The job market's confusing signals

The unemployment rate rose to 4.6 per cent in November, and the jobs report left economists' models in a muddle. Joblessness jumped even as the establishment survey showed payroll gains, a contradiction that suggests the usual seasonal adjustment broke down after the long government shutdown.

Two surveys, two stories

The monthly jobs report is built from two separate surveys. The establishment survey asks employers how many people are on their payrolls; the household survey asks a sample of families whether their members are working or looking for work, and produces the unemployment rate. The two usually move together over time, but they are collected differently, and the shutdown hit them in different ways. The household survey is also the noisier of the two: it is drawn from a sample of about 60,000 households, so a monthly move of a tenth of a percentage point in the unemployment rate is well within its margin of error.

Employers keep records, so the statisticians could go back after the government reopened and gather October payroll figures late. Those showed a fall in October, driven largely by federal employees who had accepted deferred resignation offers and left the payroll at the end of September, followed by a modest gain in November. Households keep no such records. The October household survey was never conducted, so there is no official unemployment rate for that month at all, and November's figure had to be compared directly with September's.

A gap in the series

Time-series models such as the X-13ARIMA filter used by federal agencies need an unbroken run of monthly data to calculate seasonal factors. The weeks without data in October threw the filter off. It treated hiring restarts after the shutdown as a pick-up in the underlying trend, and it overstated unemployment in the household survey.

The mechanics explain why a single missing month does so much damage. Seasonal adjustment works by estimating the normal pattern for each calendar month from several years of history, using moving averages that run forwards and backwards through the data, and then removing that pattern. The ends of the series are always the least reliable, because the averages have to be extended with forecasts. A hole near the end compounds the problem: the filter has to guess at the missing value and then use that guess to estimate the seasonal factors for the months around it. Every late-year month, when hiring for the holidays and year-end layoffs create large seasonal swings, is affected.

The pandemic precedent

The statistical agencies have faced a version of this before. In the spring of 2020 lockdowns produced movements so large that standard seasonal adjustment would have read them as a change in the seasonal pattern itself. The Bureau of Labor Statistics responded by treating those months as outliers, so that they would not distort the seasonal factors for later years. The bureau also acknowledged at the time that some workers had been misclassified in the household survey, which understated unemployment. The lesson from that episode is that the published figures around a disruption can be misleading for several months, and that revisions tend to arrive only once enough clean data have accumulated.

The labour market is loosening

Strip out the statistical noise and the signal is plain enough. The American labour market is loosening for real. The rise to 4.6 per cent is coming from permanent job losses and longer spells out of work, and few of the newly unemployed quit by choice.

Other evidence points the same way. Continuing claims for unemployment benefits, which track how many people remain on benefits after their initial claim, have stayed close to their highest levels since 2021, a sign that people who lose jobs are taking longer to find new ones. Hiring rates in the government's job openings survey have been low for more than a year. Companies have been cautious about adding staff but also slow to cut, a pattern economists describe as low hiring and low firing. In that kind of market, the unemployment rate can rise steadily even without mass layoffs, simply because those who do lose work struggle to get back in.

The case against alarm

The optimistic reading deserves a fair statement. Payrolls are still growing, if slowly, and the unemployment rate remains low by historical standards. Some of the rise may reflect people entering the labour force to look for work, which is a sign of confidence rather than distress. Slower immigration has also reduced the number of new jobs the economy needs to create each month to keep unemployment steady, so modest payroll gains may be enough.

That reading has merit, but it does not explain the composition of the rise. An increase driven by new entrants looks different in the data from one driven by permanent job losers, and the November figures point to the latter. The immigration argument cuts both ways: if fewer new jobs are needed, the fact that unemployment is still rising suggests demand for workers is weaker than the payroll numbers alone imply.

What the Fed will make of it

The Federal Reserve cut its policy rate in December to a range of 3.50 to 3.75 per cent, and officials signalled that they expect little further easing in the near term. A labour market that keeps loosening would test that stance. The missing October reading also complicates a popular recession gauge, the Sahm rule, which compares the three-month average unemployment rate with its low over the previous year, since the average now has a hole in it.

The indicators that will settle the argument are the coming monthly reports, which should give the seasonal filter clean data again; the annual benchmark revision to payrolls published with the January report in early February; and weekly jobless claims, which were collected throughout the shutdown and are the most continuous series available.

November's numbers show the limits of statistical models when data go missing. The direction is clear all the same: slack is building, and the jobless rate will probably keep drifting up into the spring. If it reaches the high 4 per cent range by March, the Fed will come under pressure to resume cutting sooner than its projections imply.