Posted by: Josh Lehner | July 8, 2011

The Effectiveness of the ADP Employment Report

UPDATE: The data, graphs and equation have all been updated through January 2012 in a new post. Click to go to the new post.

This morning’s employment report was very disappointing. Private sector gains of only 57,000 and public sector losses of 39,000, resulting in a net gain of just 18,000, along with a slight uptick in the unemployment rate to 9.2 percent. Following on the heels of the ADP employment report the day prior, which indicated 157,000 private sector jobs created in June, there is good reason to call into question the accuracy or effectiveness of the ADP report. Naturally, the stock markets increased yesterday after the ADP report was released (the reason for the increase was attributed by the press to the ADP report) and today, stock markets are down based on the BLS report. Mixed signals from the reports? Is the ADP report even useful anymore?

In theory, the ADP report should be a near perfect gauge of the private sector employment situation as outsourcing payroll processing is a very common practice. (Outsourcing as in outside the given company, not necessarily off-shoring). However, psychologically it feels that the ADP report in practice is misleading and is not a very accurate gauge of the BLS reports, especially in recent months. For instance (I realize this is cherry picking data points), in December 2010 and January 2011, ADP reported private sector gains of 256,000 and 190,000, respectively. A few days after each report the BLS reported that the private sector gains were only 167,000 and 94,000 for those two months. More recently ADP reported gains of 36,000 in May while the BLS estimates the gains to be 73,000 and the June reports differ by 100,000, as mentioned above.

Given all of this, the following information tries to look at the effectiveness of the ADP report in terms of predicting the official BLS figures. The first graph simply illustrates the monthly changes in private sector employment from the two different reports. The second figure is a simple, linear regression of the ADP report and the BLS report, with the final graph showing the fitted regression results and their residuals.

Overall, the two lines appear to be very close. There are some differences, however the same general pattern is evident.

Using a simple, OLS equation the monthly change in each series is very close. The equation fit is approximately 90% – that is, the changes in the ADP report explain about 90% of the changes in the BLS report. The coefficient of 1.017998 on the BLS change indicates that, on average, you can expect the official BLS change to be 101.8% what is reported in the ADP release, or just slightly larger.

Lastly, this graph shows the equation fit with the lower, blue line representing the residual, or error of the fit. The two dotted, gray lines represent 2 standard deviations, or approximately 95% confidence band. (Approximately 5% of the data points/errors are outside the 2 standard deviation band) Overall there is a good fit, as one would expect/hope. If you feel that the ADP report has been a poor predictor of the BLS report in recent months, you are somewhat vindicated as 3 of the past 7 months have been outside the 2 standard deviation range. (These errors are to the negative meaning the ADP report overstated the job figures from the official BLS report.)

In summary, the ADP report does a very good, big-picture job of predicting the official BLS private sector employment report. This is important to keep in mind as the ADP report is released at least a day or two in advance of the official figures. The magnitude of the change can be called into question, however, as the discrepancies between the series do fluctuate quite a bit from month to month. Over the entire sample, the average difference between the two reports is 200 jobs, however one standard deviation of these differences is 76,400 (and two standard deviations are 152,800). This means with 95% confidence you can predict the official BLS report to have the same number of jobs that ADP reports +/- 152,800. Now, that difference is quite a big range, however the direction of the movement is much more reliable. Out of the 126 months for which we have data (going back to Dec 2000), only 7 times has the sign, or direction, been different between the two reports. That is, one report says a gain while the other says a decline in the number of jobs, or vice versus.


Responses

  1. You write: “Over the entire sample, the average difference between the two reports is 200 jobs.” Just 200? Is that a typo?

    But very interesting post!


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Categories

%d bloggers like this: