The jobs report that disagrees with itself, again

Non-farm payrolls rose by 339,000 in May, well above a consensus that had expected a gentle slowdown. The household survey told the opposite story. It showed 310,000 jobs lost, and the unemployment rate rose to 3.7 per cent. The jobs report often contradicts itself, though rarely this neatly, and policymakers are left with two readings of one economy.

Two surveys, two stories

The gap between the payroll count and household employment usually widens when the cycle is turning. People with more than one job, swings in farm employment and the self-employed all distort the household figure. The payroll survey has its own weakness, since its birth-death model tends to overstate hiring when fewer new businesses are being formed. If company payrolls really are growing this fast, the Federal Reserve cannot afford a long pause. If the household survey is the better early signal, further tightening could break household and company finances.

Counting jobs versus counting people

The two surveys measure different things. The payroll survey asks employers how many jobs they filled, so a worker holding two part-time posts counts twice. The household survey asks about 60,000 households whether their members worked, so that worker counts once, while farm hands, the self-employed and domestic staff, whom the payroll survey leaves out, count too. In a month when employers add second jobs and the self-employed drift into payroll work, the payroll count can rise while the number of people in work falls. Some of May's gap probably reflects exactly that.

A large part of the household decline did come among the self-employed, a group the payroll survey never sees. That narrows the true disagreement between the two measures without closing it. Self-employment is volatile from month to month, but it also tends to shrink when small traders and contractors find work drying up, so the fall is hard to dismiss as noise.

The birth-death adjustment is the payroll survey's blind spot. New firms cannot be sampled until they exist on the survey's lists, so the Bureau of Labor Statistics adds a modelled estimate of jobs created by businesses opening, net of those closing. The model is built on past patterns, which is why it tends to flatter the figures when the pace of company formation turns down. The error is corrected only when the annual benchmark to unemployment-insurance tax records arrives, with a preliminary estimate in late summer and the final version the following February.

Other parts of the report did little to settle the dispute. Average weekly hours slipped, which is often how employers cut labour before they cut staff. Revisions, meanwhile, added to the payroll gains for March and April instead of trimming them. A labour market that was cracking would normally show it first in shorter hours and in downward revisions, and in May only the first of those appeared.

Where the precedents point

Turning points are where the payroll count has gone most wrong. The benchmark published in early 2010 cut hundreds of thousands of jobs from the payroll tally for the year to March 2009, long after policymakers had made their calls on the preliminary numbers. Household data, for all their noise, include the unemployment rate, and a 0.3-point jump in a single month is the kind of move that has tended to come early in a slowdown. It remains well short of the half-point rise in the three-month average that has historically signalled a recession, so this is a warning rather than a verdict.

This is also the second time in a year that the surveys have parted company. For much of last year household employment barely grew while payrolls climbed month after month. In December researchers at the Philadelphia Fed, working from state unemployment-insurance records, argued that job gains in the second quarter of last year had been far smaller than first reported. The official benchmark that will settle that argument is part of this summer's preliminary revision, so one of the two surveys may have been misleading for a year without anyone yet knowing which.

The strongest objection is statistical. The household sample is small, and its margin of error on monthly employment changes is several times that of the payroll survey, so a single month of job losses tells little on its own. Economists have long leaned on the payroll figure for that reason. The objection holds for one month's data. It is weaker against the combination of a rising jobless rate, a model known to overshoot at turning points, and a Fed that has already raised rates by five percentage points since March 2022.

Markets pick the bigger number

Traders chose the payroll figure. Treasury yields rose after the release, led by the two-year note, the maturity most sensitive to Fed policy, and futures leaned further towards another increase by July while still treating a June pause as the likelier outcome. Shares rallied too, helped by relief that Congress had just passed the bill suspending the debt ceiling. The division of gains and losses follows from which survey proves right. Floating-rate borrowers and holders of long-dated bonds lose if the payroll reading wins and rates stay higher for longer, while workers gain, since a market adding jobs at this pace leaves employers little room to hold down pay.

What the Fed will weigh

The Fed meets on 13-14 June, and futures markets had been leaning towards a pause even before the report. Philip Jefferson, a Fed governor, said this week that holding rates at the next meeting would give officials more time to assess the data. Revisions to May's payrolls arrive with the June and July reports, and wage growth, the figure most closely linked to services inflation, showed no acceleration in May. A second month of household weakness, or downward revisions to the payroll gains, would tilt the argument towards the pessimists.

The May consumer price report, due on the first day of that meeting, may matter more than either survey. Officials have said they care about the labour market mainly for what it implies about inflation, and a firm core reading would let them discount the household data as noise. A soft one would give the doves the cover they need.

A hiring boom in one survey and job losses in the other suggest a labour market that is weaker underneath than the headline shows. The Fed should not bet heavily on either figure until the revisions arrive, and the most likely result is a pause in June with the July decision left hostage to the next two reports.