Historical Payrolls and Employment Growth Revised Downward. As shown in the accompanying graphs, benchmark revisions to the payroll employment series of the Bureau of Labor Statistics (BLS) showed meaningful downside revisions both in terms of employment level and of year-to-year growth.

With the not-seasonally-adjusted March 2007 payroll survey benchmarked against state unemployment insurance filings of the time, the BLS estimated that March 2007 payrolls had been overstated by 293,000 jobs, which translated into a 284,000 downside revision to the seasonally-adjusted March number. By November 2007, such meant a 440,000 downside revision in the adjusted numbers. December should have been even worse, but as discussed in this month’s employment section, the December number may have been altered so as to keep the month-to-month payroll decline in January 2008 to a minimum.

Annual growth now appears to have slowed more sharply than previously reported, slipping to 0.72% as of January 2008. Growth that low, in a downtrend, has always been tied to a coincident recession.

The third graph shows some of the variability in the month-to-month payroll gain that is so heavily followed by the financial markets. Aside from the historical month-to-month numbers averaging a little less than previously reported (95,000 instead of 111,000 in 2007), the revision patterns showed a tendency for initial reporting (an average of 103,000 in 2007) to be revised upward in the standard two months of revision that follow initial release, exclusive of the benchmark revisions. In general, the BLS estimates a 95% confidence interval around the seasonally-adjusted monthly change of +/- 129,000 jobs.  

With the exception of the benchmark revision one year ago, most recent benchmark revisions have been to the downside, and most have been significant, regardless of direction. The process highlights some of the basic weaknesses in the accuracy of the regular payroll surveying and reporting, and it should raise issues for those in the private sector attempting to make use of the numbers.

ADP, ISM and Efforts at Mimicking Government Data. There was a time — a couple decades back — where it may have made sense for private sector publishers of economic data to attempt to link the reporting of the their series to popularly followed government statistics. If an indicator such as the purchasing managers survey proved to be a good predictor of broad economic activity, as measured by the GNP/GDP, then economists and analysts would come to rely heavily on the significance of the numbers. 

In recent decades, however, some government economic data have been gutted of economic significance thanks to methodological changes that have structured upside biases into the GDP and employment data, and downside biases into inflation reporting. Accordingly, private enterprises attempting to mimic today’s government reporting, unwittingly may be distorting and biasing their own otherwise very good and valuable data. Such enterprises can put out meaningful reports that have the respect of the markets, without having to tie themselves too closely to tainted government data. Two recent cases come to mind: Automatic Data Processing, Inc. (ADP) and the Institute for Supply Management (ISM).

Relatively new to the economic reporting field is ADP, a company that processes a very large number of company payrolls. Their unique database has enabled the creation of ADP’s National Employment Report. Given the quality of the company’s data, the high level of technical and economic skill behind the data analysis, and presumed political neutrality, significant trends should be signaled by the ADP data with a higher accuracy than often are available from the government’s monthly survey.

The issue is that ADP adjusted the reporting of their monthly data in an early effort to mimic the monthly payroll change as reported by the government. Although this move appears to have been quite popular with financial traders, who always look for an edge in anticipating a key market-moving statistic such as the monthly payroll survey change, there can be problems trying to use high quality data to imitate a highly volatile number that is subject to political massaging, birth-death models, etc.

Consider that the government’s monthly payroll gain comes within a +/- 129,000 range 95% of the time, and lies outside that range one time in twenty. If one used 100,000 as a base in 2007 (the average initial jobs gain reported by the BLS in 2007 was 103,000), a range from minus 9,000 to plus 229,000 easily covered all the initial reportings. One could expect results in a given month that would show something close to a random number in that range. If the numbers were random, they are by nature unpredictable. Yet, markets can respond if the monthly change is 10,000 more or less than consensus estimates. When consensus forecasts come close to the actual number, one has to wonder how random the government’s monthly reporting process is in reality.

The ISM has been publishing its purchasing managers survey of manufacturing activity for some years, and the survey has been recognized broadly as a reliable indicator of economic activity. Of particular value is the new orders component of the manufacturing survey, which I have used regularly as a leading indicator of significant economic change. It was one of the first indicators to send out a warning signal in advance of the current economic downturn and has been a much more accurate indicator of current economic activity than official GDP reporting.

On January 18, 2008, the Institute for Supply Management (ISM) announced a reformulation of its purchasing managers index (PMI), where, "The new formula more closely predicts the GDP for the past several years." The ISM explained further that, "In late 2006, ISM asked a group of supply management professionals, educators and leading economists to study the relationship between the PMI Index and GDP (gross domestic product). After extensive study, the revised formula was determined to more closely predict GDP. While the impact of the change is marginal, it does provide a more precise measure than the previous formula."

Therein lies the problem. The ISM recognized a disconnection between the results of its surveys and the ever-rosy recent performance of the GDP. The reaction was to change the ISM series, assuming the GDP was correct. Yet the problem was with the quality of the government’s GDP reporting, not the ISM’s high-quality survey.

Basically the weightings of new orders and production, which had been 30% and 25%, respectively, were changed to the 20% weighting now applied to all series, including employment, supplier deliveries and inventories. Not surprising, the best leading indicator was downgraded because it failed to perform in predicting politically-rigged results of the GDP.

While the reweightings do not alter the broad picture significantly, in conjunction with the new seasonal factors introduced in January, they did prevent the January 2008 index from falling below 50.0 and signaling a contracting manufacturing sector. Such would have happened otherwise, without the reformulated changes (see the Purchasing Managers Survey section in the Better Quality Numbers segment for a detailed table).

Fortunately, the individual series, such as new orders, will continue to be published as usual, along the regular once-per-year revamping of seasonal factors, as determined by the Commerce Department.  

As to GDP, the annualized quarterly real (inflation-adjusted) growth rate come with a 95% confidence interval of +/- 3%. Yet, consensus forecasts often come extremely close to the "advance" or initial estimate. There is nothing random about the reporting of the carefully structured GDP estimates.