Next week, when the Bureau of Labor Statistics (BLS) releases the December employment report (and for January as well), I suggest following the non-seasonally adjusted data, rather than the seasonally adjusted data. The normal seasonal patterns are not present, and the seasonally adjusted data will paint a slightly more benign picture of the job loss that will be reported in the nonfarm payroll.
An example: seasonal adjustments cannot predict statistical anomalies in unemployment claims report
The Department of Labor released a shockingly strong weekly initial unemployment claims report last Thursday: In the week ending Dec. 27, the advance figure for seasonally adjusted initial claims was 492,000, a decrease of 94,000 from the previous week's unrevised figure of 586,000. The 4-week moving average was 552,250, a decrease of 5,750 from the previous week's unrevised average of 558,000.
The advance seasonally adjusted insured unemployment rate was 3.4 percent for the week ending Dec. 20, an increase of 0.1 percentage point from the prior week's unrevised rate of 3.3 percent. The labor market is improving – at least on a weekly basis, right? Wrong. The insured unemployment rate rose to 3.4% over the week, a signal that the national unemployment – currently at 6.7% - rate will rise.
The weekly claims fell simply because of the statistical process the
The chart illustrates the seasonally adjusted and non-seasonally adjusted data since Christmas of 2003, where I highlighted the comparable report for each of the preceding five years. Every year the number of claims filed in the last week of December rises, and on average, the NSA data rises by 75,233.
The seasonal adjustments extract this cyclical pattern (always rises in December) from the data, leaving only the trend that exceeds (or falls short of) the seasonalities. New claims filed beyond what normally happens in December (the seasonal adjustment) signals a weaker-than-normal labor market, while claims filed below what normally happens in December signals a stronger-than-normal labor market.
On December 27, 2008, the NSA number of claims filed rose by just 1,892, which is 73,331 below the average. The seasonal adjustment process sees this as a good thing, and reports a stronger-than-normal labor market, -94,000 new claims filed. But what happened on December 27, 2008 was clearly not normal, as every other indicator signals an contracting labor market.
It is more plausible that a negligible amount of seasonal hiring actually occurred in November and December. Therefore, the last week of December saw very few firings, and a negligible number of initial unemployment claims were filed. Seasonal adjustments would miss such a statistical anomaly.
Beware of the seasonal adjustments in this cycle; this applies to the Bureau of Labor Statistics employment report as well.
The chart illustrates nonfarm payroll since 2004. Like clockwork the seasonal pattern is this: firms hire in October and November for the holiday season; they start firing in December; and then they slash jobs in January.
This year, firms didn't hire in October nor in November - nonfarm payroll fell by 320,000 and 533,000, respectively. Therefore, the seasonal firing pattern through December and January will not exist and the seasonal adjustements will not apply.
In December and January, the seasonal adjusted series will show a stronger labor market than actually exists. Holiday workers will not be fired in December and January because they were never hired in October and November, and the seasonal job loss will be less than the non-seasonal job loss. The non-seasonally adjusted numbers will paint a more accurate picture of the labor market.