The BLS's monthly employment numbers come out tomorrow, and economists expect a strong showing, in the neighborhood of 195,000 jobs. The top-line numbers in those BLS surveys are always seasonally adjusted - that is, they're corrected to "remove the influences of predictable seasonable patterns," like weather changes, holidays and school schedules.
Seasonal adjustment is incredibly important - it's how economists sort the noise from the signal in the monthly jobs numbers, and it ultimately affects how well policymakers are able to respond to shifts in the economy. But there's an awful lot of noise in those numbers, and some economists think we should pay more attention to that filtering process with an eye toward improving it. "Everyone just ignores [seasonal adjustment] as if it's data that comes directly from God," says Justin Wolfers of the Brookings Institution. "It's not a focus of markets or journalists, but it should be."
To get a sense of the impact of the adjustment process, check out the interactive chart below. It's currently showing the seasonally-adjusted numbers that we're all familiar with. The deep trough of the recession stands out, as does the relatively flat place of job growth since. But click on the "Non-seasonally adjusted" tab to see what the unadjusted monthly data looks like.That orderly trendline is gone, replaced by a series of deep peaks and troughs. Note the magnitude of changes - we're looking at literally millions of jobs added or removed each month."The ups and downs of the seasonal cycle are enormous - as big as major recessions and major booms," Wolfers explains.
Two major trends stand out. First, there's the massive drop that happens every January. This is due, at least in part, to employers letting go of temporary workers after the holiday season. Second, a smaller downward spike happens at the end of the school year in July, as teachers go on summer break. For tomorrow's jobs release, note that in recent years March data typically comes in around +800,000 unadjusted. This number then gets adjusted down to the neighborhood of +200,000 jobs or so.
You can see in the chart that January 2014's loss of 2.8 million jobs turned into a net gain of 129,000 jobs after adjustment. The process by which that happens is, of course, quite complicated. To make a long story short (and overly simple), BLS economists take the latest four years of jobs data and feed it into a computer program with a positively terrifying name: "The X-12 ARIMA Seasonal Adjustment Program." The program massages the current month's numbers according to past trends, and out pops that tidily-adjusted figure that everyone obsesses over.
Given the degree of adjustment that the jobs numbers are subject to, you might wonder why we obsess over them as much as we do. According to Jonathan Wright, an economist at Johns Hopkins, the sampling error in payroll numbers is comparatively large - on the order of +/- 56,000 jobs. And even minor tweaks to the seasonal adjustment algorithms would change the numbers by tens of thousands each month. Given this context it seems silly to fret when the monthly numbers miss expectations by a few thousand jobs - but we do it anyway.
"Policy makers, the press, and financial markets focus far too much on small changes in the month-over-month number," Wright says. "They often make a big deal of whether the number is (say) 30,000 above or below expectations. But there are several different reasonable ways of doing seasonal adjustment, and differences among these are often more than 30k."
For his part, Wright has made his own tweaks to the adjustment methodology to come up with numbers that he says are more accurate, filter out more noise, and are less prone to revision than what the BLS currently uses. Again, long story short, he basically extends the comparison period from the four years used by the BLS to six years. You can see the differences in the chart below.The differences are small, but again - the financial press has gone into a swoon over much, much less. Wolfers is a fan of Wright's methodology, which was presented at the Brookings Papers on Economic Activity last year. "It's a better representation of what's actually happening in the economy," he says. Richard Tiller, a statistician at the Bureau of Labor Statistics, is more cautious, but he notes that an internal BLS study done in 2007 reached essentially the same conclusions, and that the BLS "continues to work on improving the methodology."
Bottom line: that methodology matters. According to Wright and Wolfers the small monthly fluctuations matter much less than the accuracy of the filters used to arrive at those numbers. Wright says that "distortions in [seasonal trends] coming from the Great Recession may have influenced perceptions of the strength of the economy in spring 2010 and 2011, and may even have affected Fed policy." He thinks that his methodology would have led to less distortion, and a more accurate depiction of the health of the economy immediately following the recession.
Wolfers agrees that "there is a clear case we may have been led astray during the recovery. The data was overstating the strength of the recovery, which led to policymakers holding back when they should have taken stronger measures."