It's in the Footnotes

THE STUNNING BIT OF NEWS IN THE SEPTEMBER EM­ ployment report, released Friday by the Bureau of Labor Statistics (BLS), was not in the headline numbers but in one of the footnotes: The B LS made the confidence-shaking announcement that it would soon make the largest upward revision to payroll employment in recent history, if not ever.

But first the headline numbers themselves, which con­ tained a revisionist mare's nest of their own.

The unemployment rate was essentially unchanged in September, at 4.6%-01' as the markets prefer, it ''fell'' by a statistically meaningless tenth-of-a­ percentage point from the month before. The disappointment came with the payroll employment nun1ber, which rose only 51,000 in September, well below the consensus expectation of 120,000. On the other hand, upward revisions to July and August were also reported, with the net result that payroll gains over the past three months averaged 120,000 per month almost exactly.

And here's the irony: The September increase in payroll employment probably would have come in at lOO,DOO-plus, or nearly the consensus figure, if the BLS had still been reporting the monthly figures the way it used to.

The change in method, which seemed little-noticed by the markets, is known as "concillrent seasonal adjustment," and was introduced "in May 2003. It lends greater accuracy to the monthly estimates. Before conCillTent seasonal adjustment, however, the September increase probably would have been reported at more than 100,000, with virtually no revision to prior months. With conCillTent seasonal adjust­ ment, the increase is distributed more evenly over the full three months.

The only problem with this sort of accuracy, however, is that it's wasted on the markets, given their obsession with the most recently reported figm'e. In this case, the revisions to the two earlier months did have an impact, but the psycho­ logical effect is not quite the same. The irony, then, is that the improvement in accuracy can sometimes make traders less accurate in the way they interpret the numbers.

As for the unemployment rate, it first hit a 58-month low of 4.6% in May of this year. That it was back at 4.6% in September further confirms that payroll gains are both slow enough to stop the unemployment rate from falling, and fast enough to prevent it from rising. At this unemployment rate, the labor market is tight by almost any standard, with wages rising smartly. The increase in wages is clear even from a popularly-followed but flawed indicator, average hourly earnings, which tends to under­ state wage gains. It's even more apparent from the BLS' far more accurate measure of "hourly compensation," found in its data on productivity.

As of the second quarter, hourly compensation in the nonfarm business sector-which includes almost all forms, including benefits-ran 7.7% higher than a year ago. That's the fastest rate of increase since third quarter 2000, when the labor markets were even tighter.

And speaking of accuracy, the reason I called the planned upward revision of the payroll employment data "confidence-shaking" is that it shakes our confidence in the Bureau of Labor Statistics' ability to avoid the sort of errors it used to make. What used to happen is that it would generally overstate employment gains in a slow or contracting economy, and overstate them in an expanding economy.

But over the past few years, various improvements in technique were introduced that seemed to usher in a newera of greater accuracy. Those hopes were blasted away, however, when the BLS announced last week that the March 2006 tally of payroll employment would be upwardly revised by more 0.8 million. In proportionate terms, that's even larger than the huge upward revisions of 1997 and 2000, when the agency's statistical techniques were far cruder, and when gains in employment were truly going gangbusters.

The upward revision means, in effect, that from April 2005 through March 2006, payroll gains actually ran 40% higher ilian cillTently available estimates. Among other results, that closes much of ilie unexplained gap between payroll employment conventionally measured and as measured by the "payroll-compatible" data from the Household Sill'Vey (see "Missing from ilie Books," Economic Beat, Sept. 11).

It also makes you wonder if the BLS isn't underestimating gains in payroll employment right now.

By Gene Epstein