America had 818,000 fewer jobs than we thought

On 21 August the Bureau of Labor Statistics published preliminary benchmark revisions showing that US non-farm payrolls had been overstated by 818,000 jobs in the twelve months to March 2024. It was the largest downward revision to employment in fifteen years.

How the benchmark works

The monthly payrolls number comes from a survey of employers. Once a year the BLS checks it against a near-complete count drawn from state unemployment-insurance tax records, the Quarterly Census of Employment and Wages. The preliminary benchmark compares the two for March, and the final version is folded into the official series with the January employment report early next year. Revisions of a few hundred thousand in either direction are routine, and the last miss of this scale came in the benchmark for March 2009, when the survey failed to keep pace with the collapse in hiring during the financial crisis.

The news was not a complete surprise. The quarterly census is published with a lag of several months, and the Philadelphia Fed has for some time produced its own early benchmark estimates from the same tax records. Those had pointed to weaker job growth than the monthly survey reported, and forecasters had braced for a large downward revision. Markets therefore took the figure calmly; the scale confirmed a suspicion rather than creating one.

Jobs that never were

The revision confirmed what sceptical economists had long suspected. The BLS birth-death model, which estimates hiring at newly formed businesses, kept overstating the strength of the labour market while high interest rates pushed firms into bankruptcy. The economy was adding roughly 175,000 jobs a month, well short of the reported 242,000. The Fed was setting policy against a hiring boom that did not exist.

The mechanism is simple enough. A survey of existing employers cannot see a firm that opened last month, nor can it always tell a firm that has closed from one that failed to reply. The BLS fills the gap with a model that assumes openings and closures behave as they have in the past. When credit tightens and closures climb, that assumption flatters the count, and the error compounds month after month until the tax records arrive to correct it.

Spread across the year

Divided evenly, 818,000 missing jobs works out at about 68,000 a month, almost exactly the gap between the two monthly averages above. That turns a hiring boom into a solid but unspectacular expansion. The adjustment falls hardest on the parts of the count the model has to guess at, which means small and young service firms, where openings and closures are most frequent and where tight credit bites first. Professional and business services took the largest cut in the preliminary figures, with leisure and hospitality and manufacturing also marked down.

A puzzle about productivity

The revision raises a question about the rest of the economy. Output is measured separately, through spending and income data, and it has not been revised down alongside payrolls. If the same volume of goods and services was produced by fewer workers, output per worker grew faster than previously thought. That is good news for the economy's long-run capacity and for inflation, because rising productivity lets firms pay higher wages without raising prices as much. It also suggests that strong growth over the past year owed more to efficiency and less to hiring than the headline figures implied.

The alternative explanation is less comfortable. If the gap reflects workers the tax records cannot see, then employment was stronger than the benchmark says and productivity weaker. The two readings carry different implications for how much slack remains in the economy, and the data cannot yet settle which is right.

Why the Fed should care

The timing is awkward for the central bank. Through 2023 and into 2024 officials pointed to strong payroll growth as evidence that the economy could bear a policy rate of 5.25 to 5.5 per cent without breaking. Minutes of the July policy meeting, released on the same day as the revision, showed that a large majority of officials already expected to cut in September if the data held up. The revision strengthens their hand and weakens the case of those who wanted to wait.

It also bears on how the Fed reads the present. If the monthly survey overstated hiring during a period of tight credit, it may still be doing so. Officials who have relied on payroll growth to judge the labour market will give more weight to other indicators, such as unemployment, job openings and the rate at which people leave jobs voluntarily, all of which have softened.

The case for scepticism

The benchmark has flaws of its own. The tax records cover only workers inside the unemployment-insurance system, so they can miss people such as recent immigrants working informally, a group that swelled over the period. Part of the 818,000 may therefore be jobs the census cannot see rather than jobs that never existed, and the final figure could land some way from the preliminary one. The household survey, which counts people rather than payroll slots, had already been showing weaker hiring, and the unemployment rate had risen to 4.3 per cent by July. Whatever the exact split, every measure points the same way.

Futures markets had already priced a September cut as all but certain, so the argument now is about its size, and the August payrolls report due in early September will settle much of it. Removing 818,000 jobs shows the labour market was weaker than the official figures said, and leaves the Federal Reserve open to the charge that it waited too long. Pressure for bigger cuts this autumn will probably grow. A weak August report would make a half-point cut in September the likelier outcome, with borrowers the winners and a Fed that trusted the payroll count carrying the blame.