The Bureau of Labor Statistics (BLS) released its most recent employment report on March 8: In February of this year, the nonfarm economy, on net, created only 25,000 private-sector jobs and 20,000 jobs overall. One part of this report is the establishment survey, which contributed some of the weakest numbers since the past recession.
Forecasters failed to predict these anemic jobs numbers. In fact, before the report’s release, consensus market expectations foresaw 180,000 jobs being created in February. Thus, consensus expectations “missed” the February payroll number by 160,000 persons on the downside. The household survey component of the report was stronger, with the unemployment rate declining by 0.2%. According to the BLS, this decline in unemployment “reflects, in part, the return of federal workers who were furloughed in January due to the partial government shutdown.”
Another well-known statistic used to predict the BLS jobs number is the ADP national employment report. It’s produced by the ADP Research Institute, which is part of ADP. (ADP is an American company that provides human resource and payroll management software and measures nonfarm private sector employment using anonymous data from its clients.) The ADP report, which is released monthly a few days before the BLS employment numbers come out, predicted an increase in private nonfarm payroll of 183,000 in February. Thus, the ADP report differed from the BLS number by 158,000. This difference was one of the largest in 14 years, relative to forecast errors in months outside of U.S. recessions.
FRED provides an integrated picture of all this in the graph above, which combines three monthly series: BLS total nonfarm payroll, BLS government employment, and ADP total private nonfarm payroll. All three are presented as their month-over-month changes in thousands of persons. We combine the three series by first taking the difference of the first two. This gives the change in private nonfarm payroll according to the BLS. Second, we construct a “forecast error” by subtracting the ADP number (which we use as our forecast) from the actual BLS private payroll number.
Excluding five months during the recession (the shaded section in the graph), the downside difference between the BLS and ADP numbers for February employment was larger than the downside difference of any other month since 2003.
How this graph was created: Search for “total nonfarm payrolls,” select the series “All Employees: Total Nonfarm Payrolls,” and click “Add to Graph.” From the “Edit Graph” panel, use the “Edit Line 1” feature to “Customize data”: In this field, enter “government employees.” From the list of options, choose “All Employees: Government” and click “Add.” Again under “Customize data,” search for and add “Total Nonfarm Private Payroll Employment.” The first two series are from the BLS and the third is from ADP. For each of these three series, adjust the units to “Change, Thousands.” Next, compute the series shown here by subtracting the other two. FRED denotes the variables for each series in order: So, enter “a-b-c” into the “Formula” box and click “Apply.”
Most people look forward to Fridays in general, but data analysts and economists eagerly await one in particular: the Friday when the BLS’s employment situation is published. Two headline figures in this release are the unemployment rate and total nonfarm payrolls. These numbers are still subject to revision after their initial release. For example, the payroll numbers are based on about 70% of the surveyed businesses, and that number gradually increases to about 94% through revisions. Thus, relying on this first release to tell the whole story may be a bit premature, given that something could be changed by the revisions.
ADP is a company that provides payroll services to many businesses. It uses its internal data, as well as other economic indicators, to predict a few days before the BLS’s release what the final payroll number will be. The graph here compares the BLS series (in red) and the ADP series (in blue) and shows that there are some spectacular hits…and misses. Note that the misses could be on either side―too high or too low―as both are imperfect measures. Yet, if both measures agree, that’s a strong indication that they hold some truth.
How this graph was created: Search for nonfarm payrolls, select the relevant series, and click on “Add to Graph.”
