The job numbers, rewritten again

The January jobs report came with bad news about the past. The Bureau of Labor Statistics published its final annual benchmark revision, wiping hundreds of thousands of jobs off the historical record and redrawing the path of the labour market. The size of the markdown was no surprise to anyone who read the preliminary estimate in September, which put the overstatement of payrolls in the year to March 2025 at 911,000, the largest such preliminary revision on record.

It was the second big markdown in a row. A year earlier the preliminary estimate for the year to March 2024 had cut 818,000 jobs, at the time the largest since 2009. Two consecutive overstatements of that size point to something systematic in the method rather than bad luck in a single year.

How the count goes wrong

The monthly payroll figure comes from a survey of employers, and it is only an estimate of a much larger count. Once a year the bureau lines it up against the Quarterly Census of Employment and Wages, which draws on the unemployment-insurance tax records filed by nearly every employer in the country. The gap between the two is the benchmark revision. Because the census data arrive months late, the monthly survey has to guess in the meantime how many jobs are created by businesses too new to be sampled and lost by those that have closed, and that guess is the birth-death model.

Response rates make the guessing harder. Fewer employers return the survey on time than a decade ago, so the first estimate for each month rests on a thinner sample and leans more on the model to fill the gaps. The first two monthly revisions catch some of the error as late responses arrive. The benchmark catches the rest, a year or more after policymakers have acted on the original figure.

Two years of hiring, marked down

The revision shows that the labour market looked stronger over the past two years than it was. The birth-death model misjudged how many businesses were opening and closing, and establishment surveys came in late. The lower numbers show that non-farm hiring slowed to stall speed much earlier than policymakers realised, in an economy where companies had stopped adding staff and hours worked were falling. Such models are built on the recent past, and they tend to overstate hiring at turning points, when new firms stop appearing at their old rate.

That pattern has form. In 2008 the first estimates of monthly job losses were far smaller than the revised figures, and the Fed spent much of that year reacting to a labour market that was weakening faster than the data showed. The error runs the other way in recoveries, when the model is slow to count new firms, so revisions tend to deepen whatever mistake policymakers were already making.

A year of damaged statistics

The revision lands on a statistical agency that has had a bruising year. The president dismissed the bureau's commissioner in August, hours after a weak July report that included large downward revisions to earlier months. The government shutdown in the autumn then interrupted data collection, leaving gaps in the household survey that cannot be filled after the fact. Each episode made it harder for the Fed to read the labour market in real time, and the benchmark shows that the real-time picture it relied on was too flattering.

The details of that July report explain the reaction. It showed only 73,000 jobs added and cut 258,000 from the estimates for May and June, and the commissioner, Erika McEntarfer, was gone the same afternoon. The shutdown that ran from 1 October to 12 November lasted 43 days, the longest on record, and left the bureau reconstructing what it could. Firing the messenger did nothing to change the numbers, as this revision shows, but it did give markets a reason to question every release since.

Room for Warsh to cut

For the Fed's incoming leadership, the revised figures make a clear case for cutting rates faster. The labour market has less of a cushion than the hawks assumed. Kevin Warsh, nominated to succeed Jerome Powell, has already argued for lower rates, and revised data showing a weaker job market hand him an argument that does not depend on politics. Lower rates would help most the sectors that borrow to hire and build, such as housebuilders and small firms, which have been squeezed hardest by high borrowing costs.

The committee has already moved some of the way. It cut three times between September and December, taking the federal funds range to 3.50–3.75 per cent, citing downside risks to employment each time. The benchmark suggests those risks were larger than the committee could see when it voted, and it weakens the case of members who argued at the December meeting that the easing had gone far enough.

What the hawks say

The strongest objection is that fewer jobs do not mean a weaker labour market if fewer workers are looking for them. With immigration sharply lower, the number of new jobs needed each month to keep unemployment steady has fallen, so modest payroll gains can coexist with a stable jobless rate. Inflation also remains above the 2 per cent target, partly because tariffs have pushed up goods prices, and hawks on the committee will argue that cutting quickly risks cementing it. That argument weakens if the unemployment rate starts to climb, which is why it is the number to watch.

Numbers that settle it

The unemployment rate has a useful property here: it comes from the household survey, not from employer payrolls, so it does not depend on the birth-death model. A rise in the three-month average of half a percentage point above its low of the previous year has preceded every recession since the 1970s, a relationship known as the Sahm rule, and it briefly triggered in the summer of 2024 without a downturn following. The other signals are weekly jobless claims, the job openings survey, and the quarterly census releases that will show whether the overstatement has continued beyond March 2025. The Fed kept policy too tight against job gains that turned out not to exist. The revisions leave it little choice but to cut rates repeatedly in 2026, probably starting soon after Warsh takes the chair.