The jobs report that contradicts itself

The October jobs report looked like a portrait of two different economies. According to the establishment survey of employers, non-farm payrolls rose by a solid 261,000, comfortably beating forecasts and suggesting companies still want to hire. The household survey, from which the unemployment rate is calculated, showed the number of people in work falling by 328,000, and the jobless rate rose by two-tenths of a percentage point to 3.7 per cent. When the two main measures of American employment differ by nearly six hundred thousand workers in a single month, investors need to look past the headlines at how the surveys are built.

The gap matters for more than statisticians. It goes to the heart of the Federal Reserve's thinking. If payrolls really are growing by more than a quarter of a million a month, wage growth will keep services inflation going and the Fed will have to push rates well above 5 per cent. If the household survey is right and jobs are being lost, the economy is cooling much faster than the hawks think.

A report landing on a hawkish week

The numbers arrived two days after the Fed raised its policy rate by 75 basis points for the fourth meeting running, to a range of 3.75 to 4.00 per cent. Chairman Jerome Powell used his press conference to say that rates would probably peak higher than officials had projected in September, and that it was premature to talk of a pause. Markets had been hoping the labour market would give him a reason to soften. The payroll figure did not.

Other indicators also point to a tight market. Figures published on 1 November showed job openings rising to 10.7m in September, still well above the number of unemployed people. Average hourly earnings were 4.7 per cent higher than a year earlier in October, a slower pace than in the spring but still too fast for inflation to settle near the Fed's 2 per cent target. It is against this backdrop that the household survey's fall looks odd.

How the two surveys count

The two series differ in what they count and how they sample. The establishment survey counts filled jobs on company books, drawing on about 122,000 businesses and government agencies. A worker with two part-time jobs is counted twice, and the totals are adjusted with a birth-death model that estimates how many new businesses have opened, net of closures. At a time when inflation is pushing people into second jobs to pay the bills, the payroll survey can mistake extra jobs for a growing economy.

The household survey counts people in work, based on interviews with roughly 60,000 households. It picks up the self-employed and farm workers, as well as domestic staff, none of whom appear on company payrolls. Someone with several jobs is counted once. At turning points in the cycle the household survey has tended to move first. Small business owners and contractors often feel slowing demand and higher borrowing costs weeks before big companies formally freeze hiring.

Noise, and how to filter it

The household survey has a well-known weakness. Its sample is about half the size of the payroll survey's in units and far smaller in coverage, so its monthly change carries a margin of error several times as wide. A single month's fall of 328,000 could be statistical noise, and economists routinely warn against reading much into it.

The sensible test is persistence. One month of divergence proves little; three or four months of it, in the same direction, would be hard to dismiss. Analysts can also adjust the household figures to the payroll concept, stripping out the self-employed and counting multiple jobs, which narrows the gap if the difference comes from definitions and leaves it in place if it comes from a real loss of work.

Which number the Fed trusts

The Federal Open Market Committee is in an awkward spot. Officials almost always give priority to the payroll series, citing its bigger sample and lower monthly volatility as reasons to treat it as the better guide to the economy. Solid payroll gains also give them cover to keep tightening, since they can point to strong hiring as the justification for higher borrowing costs.

If the fall in household employment is the first sign of companies pulling back, the Fed risks pushing rates deep into restrictive territory on payroll data that will be revised away a year from now. The birth-death model is known to amplify the cycle, assuming new businesses are being created just as higher borrowing costs are pushing small firms under.

When the revisions come

The payroll survey is eventually checked against a near-complete count of jobs drawn from state unemployment-insurance tax records. The Bureau of Labor Statistics publishes a preliminary estimate of that annual benchmark revision each summer and folds the final version into the data every February. Those revisions have at times moved the level of employment by several hundred thousand, and they are largest near turning points, when the birth-death model is most likely to be wrong.

The costs of an error are lopsided. If the payroll figures are right and the Fed eases too early, it risks a repeat of the 1970s, when inflation came back after each pause. If the household figures are right and the Fed keeps tightening, the damage lands on the workers the household survey sees first: contractors, the self-employed and employees of small firms with bank loans rather than bond financing.

The contradictory report suggests the American labour market is changing direction. Payrolls are being propped up by second jobs and hoarding, and if the household survey is right the payroll figures will probably be revised down next year, well after the Fed has acted on them. Until the establishment survey falls into line with the household data, the Fed will treat the labour market as uncomfortably strong and keep its foot on the brake, most likely through at least two more rate rises before the benchmark revision can tell it whether the brake was needed.