One might think that yours truly, who has been nagging the establishment press for years over its blind acceptance of seasonally adjusted data in government economic and employment reports, would be pleased to see that the Associated Press's Christopher Rugaber finally got around to making such adjustments the primary focus of his final report on the most recently released unemployment claims numbers on Thursday. His story's headline at the AP's national site even noted that "Seasonal adjustments to economic data can mislead."
That's fine, but it's not yesterday's full story. Rugaber noted that Thursday's report from the Department of Labor (DOL) -- that 350,000 initial jobless claims were filed after seasonal adjustment -- was influenced by the relatively light level of summer shutdown-related layoffs in the auto industry. But he totally and all too conveniently missed the fact that this year's number looked better after seasonal adjustment than last year's comparable week primarily because, as will be seen later, this year's seasonal adjustment factor was so inexplicably different. First, some excerpts from Rugaber's report:
Seasonal adjustments to economic data can mislead
At first, it seemed like cause to cheer: The number of people who sought unemployment benefits last week tumbled to a four-year low, the Labor Department said Thursday.
Did that mean the job market is surging back to health? Is the unemployment rate about to sink?
Not necessarily. Turns out, last week's drop in unemployment applications reflected mainly a strengthening auto industry - and the Labor Department's imperfect means of calculating unemployment applications. Economists who analyze the job market cautioned against reading too much into Thursday's data.
The problem involves the department's effort to account for seasonal factors in its calculations. It adjusts the data to factor in such annual trends as holiday hiring by retailers and summer layoffs by automakers. Without such adjustments, these seasonal events would distort the health of the job market.
No Chris, without such adjustments, people would have to dig into the raw data and interpret what it means in historical perspective (e.g., compared the same period last year and two years ago, etc.) to come up with an interpretation of the data. Seasonal adjustment is an often effective but by no means infallible attempt to shortcut that process. It doesn't eliminate the likelihood that the raw data will tell you more if you take the time to look at and evaluate it.
Often, an unexpected event will throw off the seasonal adjustments. And economists and investors who try to interpret what the data signify are left scratching their heads.
This spring, for example, auto companies decided to close fewer factories than they normally do in summer. They kept those plants open to keep up with rising demand for cars. Normally, a summer lull in production prompts many temporary plant closings and layoffs.
Ford Motor Co. said in May that it would reduce its usual two-week closing to only one week. And Chrysler announced in spring that it was canceling its normal two-week shutdowns at three factories.
The Labor Department's seasonal adjustments had anticipated more temporary auto layoffs than occurred. As a result, the department reported Thursday that applications for unemployment benefits plunged 26,000 last week to a seasonally adjusted 350,000. (Excluding the seasonal adjustment, applications rose 70,000.)
The real question isn't why "the Labor Department's seasonal adjustments had anticipated more temporary auto layoffs than occurred." It's why it anticipated more layoffs this year than it did last year. For some reason, even though both the week ended July 9, 2011 and the week ended July 7, 2012 contained the Fourth of July holiday (in 2011 it was on Monday; this year it was on Wednesday), it changed the seasonal adjustment factor from 115.7% last year to 125.8% this year (verifiable by join to this interactive DOL web page). This means that it thought that this year's layoffs would ordinarily be expected to be almost 26% higher than the average of all weeks during the year, compared to its assumption that they would be just shy of 16% higher than a full-year weekly average last year.
As I pointed out yesterday at my home blog, the effect of this change on Thursday's reported seasonally adjusted result was huge, and in a downward direction:
- This year’s math (week ended July 7, 2012) — 439,743 actual claims (before the virtually inevitable upward revision next week) divided by 1.258 equals Thursday's seasonally adjusted 350,000 (rounded).
- Last year’s math (week ended July 9, 2011)— 473,963 actual claims divided by 1.157 led to a seasonally adjusted result of 410,000.
- But if DOL hadn't changed its adjustment factor by almost 9% for reasons which escape me (Rugaber noted that "The Labor Department did not respond to requests for comment"), this year math would have been 439,743 raw claims divided by 1.157, yielding a result of 380,000.
So even before considering specific industries, Thursday's seasonally adjusted result would have come in slightly worse than the previous week's (upwardly revised, of course) 376,000, and only about 8% below last year's awful comparable week value of 410,000.
If this year's seasonal conversion done consistent with last year's is slightly higher only because of the carmakers keeping their production going, the rest of the economy's labor market would appear to be about as bad now as it was a year ago -- and significantly worse than it was just a week ago.
Future weeks will tell the tale, but one clue that there is reason for grave concern can be had by comparing the week-to-week change in raw claims from last year to this year:
- Last year (from the week ended July 2 to the one ended July 9), claims increased by over 48,300 from 425,640 to 473,943.
- This year (from the week ended June 30 to the one ended July 7), claims increased by just under 70,000 from 369,772 to 439,743.
Again, if this year's increase would have been much higher except for auto industry layoffs, as everyone seems to be saying, things will get really ugly in the coming weeks when the industry's traditional but unobserved shutdown period ends in a week or two.
So even though it was nice that Christopher Rugaber and his wire service finally acknowledged the pitfalls of blind reliance on seasonally adjusted, they missed the big story, which was the big year-over-year change in the adjustment factor. The fact that July 4 was on a Wednesday this year and a Monday last year -- the only conceivable justification -- doesn't seem seem satisfactory. AP was not alone in failing to pick up the factor change; Bloomberg didn't. Nor did Reuters, which headlined and highlighted how this week's strangely calculated 350,000 represents a four-year low.
Readers can judge for themselves whether they believe AP's failure to note the factor change while supposedly presenting a comprehensive report on why Thursday's claims number came in as it did occurred because of ignorance or because of the desire to create a "hey, look over there!" distraction.
Cross-posted at BizzyBlog.com.