With the benchmark showing that hiring was already weak through the first quarter of this year, the pause looks longer than the data justified, and the weak summer figures look like a continuation rather than a sudden break. Officials who argued for patience will find that case harder to make.

Credibility and the calendar

The revision lands in a politically charged setting. The president fired the head of the Bureau in August after a weak jobs report, and the latest numbers will be read by some as vindication and by others as proof that the statisticians were doing their job. The methods behind the benchmark have not changed, and the final figure will be folded into the official series with the January employment report, published in February.

The nearer dates are the consumer price index for August, due later this week, and the Fed's meeting on 16 and 17 September. A quarter-point cut is fully expected; the revision strengthens the argument of those who want more. Removing 911,000 jobs from the record shows the Fed kept policy tight on the strength of jobs that did not exist, and it will probably have to cut faster over its next few meetings to stop the slowdown turning into a contraction.