By overshooting the federal government's June job growth estimates threefold, payroll outsourcing giant Automatic Data Processing was schooled in the downside of economic forecasting.
ADP processes paychecks for one in six American workers, so the idea of using the company's data to create a forecasting model seemed like an obvious way to raise the company's profile and increase its market share.
"It's very much to get our name out there," says Victor Saliterman, senior vice president for marketing for ADP Employer Services.
Last summer, ADP partnered with Macroeconomic Advisers, an economic modeling company, and took five and a half years worth of ADP payroll data, representing about 60 million employees, to help create the ADP National Employment Report, a monthly forecast that comes out two days before the Bureau of Labor Statistics' monthly employment report.
When the report debuted in May, ADP's figures correlated with those of the BLS 90 percent of the time when the two were compared over a period dating back to January 2001, ADP says.
Then, on July 5, ADP announced what came as a shock to many: that June job growth in the private sector was estimated at 368,000. That would have represented the most robust monthly job growth since March 2000. The news set off fears among investors of inflationary growth and rising interest rates.
Two days later, the federal government's job growth number came out at 121,000, surprising even the architect of the ADP report, Joel Prakken, chairman of Macroeconomic Advisers. It also put a wrinkle in the company's well-conceived marketing plan.
"The divergence was several times the typical divergence between the two numbers" and the largest divergence to date by a factor of three, Prakken says.
Despite the different results, and the different data sets used by ADP and the BLS, ADP stands by its numbers, Prakken and Saliterman say.
Nonetheless, the disparity has raised questions about the accuracy of ADP's report. By focusing exclusively on its own data—8 million employees culled from 225,000 businesses—ADP might have missed other signs that job growth was not as hot as ADP's number indicated, says Ken Goldstein, an economist with the Conference Board. "There were no corroborating numbers to suggest it," Goldstein says.
Despite the huge data sets available to ADP and the federal government, which surveys 160,000 companies and government employers about the number of jobs added to their payrolls in a month, employment numbers are always educated guesses.
"There is some art and there is some science here," says Bruce Murray, CEO of Corzen, an online data analysis company. "If it's a consistent pattern of being wrong, then it's a problem. But one number is not a problem."
Still, ADP "better get it right next time," says IDC analyst Lisa Rowan, "or no one is going to be paying attention."
ADP's numbers represent one truth, though it might not be of much use to economists. "ADP's customers were hiring people faster than firms surveyed by BLS," Prakken says.
ADP numbers may eventually be vindicated if job growth data gets revised upward, Prakken says.
With the company's reputation on the line for next month, Prakken and ADP are not planning to change their methodology, nor are they worried about being off the mark. "All we're doing is collecting numbers and adding them up," Prakken says. "The data are what the data are."