Thursday, February 4, 2010

A Review of the Benchmark Revision and Seasonal Adjustment for January

Tomorrow, the Bureau of Labor Statistics (BLS) is going to publish its employment situation report for January and it will show job losses exceeding 2 million (and quite possibly over 3 million jobs lost). They (the BLS) are going to gloss over this with their seasonal adjustment, which may not only hide these losses, but show a potential gain in jobs for January. Do not be fooled, the seasonal adjustment is just an internal BLS code for fudging the numbers.
The seasonal adjustment to the employment situation report for the January job numbers has averaged +1.62% over the last five years and is used to offset the holiday firing that typically is associated with January (which the BLS also discounts on the hiring side by large negative seasonal adjustments in November and December). That 1.62% may look small, but remember we are dealing with roughly 130 million employed persons, so that 1.62% is a huge adjustment. Since 1999 the average January lost 2.8 million jobs on an unadjusted basis. The entire point of using seasonal adjustments is to eliminate the well-known effects of seasonality on job creation so we can see what the underlying economy is doing (we also subtract a ton of jobs in May and June every year to account for summer hiring). There is nothing wrong with this adjustment and it is based on years of data.

Now, the number to examine in this report will be the adjustment itself. In both November and December, the BLS used higher than normal negative adjustments (ie they subtracted more jobs than normal), which seemed odd to me since holiday hiring was expected to be down last year and last year the January adjustment was +1.535%, so we need to look for an outlier seasonal adjustment before critiquing the data (ie greater than 1.65 or less than 1.535). Any number in that general range (I used the 1.65 since that was the average without last year included) would seem to be inline with past adjustments for seasonal firing, while a number outside that range would need to be accounted for through additional data collection/reasoning. Do not believe the case that the BLS is "covering up" more job losses through its seasonal adjustment, as it is simply not true.

On to the much publicized benchmark revision that will be included with the January Employment Situation Report (like it always is).
Each year, the Current Employment Statistics (CES) survey employment estimates are benchmarked to comprehensive counts of employment for the month of March. These counts are derived from state unemployment insurance (UI) tax records that nearly all employers are required to file.

What this is intended to do is adjust the survey numbers (remember the jobs numbers come from a survey of employers, which is not combined with the unemployment rate that comes from a separate survey of households) to the actual job counts from filings (ie the revision is essentially the actual margin of error for the establishment survey for the year). This year, the benchmark revision is going to be huge, likely coming in at 824,000 jobs that were lost through March of 2009. AS you can see by the table in this link the revisions have been all over the place (ie both positive and negative), which shows there is no inherent bias in the survey towards one side (the data goes back to 1979).

Just to be very clear, the benchmark revision has nothing at all to due with the unemployment rate. These numbers come from two completely different surveys (the establishment survey interviews employers, and the household survey interviews households). The household survey (to which the benchmark revision is not applicable) gives us the unemployment rate and thus the benchmark revision does not imply at all that the unemployment rate should have been different because of the revised establishment survey numbers. Anyone claiming that the benchmark revision means that the unemployment rate was artificially low last year has no clue what they are talking about.

As for tomorrow's actual jobs number, the consensus is for 0 job gains/losses with a range (from surveyed economists) of (40,000) to +75,000 jobs. Keep in mind that the margin of error for a given month is about 100,000, so the multi-month direction of the report is going to be much more important that the actual number (assuming it falls within +/- 100,000 of 0).

Hopefully this was helpful as an educational piece ahead of tomorrow's important employment situation report.