Personal spending is the bulk of US GDP, accounting for roughly 70%. Employment, thus, is a key driver of economic growth. For this reason, investors and economists alike carefully watch the statistics that provide some insight to the dynamics of the labor market, particularly the ADP National Employment Report and the Bureau of Labor Statistics Employment Situation.
Between Thursday and Friday, we saw how quickly investors’ attitudes toward the economy changed when we received conflicting messages about the strength of the jobs recovery. On Thursday, the stock market rallied, thanks, in part, to the relatively solid results in the ADP report. On Friday, stocks dropped on a weaker-than-expected BLS report. The dissension between the two measures gave me a reason to take a look at how well correlated these measures really are. Here is what I found.
I recently downloaded the historical ADP data, which is available here. I calculated the change in total non-farm payrolls. The historical ADP time series goes back to December 2000, which is good because it gives us data for a couple of recessions and recoveries. I compared these numbers with the BLS non-farm payroll data, which is available here.
A quick graph suggests that the two data series track each other relatively well. Indeed, the correlation for the full period of January 2001 through May 2011 is 0.92.
click on image to enlarge
With that in mind, there is little wonder why investors keep an eye on the ADP figures to gauge expectations for the BLS results.
When we take a closer look, however, we find out how closely these numbers correlate changes over time.
During the 2001 recession (March – November 2001 according to the NBER), the correlation of the changes in payrolls stood at approximately 0.73. Not bad, but not terrific. Figure, a 73% is a passing grade, but just barely. Our kid has about a “C” average.
What about the correlation during the jobless recovery, from December 2001 through the first quarter of 2005? There, we see the correlation rose to about 0.83. Now we’re talking “B-/B” territory.
If we expand that to include the entire recovery and bubble period, so now we’re talking about December 2001 through November 2007, the correlation of the change in payrolls drops to 0.80. Our kid is looking at a “B-“, but just barely.
Of course, that suggests a relatively weaker relationship during the real recovery, from the point when the economy made up for the jobs lost before, during, and immediately following the 2001 recession, right up to the most recent recession. So, we will start counting with the beginning of the second quarter (April) of 2005, through November 2007. The correlation of the change in payrolls during this period is 0.68. Guess who’s going to bed without any supper.
During the last recession, from December 2007 through June 2009 according to the NBER, the correlation jumps to 0.97. A solid “A”. Nice. Harvard here we come.
What about the most recent jobless recovery, from July 2009 through June 2011? The correlation plummets to 0.74. Our “A” student is back to goofing off, and has a solid “C”.
The table below shows these correlations:
|Full Period January 2001 - June 2011||0.92|
|2001 Recession March - November||0.73|
|Recovery & Bubble December 2001 - November 2007||0.80|
|2001 recession jobless recovery December 2001 - March 2005||0.83|
|Recovery since end of "jobless" portion of recovery April 2005 - November 2007||0.68|
|2007-2009 recession December 2007 - June 2009||0.97|
|Present Jobless Recovery July 2009 - June 2011||0.74|
The ADP Employment Report provides some good insight to the labor market, yet it is not as comprehensive as the government release. For example, in recent months, the ADP report covered just over 108,000,000 workers, while the BLS report covers an employed civilian labor force of just under 140,000,000. This means that we can glean some useful insight from the ADP information. Of course, this also suggests that, perhaps, we shouldn’t be too surprised when the government figures come in considerably different from the ADP numbers.