What Today's Unemployment Report Means
by Nate Silver
Last month, I boldly proclaimed, based on a fairly simple statistical model, that there was only about a 1 in 3 chance that unemployment would hit 10.0 percent. People who took the long side of that bet are looking a bit smarter today after a mediocre employment situation report that showed unemployment rising to 9.7 percent last month. The model had expected an increase to 9.5 percent.
First, the good news. (And no, I don’t feel any need to be politically correct by using scare quotes around the term “good”). The establishment survey – what is usually reported as the “payrolls” number – showed a decline of 216,000 jobs. That was slightly better than consensus expectations, and slightly worse than my model’s expectations, but broadly in line with what everyone was projecting. It’s also the best figure since last August, when the economy lost 175,000 jobs. (There were also some upward revisions to the job loss numbers for June and July, although they were fairly trivial.)
The problem is that the other of the two surveys the government conducts every month – the household survey – showed a decline of 392,000 employed persons. It also showed a small increase in the size of the labor force (which is not unexpected). The unemployment rate is keyed off the household survey – and so it jumped to 9.7 (actually 9.66) percent.
As I wrote last month, when the two surveys conflict with one another, the establishment survey is generally the more reliable one by a fair margin. That's why Wall Street is trying up on today's news -- it trusts the establishment survey a lot more, which is consistent with the story that the economy is recovering. But, both surveys provide some useful information and the household data should not be dismissed out of hand.
Our model still shows unemployment avoiding the 10 percent threshold – but barely – peaking at 9.91 percent in November. Since there’s a definite margin of error in these calculations, and a de facto margin of error in the unemployment rate figures as actually reported by the government, the model probably implies that there’s a 50:50 chance, or a little higher, that unemployment will indeed be reported at 10.0 percent or higher in at least one month.
There are other models out there that are a lot, lot more sophisticated than mine, and most of them show an unemployment peak that is both higher (at least 10 percent) and later (occurring early in 2010 rather than late in 2009) than mine does. On the other hand, the parameters of this particular economic slump are unusual. Recent recessions have been proceeded by "jobless recoveries". But, recent recessions have been fairly shallow whereas this one was (is?) quite deep, and generally speaking the deeper the recession the brisker the recovery in jobs figures. So there's a lot of uncertainty in everyone's estimates. The bottom line: we’re now somewhere in between the rather pessimistic picture implied by June’s employment report, and the more optimistic one implied by July’s.