Friday, February 3, 2012

Did seasonal adjustments distort the January Job Report?

This post is not designed to be cynical or pessimistic, but to explore possible reasons that the report overestimated the strength of the job market.  Most economic data, including employment data, are seasonally adjusted.  As part of a recent analysis of the job market relative to GDP, Mark Vitner of Wells Fargo examined how seasonal adjustments may be misleading.  It should be noted that this was not written as a way to explain away this morning's job report, but was released on January 31.  Here are a few of his key points:

"The 2007-2009 recession was the deepest recession since the Great Depression, when considering peak-to-trough declines in real GDP. Moreover, some of the largest declines in employment, as well as the largest increases in the unemployment rate, occurred during the fall and early winter months of 2009-2010, with nonfarm payrolls reaching a bottom in February 2010...  This occurrence has likely set the bar very low for employment gains during the winter months in such a way that even normal hiring gains related to the holiday shopping season are showing up as seasonally adjusted employment gains, leading to exaggerated estimates of employment growth.

The seasonal factors used to adjust the unemployment rate were also impacted by the severity of the 2007-2009 recession. The result of this amplitude change is that the seasonal adjustment process has tended to exacerbate downward movements in the unemployment rate during the fall and winter months, while exerting upward pressure on the unemployment rate in the spring and summer months."

In essence, hiring patterns in recent years are used to estimate seasonal adjustments.  Given that some of the worst job losses during the Great Recession took place in the winter (Dec 2008-March 2009), the seasonal adjustments for those months may be biased, resulting in an overestimate of seasonally-adjusted employment growth for those months.  How much of a difference did this make?  If one seasonally adjusts last month's data using the average seasonal adjustment for January for the five years prior to the Great Recession, there would have been 40,000 fewer jobs reported (resulting in an increase of 203,000 jobs as opposed to 243,000).  Thus, a significant portion of the upward surprise appears to be linked to seasonal adjustments that may be biased.

Update (Saturday):

Here’s the math underlying the analysis. The average seasonal adjustment for January data between Jan 2002 and Jan 2009 was about 1.583% (these adjustments were not affected by the financial crisis).  That is, non-seasonally adjusted data are adjusted upward by 1.58% to get the seasonally-adjusted numbers).  Meanwhile, the January 2012 was adjusted upward by about 1.65%.  What difference does that make?
Non-seasonally adjusted employment for January 2012 was 130.263 million and the seasonally-adjusted number was 132.409 million. If one applies the average adjustment from 200522009 (1.58%), the seasonally-adjusted number for January 2012 is 132.321 million, 40,000 less than the official figures.  Thus, it's still a strong report, but probably reflects some bias from the new seasonal adjustment factors (note: this does not account for issues related to warmer than normal temperatures in Jan 2012, etc.).