|Total Non-Farm Employment |
(No Seasonal Adjustment)
Birth-death Adjustments: A second adjustment by the BLS is an attempt to estimate how many new businesses have been formed, but are not yet part of the list used by the survey team. The BLS makes assumptions about how many businesses are created in different industries, in different months.
Bureaucracy rather than conspiracy: Seasonal adjustments are not "fudging". They can also make the numbers look worse. Private business do this all the time. Birth/Death adjustments can make numbers look good (or bad) for months or years, if we're in a secular "new normal". A private investor gathering data, seeing every month being revised down by 2000, may come up with an adjustment: something has changed, we don't know what it is... but, lets start subtracting 2,000 right off the bat until this does not happen any more. Bureaucracies don't work that way, least of all government ones.
Married to models: It is not just bureaucrats. People can be wedded to their "models". Paul Samuelson's famous textbook showed the Soviet Union would surpass the U.S. When his past forecast for commie success was too high, he did not change his model. Instead, he assumed the growth revolution had simply been delayed. So, his revised text books continued to predict Soviet economic growth shooting up in the future.
A literal lie: Taken literally, the BLS does lie when is says (e.g. July 2012) "Total non-farm payroll employment rose by 163,000 in July". It ought to say "Seasonally-adjusted total...". For instance, an accurate description of that July report would have said: "Total non-farm payroll fell by 1.2 million, but adjusted for seasonality, it rose by 163,000". (That is not a typo.)
This time is different: The two adjustments above work reasonably well when times are close to "normal". However, adjustments fail when there are significant one-off influences. Particularly warm or cold weather that moves shopping patterns by a few weeks this way or that; or, a government cash-for-clunkers program that induces purchase that would otherwise have delayed for a few months; or, historically low new-business formation.
Terrible at turning points: At turning points (when you would need them most), the BLS's adjustments turn out to be very wrong in retrospect. From one of John Hussman's superb weekly commentaries: For example, if you look at the originally reported data for May through August 1990, you’ll see 480,000 total jobs created (see the October 1990 vintage in Archival Federal Reserve Economic Data). But if you look at the revised data as it stands today, you’ll see a loss of 81,000 jobs for the same period. Look at January through April 2001, at the start of that recession. The vintage data shows a total gain of 105,000 jobs during those months, while the revised data now shows a loss of 262,000 jobs. Fast forward to February through May 2008, and though you’ll actually see an originally-reported job loss during that period of 248,000 jobs, the revised figures are still dismal in comparison, now reported at a loss of 577,000 jobs for the same period. As other good economic analysts have recognized, economic time series tend to be revised after-the-fact, with upward revisions in periods just before the recession begins, and downward revisions in periods just after the recession begins. I continue to believe that the U.S. joined an unfolding global recession, most probably in June of this year. [emphasis added]
Summary: BLS data is fine when one is studying history; but, any one or two months data does not tell us much if anything. Nietzsche warns: "Danger, disquiet, anxiety attend the unknown –the first instinct is to eliminate these distressing states. First principle: any explanation is better than none… The cause-creating drive is thus conditioned and excited by the feeling of fear!…” Don't grasp at the data merely because you want to grasp at something. Acknowledged ignorance is sometimes closer to reality.