600,000 jobs erased from the record

The Bureau of Labor Statistics has taken nearly 600,000 jobs out of the 2024 payroll count in its annual benchmark revision to non-farm payrolls. The size of the cut backs up what sceptics had said for months. The monthly establishment survey had been making the labour market look stronger than it was.

A revision flagged in August

Nobody should have been surprised. Last August the bureau published a preliminary estimate that payrolls in the year to March 2024 had been overstated by 818,000, the largest such revision since 2009, and the final number released with the January jobs report is smaller but of the same order. The benchmark comes from the Quarterly Census of Employment and Wages, which counts jobs from the unemployment-insurance tax filings that almost every employer must make. Those records arrive months later than the monthly survey but are far more complete, which is why the survey is realigned to them once a year.

Spread across the twelve months to March 2024, the cut works out to roughly 50,000 fewer jobs a month than first reported. That is enough to turn a labour market that looked comfortably strong into one that was growing at a more ordinary pace, and it changes the story told by every monthly headline over the period.

How the rewrite is done

The adjustment is mechanical. The bureau resets the level of employment in March 2024 to the census count, then spreads the gap back evenly across the previous eleven months, so each month's reported gain shrinks by a similar slice. The months after March are rebuilt too, because the birth-death estimates for that period are recalculated with the new information. The result is a whole new history for 2024, released on the same morning as a fresh month of data, which made the January report unusually hard to read. The headline for January itself, a gain of 143,000, was the least important number in it.

Businesses that never opened

Most of the blame falls on the BLS birth-death model, which estimates how many firms open and close when the economy is changing direction. With interest rates high and commercial bankruptcies elevated, the model kept adding new businesses that were never born and missed closures in retail and hospitality. Central bankers and corporate planners were therefore working from a picture of the economy with more momentum than it really had.

The model exists because the survey is blind at both ends. New firms cannot report until they are added to the sample, and dead ones do not return questionnaires, so the bureau imputes both from past patterns. That method works tolerably in steady times and fails at turning points, which is exactly when accurate numbers matter most.

The Fed was tighter than it knew

For the Federal Reserve, losing 600,000 jobs from the record suggests its tightening bit harder on the real economy than the first headline numbers implied. The central bank held rates at their cyclical peak while relying on employment figures that turned out to be wrong. It kept its policy rate at 5.25–5.5 per cent from July 2023 until September 2024, when it cut by half a point, and two quarter-point cuts followed before it paused in January.

The warning signs were visible before the official admission. Unemployment rose to 4.3 per cent in July 2024, enough to trigger the Sahm rule, a recession indicator based on the jobless rate, and the market scare that followed came only weeks before the Fed's outsized first cut. With hindsight, the survey data had been flattering the economy for the whole period in which officials were debating whether policy was restrictive enough.

Others relied on the same flattering numbers. Bond traders price the path of interest rates partly off each payroll release, so the market spent much of 2023 and early 2024 expecting fewer cuts than a truer count might have justified. Borrowers on floating rates, from small firms with bank loans to households carrying card balances, paid for that. Companies setting pay and hiring budgets, and state governments projecting income-tax receipts, also planned around a labour market that was hotter on paper than in practice.

Old errors, new data

Real-time figures have misled policymakers before. The first estimates of American output in late 2008 understated how fast the economy was shrinking, and the Fed of the 1970s set policy on estimates of spare capacity that later proved badly wrong. The lesson each time was the same: when the model fills in the gaps, it tends to assume the recent trend continues.

There is a fair case against reading too much into the cut. Payrolls still grew throughout the period, only more slowly, and unemployment in the latest report stood at 4 per cent, low by historical standards. Some economists also argue that the benchmark undercounts, because recent immigrants working without authorisation rarely appear in unemployment-insurance records, which would make the true revision smaller.

The Fed also had a second source it could trust. The unemployment rate comes from the household survey, which is not revised by the benchmark, and it was already drifting up through 2024. Officials who weighted it over the payroll count were not misled for long. The household survey has its own problem, though, since its population weights depend on immigration estimates that the Census Bureau raised sharply in December, and the January release was adjusted to reflect them.

What to watch next

The quarterly census now offers an earlier check on the monthly survey, and the next preliminary benchmark, due in the late summer, will show whether the model's errors carried on after March 2024. Weekly jobless claims and the quits rate are less prone to this kind of drift and deserve more weight than the headline payroll number for now. The Philadelphia Fed publishes early benchmark estimates for individual states from the same census data, and they give a running read on whether the national survey is drifting again.

The policy lesson points backwards as much as forwards. The revision shows the Fed kept policy restrictive while hiring was already softening, and the real economy was probably slowing for most of 2024 before the official data admitted it. Expect officials to lean less on the monthly payroll headline this year and more on unemployment and claims, which makes the jobless rate the number most likely to move rates by the summer.