Don't count on ADP jobs figures to predict the Friday report

Daniel Acker | Bloomberg | Getty Images

A job seeker looks over job listings during the Quad Cities career fair in Moline, Illinois.

Before the market gets a glimpse at the latest government jobs data each month, payroll management company ADP gives it a preview with its own monthly employment report. The metric is closely followed by the market, but the two numbers often differ by thousands of jobs.

So how useful is the ADP report for predicting the Bureau of Labor Statistics figures each month? Over the long term, it's clear that the two indicators tend to fall in line, but over the short term, it's hard to look at the ADP National Employment Report figure and know exactly what's coming in the official report.

The difference between the number of new jobs reported by ADP every month and the BLS private nonfarm jobs number that comes two days later is greater than 40,000 about half the time, according to a CNBC analysis of the figures since November 2012. That means that the disappointing 154,000 figure reported Wednesday by ADP for September could easily be a healthier number near 200,000 or an anemic one closer to 100,000.

ADP's figure for September is about 20,000 jobs under the number of jobs added in August, and well below the monthly average so far this year of around 185,000. So can we at least predict a decline compared to last month's BLS figure of 126,000?

That, too, appears to be about as reliable as a coin flip. The monthly change in job gains reported by ADP goes in the same direction as the change in the BLS only about half the time. So the decline in the ADP doesn't mean we should expect a lower number from the BLS on Friday. Indeed, Dow Jones puts the median forecast for all nonfarm jobs at 169,000, up slightly from 151,000 last month.

So what can the ADP report help us predict?

Well, it's very unlikely that the BLS will find that the economy lost jobs last month. Based on ADP's performance since it partnered with Moody's and initiated a new methodology in late 2012, we can be 95 percent sure that the BLS will find that the final number of new private jobs is somewhere between a paltry 46,000 and a robust 294,000.

That isn't a precise prediction, but it's not bad considering that the two indicators come from entirely different sources. ADP counts all employees who are listed as active on an employer's payroll, while the BLS surveys companies to tally employee who are actually paid. That means that real-world structural changes to the economy or temporary labor force changes can cause the figures to diverge.

For example, ADP told CNBC that the Verizon strike earlier this year put a wedge between the two numbers (ADP counted the striking workers, but BLS would not), and even weather events can cause a gap.

So both numbers can be "right" about what they're counting and still have different results over the short term, simply because they're counting slightly different things. That's why they tend to trend together over the long term.

The ADP may be a valuable measurement reflecting economic fundamentals, but for an investor looking to predict what the market will do on Friday, the headline number is probably not very useful for specific predictions — such as whether the BLS will beat consensus estimates.

In a paper published in Applied Financial Economics in 2014, researchers found that including the original ADP data in a forecast model improved the accuracy compared with a standard model that lacked that information, but that the advantage is no better than simply using Bloomberg consensus forecasts, which are available before the ADP figures are published. In other words, ADP doesn't seem to provide information that is not already present in the market in the form of analyst expectations.

"It is of considerable interest for future research why the market now pays so much attention to the ADP release since it conveys no additional and useful information in comparison to the current consensus," the authors wrote.

Another study released this September by Goldman Sachs found similar results for the report, since ADP's substantial methodological upgrades in late 2012.

"In sum, our finding suggest that the ADP employment report has only limited value in forecasting the CES [the BLS's Current Employment Statistics report], as much as its marginal information content appears to come from other publicly available data such as jobless claims that are incorporated by ADP into their official figures," the analysts wrote.

However, the Goldman Sachs report did find that surprises in the two measures tend to point in the same direction (62 percent of the time), so forecasters may be able to improve their models by taking into account big ADP surprises. The ADP measure may be very close to other data available during periods of steady growth, but it could also have more potential to add more value during periods of flux in the labor market.

"With regard to the 2014 paper, it is difficult to distinguish between the information from ADP and the consensus in the past several years given that monthly job growth has been remarkably stable," said Mark Zandi, chief economist at Moody's Analytics, which prepares the report for ADP. "The real test will occur when job growth slows significantly or actually declines in a consistent way."

Over the long haul, investors should expect the two indicators to move together, and investors may even be able to glean some information from sudden swings, but they shouldn't put too much stock in small changes in the monthly number. The ADP report is just one more piece of information that can help investors forecast the scope of job growth each month.