The jobs numbers keep getting revised down

Employers added 142,000 non-farm jobs in August, short of consensus forecasts, and the previous two months were revised down by a further 86,000. Downward revisions keep coming, and there are now too many to dismiss as a statistical quirk. They confirm that the economy is slowing.

June and July cut again

July's gain, first reported at 114,000, now stands at 89,000, and June's has been cut to 118,000. That leaves average payroll growth over the past three months at about 116,000, well below the pace reported earlier in the year. The unemployment rate edged down to 4.2 per cent in August, which offers some comfort, but only after a string of increases that had already triggered the Sahm recession rule.

The detail beneath the headline was mixed. Much of August's gain came from construction and health care, while manufacturing shed jobs, and temporary-help agencies, which employers cut first when they grow cautious, continued to shrink. The slight fall in unemployment partly reflected people counted as on temporary layoff in July returning to work, which suggests July's jump was exaggerated by one-off factors such as the hurricane that struck Texas.

A pattern, not a blip

The latest revisions follow the Bureau of Labor Statistics' preliminary benchmark on 21 August, which found that payrolls in the year to March had been overstated by 818,000. August is far from the first month this year in which earlier estimates have been marked down. One revision is noise. A run of them in the same direction says something about the economy.

The errors matter beyond the statistics office. Companies' hiring plans, wage settlements and the Fed's own projections all lean on the first print, so an economy that looks stronger than it is ends up with tighter policy than it needs. Bond traders have drawn the conclusion faster than officials: two-year Treasury yields have fallen sharply since the spring, and the debate has moved on to whether the first cut should be a quarter-point or half a point.

First prints run hot

When the economy turns, first survey estimates lag behind reality. By the time the final revisions are in, reported job growth has routinely been cut by thirty per cent. The Federal Reserve cannot rely on the headline figures to judge the economy's health. Demand for labour is weaker than the published numbers suggest.

The reasons are mechanical. The first estimate rests on the employers who reply in time, and response rates to the payroll survey have fallen since the pandemic, so the early sample is thinner than it used to be. Firms under strain are more likely to reply late, which means bad news tends to arrive in the second and third estimates. The model the BLS uses to estimate hiring at new businesses extrapolates from past patterns, and so keeps adding jobs for a while after conditions have changed.

What 2008 taught

The same thing happened on the way into the financial crisis. Through 2008 the first estimates repeatedly understated job losses, and the depth of the recession became clear only as revisions came in. The benchmark for March 2009 was the last to be revised down by more than this year's preliminary figure. Revisions run the other way in recoveries, as they did in 2021, which is why their direction is a useful guide to where the economy is heading.

Who is misled by the first print

The cost of optimistic first estimates falls on those who act on them. Borrowers pay rates set by a central bank that believed the economy was stronger than it was. Firms that plan hiring and investment around the monthly figures may keep expanding into a slowdown. Workers negotiating pay lose leverage later, when the true picture emerges and employers pull back more sharply than they would have done with accurate information.

The household survey, which asks people rather than employers whether they have work, has shown weaker job growth than the payroll survey for much of this year. It is noisier month to month, but it does not depend on a birth-death model, and its message has been consistently gloomier. When two independent measures disagree and the revisions keep moving one towards the other, the weaker reading deserves more weight.

The case against

Some of the weakness may reflect supply as much as demand. Immigration has swelled the labour force, and unemployment has risen partly because more people are looking for work rather than because more are being laid off. August's dip in the jobless rate suggests the labour market is cooling rather than collapsing. Those points argue for steady quarter-point cuts rather than panic. They do not argue for keeping policy where it is.

The next pieces of evidence arrive quickly. August's consumer price figures next week will show whether inflation leaves room for a bigger move, and the committee's new projections will reveal how far officials have marked down their view of the economy. The final version of the annual benchmark will be folded into the official series early next year, and the September jobs report in early October will bring another round of revisions to the summer months.

The Federal Open Market Committee meets on 17 and 18 September with its policy rate still at 5.25 to 5.5 per cent, the highest in more than two decades. The steady run of downward revisions shows the jobs market cooling fast. The Fed has little reason left to keep policy this tight, and should start cutting this month. Expect the summer's payroll figures to be marked down again in coming reports, and expect a cut on 18 September of at least a quarter-point, with job-seekers the main losers from every month of delay.