Misleading Payroll Employment Reporting Has Resulted from Unpublished Revisions of Concurrent-Seasonal-Factor Adjustments.  Further to Commentaries No. 336 and No. 331, this morning’s (December 3rd) reporting of November 2010 payrolls included misleading data, where historical revisions inherent in the Bureau of Labor Statistics’ (BLS) concurrent-seasonal-adjustment process are not published, even though they are related meaningfully to the published, seasonally-adjusted headline data.

As cited in No. 336, the BLS’s stated reasons for not publishing these revisions are tied to publication costs and to a desire not to confuse data users, although it is the non-publishing of the data that has caused misperceptions and confusion.  The BLS, however, does make available enough material for anyone interested to run the concurrent-seasonal-adjustment process on the raw payroll data, and we have done so, along with historical revisions and with results that are within single digits of official reporting.  Some detail from today’s run follows.

Reported Payroll Employment Trough in December 2009 Dragged into Early 2010.  The BLS Web site shows the trough of seasonally-adjusted nonfarm payrolls in the current cycle to have been December 2009.  The revised data from the November 2010 concurrent-seasonal-adjustment run now show January 2010 to have been slightly weaker, with a bottoming process running from December 2009 to February 2010, with roughly 6,000 jobs separating the adjusted monthly totals.  Importantly, this does not include adjustments from the downside benchmark revision to the unadjusted data that will be released in February 2011.  That revision — which will recast the historical seasonally-adjusted as well as the unadjusted series — already was likely to move the near-term seasonally-adjusted trough into 2010.  

Concurrent-Seasonal-Factor Adjustments Boosted September and October 2010 Payrolls at Expense of February to April 2010 Reporting.  Abnormal concurrent seasonal adjustments boosted last month’s reporting of September and October payroll levels, and the resulting revised historical seasonal factors reduced the levels of previously-reported payroll levels in February, March and April as an offset.  The problem is that the revisions to the earlier numbers have not been published, so that the September and October jobs gains appear falsely to be standalone results.  With today’s reporting, those September and October excesses remained, but November’s muted gain appears to have been something of an offset or catch-up, bringing the seasonals more into balance.  Again, these numbers are independent of the pending benchmark.

Extraordinary Downturn Has Damaged Reporting Quality Severely.  When first introduced in 2004, the BLS’s use of concurrent seasonal factors may have made some sense.  As with the Birth-Death Model that was torn apart by the severe loss of jobs, however, the extraordinary duration and depth of the economic contraction, and resulting disruptions to the normal cycles of commerce, have made recalculating seasonal factors every month a fruitless exercise with unstable results.  Modern economic reporting never was designed to handle an ongoing economic catastrophe.  Unfortunately, the collapse of reporting quality with a number of key series has coincided with a period where accurate and meaningful economic reporting could not be more important.

 

November Labor Data Were Consistent with a Weakening Economy.  With results weaker than consensus — a payroll gain that is statistically indistinguishable from a contraction and a jump in unemployment that is borderline significant — today’s labor report generally was more consistent with a weakening than a strengthening economic environment.  

Payroll Survey Detail.  The BLS reported a statistically-insignificant, seasonally-adjusted November 2010 jobs gain of 39,000 (a gain of 77,000 before prior-period revisions) +/- 129,000 (95% confidence interval).  October payrolls showed a revised 172,000 increase, which had been reported initially as a gain of 151,000.

As shown in the next two graphs, the level of payroll employment still stands below where it was a decade ago, despite the U.S. population growing by more than 10% in the same period.  The structural impairments to U.S. economic activity continue to constrain normal commercial activity, preventing any meaningful recovery in business activity, as discussed in Special Commentary No. 333.

 

 

 

 

The first of the preceding graphs reflects payrolls net of temporary census hires, with the red line reflecting the likely benchmark revision due for publication with the January data in February 2011.  The magnitude of the revision was estimated by the BLS along with September reporting.  Such shows no meaningful recovery of upturn in the economy.  Industrial production numbers and likely revisions to the payroll data suggest that the onset of timing for an official double-dip in this recession will be in the August/September 2010 timeframe.  The second graph is similar to the first one, except it includes the census hires, as officially reported.

From peak-to-trough (the peak month was December 2007; December 2009 is the short-lived official trough of the current cycle), payroll employment declined by a seasonally-adjusted 8,363,000 jobs, or 6.1%.  As of November 2010 reporting, payrolls purportedly have gained 0.7% or 951,000 jobs since the December 2009 trough, if one uses the published seasonally-adjusted payroll estimate for December 2009.  Distortions from the concurrent-seasonal-factor process are discussed in the opening comments.  Separately, the current December 2009 trough should shift into 2010 with the upcoming benchmark revision, along with a reduction in current employment levels by more than 600,000.

In terms of annual change, November 2010 was up by 0.64%, versus October’s revised 0.54%, (initially 0.48%).  Net of likely benchmark revisions, annual change was about 0.2%.

 

 

The preceding graph of long-term year-to-year payroll change reflects the numbers as reported, with no adjustment for census hiring variations or the pending benchmark revision.  Thanks to recent, protracted bottom-bouncing in the payroll series, current annual growth has recovered from the post-World War II record 4.96% decline in July 2009.  The July 2009 decline was the most severe annual contraction seen since the production shutdown at the end of World War II, which reflected a trough of a 7.59% annual contraction in September 1945.  Disallowing the post-war shutdown as a normal business cycle, the current annual decline remains the worst since the Great Depression, and should deepen further, net of the pending benchmark revision.

The next graph shows the year-to-year detail both with and without the census hires, but still without benchmark considerations.

 

 

Again, while the patterns of year-to-year change have recovered their pre-collapse levels, such is due only to the particularly protracted nature of the downturn, with year-ago comparisons against the bottom-bouncing.  In the above graph, keep in mind that the level of payrolls in November 2010 still is 5.4% below the level seen when annual growth in payrolls began to tumble.    

Benchmark Revision — About Minus 610,000 and Counting.  Announced along with the September 2010 payroll release was an early estimate for the 2010 benchmark revision, which indicated the not-seasonally-adjusted March 2010 payrolls were overstated by 366,000.  As the data are re-worked for that estimate, changes will be carried back to the prior revision as of March 2009, as well as carried forward to present reporting.  Such suggests that the overstatement of the level of payrolls as of November 2010 reporting is about 610,000 jobs.  The formal benchmark revision and restated economic history will be published with the January 2011 employment report, due for release on February 4, 2011.

Birth-Death/Bias Factor Adjustment. Where the BLS cannot measure the impact of jobs loss and jobs creation from employers starting up or going out of business, on a timely basis (within at least five years, if ever), such information is estimated by the addition of a bias factor generated by the Birth-Death Model (a model of the effects of new business creation and old business bankruptcies).  The unadjusted November 2010 bias factor was a monthly subtraction of 8,000 jobs, versus 23,000 jobs subtracted in November 2009, and against a monthly addition of 61,000 jobs in October 2010.

The BLS announced in last month’s release that it would begin estimating its monthly bias factors on a quarterly basis, instead of on an annual basis, along with the publication of the 2010 benchmark revision in February.  This appears intended to provide more frequent and accurate adjustments to the biases, reflecting data from the BLS Quarterly Census of Employment and Wages.  While such has to be taken as a positive move, it does not appear to address the fundamental flaws of the Birth-Death Model, discussed as usual in the ensuing paragraphs, and it is not likely to improve the quality of current reporting meaningfully.

Based on the unsupportable assumption of sustainable economic recovery, the bias factors used in the last eight months of payroll reporting have been 31.3% ahead of what they were reset to after the most recent and disastrous 2009 benchmark revision.  Positive assumptions — commonly built into government statistical reporting and modeling — can become self-fulfilling prophesies, with "stronger" economic data being reported as a result of happy guesstimates, or underlying assumptions of ongoing economic recovery.

Historically, the Birth-Death Model biases have tended to overstate payroll employment levels — to understate employment declines — during recessions.   There is a faulty underlying premise here that jobs created by start-up companies in this downturn have more than offset jobs lost by companies going out of business.  So, if a company fails to report its payrolls because it has gone out of business, the BLS assumes it still has its previously-reported employees and adjusts those numbers for the trend in the company’s industry. 

Further, presumed additional "surplus" jobs, created by start-up firms, get added on to the payroll estimates each month as a special add-factor.  These add-factors are running at about 50,000 per month (seasonally-adjusted), at present.  I still estimate this monthly bias should be negative by 200,000 or so, on average.  Since it is not, the BLS continues regularly to overestimate monthly growth in payroll employment by roughly 250,000 jobs.  Most of that misreporting, however, now will not be corrected until at least the 2011 benchmark revision to be published in February 2012.

Household Survey.  The usually statistically-sounder household survey, which counts the number of people with jobs, as opposed to the payroll survey that counts the number of jobs (counting multiple job holders more than once), showed a seasonally-adjusted monthly employment decline of 173,000 in November, following a reported loss of 330,000 in October.

The November 2010 seasonally-adjusted headline (U.3) unemployment rate increased by a statistically-insignificant 0.18 percentage point to 9.82% +/- 0.23% (95% confidence interval), up from the 9.64% U.3 in October.  Not seasonally adjusted, November’s U.3 unemployment rose to 9.3% from 9.0% in October. 

The November U.6 unemployment rate held at a seasonally-adjusted 17.0% versus October, but the unadjusted rate rose to 16.3% in November from 15.9% in October.  The broadest unemployment rate published by the BLS, U.6 includes accounting for those marginally attached to the labor force (including short-term discouraged workers) and those who are employed part-time for economic reasons (they cannot find a full-time job).

In 1994, during the Clinton Administration, "discouraged workers" — those who had given up looking for a job because there were no jobs to be had — were redefined so as to be counted only if they had been "discouraged" for less than a year.  This time qualification defined away the long-term discouraged workers.  The remaining short-term discouraged workers (less than one year) are included in U.6. 

Adding the SGS estimate of the excluded long-term discouraged workers back into the total unemployed and labor force, unemployment — more in line with common experience as estimated by the SGS-Alternate Unemployment Measure — notched higher to about 22.6% in November, versus 22.5% in October.  The November reading was the highest of the current cycle.  The SGS estimate generally is built on top of the official U.6 reporting and tends to follow its relative monthly movements.  See the Alternate Data tab for a graph and more detail.

As discussed in earlier writings, while 22.6% unemployment might raise questions in terms of a comparison with the purported peak unemployment in the Great Depression (1933) of 25%, the SGS level likely is about as bad as the peak unemployment seen in the 1973 to 1975 recession.  The Great Depression unemployment rate was estimated well after the fact, with 27% of those employed working on farms.  Today, less that 2% work on farms.  Accordingly, for purposes of a Great Depression comparison, I would look at the estimated peak nonfarm unemployment rate in 1933 of 34% to 35%.

Week Ahead.  Given the unfolding reality of an intensifying double-dip recession and more-serious inflation problems than generally are expected by the financial markets, risks to reporting will tend towards higher-than-expected inflation and weaker-than-expected economic reporting in the months ahead.  Increasingly, previously unreported economic weakness should show up in prior-period revisions.

SGS Ongoing M3 Estimate (November 2010).  A preliminary M3 estimate for November 2010 will be published over this weekend (December 4th) on the Alternate Data Tab.  The series appears to be on track to show a 2.8% year-to-year contraction for the month.  A review of liquidity conditions will be included in the next Commentary.

Trade Balance (October 2010).  The October trade deficit is due for release on Friday, December 10th.  While expectations are for no meaningful monthly change (per Briefing.com), any major movement in the monthly data would give the first indication of reporting pressure on fourth-quarter GDP from the standpoint of trade activity.  I expect continued monthly trade deficit deterioration, with corresponding early downside indications for fourth-quarter GDP reporting.