Hazards and Head Fakes in Today's Employment Report
The October employment report seems, at first glance, to tell a very strong story, and it could be good news -- perhaps very good news. But given the asterisk next to today's numbers, anything more than cautious optimism is getting ahead of the data.
The payrolls survey suggests that the U.S. economy added 204,000 jobs in October, and upward revisions to previously released estimates for August and September added a further 60,000 jobs. The average pace for the past three months, which usually gives a better sense of the underlying trend, suggests that the U.S. economy is adding slightly more than 200,000 jobs a month. If that continues, the U.S. will make steady progress toward reducing joblessness.
But this report was complicated by the government shutdown, so it's important to be even more careful than usual in reading the tea leaves. At a purely statistical level, how did the shutdown distort the numbers? And at an economic level, did the shutdown mask even greater underlying strength?
Fair warning: This will get wonky. It will also lead to two conclusions. First, the payroll numbers look good, but temper your optimism: The margin of error is simply much larger than usual. And second, the household survey tells a far more confusing story -- providing ammunition to both optimists and pessimists.
If you don’t want to get down into the weeds, stop reading now, and come back next month for a cleaner reading of the state of the labor market.
Let’s start with the statistical issues. How were hundreds of thousands of furloughed government workers counted? Because they eventually were paid, the payrolls report considered them employed. By contrast, in the household survey, those who weren’t working due to a temporary layoff were counted as unemployed.
This suggests that the payrolls number may be a more accurate guide to what’s happening in the economy. Still, some issues remain.
It's unlikely that the many government contractors who were also laid off during the shutdown were paid while off work. That said, those who are paid monthly or bi-weekly probably got paychecks for the period covering the reference week, so only contractors who are paid weekly would be missing. So perhaps this report slightly understates job growth.
Second, the Bureau of Labor Statistics released this report a week later than normal, giving it more time to collect payroll reports from employers. As a result, the response rate in October was higher than it has ever been. There’s some reason to believe that this may lead to an upward bias in the report. Why? Revisions, which typically reflect firms that are late responders, are usually pro-cyclical. The higher response rate, then, may have captured more firms in a hiring phase than usual.
Phrased alternatively: In a normal recovery, it’s reasonable to expect positive revisions to the initial jobs numbers; those revisions may already be baked into today’s report.
As for the household survey, it shows the unemployment rate for October was 7.3 percent, a slight rise from 7.2 percent. Believe it or not, this is good news. It’s just unclear how good.
If furloughed workers responded appropriately to the household survey, they would report being on temporary layoff, and hence counted as unemployed. This more than accounts for the rise in unemployment. Indeed, 13.4 percent of the unemployed reported being on temporary layoff in October, well above the average over the last year of 9.4 percent. That is, around 4 percent of the unemployed were on shutdown-related furloughs. (Only about half of these extra layoffs were among federal workers, suggesting a very real effect on government contractors.)
If this rise in temporary layoffs hadn't happened, the unemployment rate would have been 7 percent, a substantial improvement.
But that isn't the whole story. A separate BLS memo says that, because people don't always respond to surveys as statisticians expect, about 200,000 furloughed workers were misclassified as employed rather than laid off. Their mistaken answers actually make the data a bit more accurate.
It isn't all good news, though. Economists draw a distinction between the unemployment rate dropping because of the happy event of the jobless finding work, versus a decline that results when the unemployed give up looking. This report very much reflects the latter, with the labor force participation rate declining from 63.2 percent to 62.8 percent.
According to the household survey, total employment fell by 735,000 (although the corrections above suggest that we should add back in about 450,000 people on temporary layoff, yielding a corrected decline of 285,000). Declining employment, even in the noisy household survey, is a bad signal.
What, then, is the underlying rate of employment growth in the economy? Typically, I suggest putting an 80 percent weight on the payrolls survey and only a 20 percent weight on the household survey. Plugging in my corrected number for the household survey suggests that overall employment growth in October was a disappointing 106,000.
There’s yet one more asterisk to add. The household survey typically asks people about their activities last week. But because survey staff were furloughed, they couldn't do their work on the usual schedule and had to go back a week later to ask people about their activities two weeks earlier. A key lesson of survey research is that even small changes in methodologies can have big effects, which suggests even more caution in interpreting the household survey.
Let’s step back out of the weeds. The payrolls survey suggests robust jobs growth and, if anything, may understate strength. The household survey says that the unemployment rate would have fallen if we counted the furloughed as employed, although it also reports weak employment growth. Depending on what signals you emphasize, it’s reasonable to conclude anything -- from a moderately weak to a moderately strong jobs picture.
And what was the effect of the shutdown? To some, the headline numbers suggest that the shutdown didn’t matter. That’s too hasty a conclusion. We don’t know what would have happened if it weren’t for the shutdown. We don’t know how much of our current optimism is due to statistical noise. We aren't sure how to reconcile the competing signals given by the data. And we don’t know how much current numbers reflect temporary distortions.
What's more, the shutdown may yet have a bigger effect on November’s data. The October data measure the first part of October, when the shutdown was ongoing. To the extent that employers pulled back on hiring, it seems plausible it came after the mid-shutdown survey week, and so we may yet see a weaker number next month.
My advice to Federal Reserve policy makers considering pulling back on stimulative bond purchases is simpler: Do nothing. Ignore this report. Sit on your hands. Wait for the November data.
(Justin Wolfers is a Bloomberg View columnist. Follow him on Twitter.)