Gauging Employment: Is the Professional Wisdom Wrong?

The bureau of Labor Statistics' (BLS) monthly report on labor market developments is the government's most widely anticipated statistical release and the one that most influences markets, forecasters, and policymakers. The report provides detailed information on employment from two sources: the Current Population Survey of households, which also provides data on unemployment, and the Current Employment Statistics survey of payrolls from nonfarm business establishments and government. These two sources often produce very different estimates of the monthly change in aggregate employment, and the payroll data are widely accepted as the more reliable. They are featured on the first page of the monthly release and are the employment data most frequently discussed in the business press and other media. The annual Economic Report of the President has recently reaffirmed their reliability relative to the household data, as has Federal Reserve Board Chairman Alan Greenspan. Simply put, to judge by what the experts say and what the media report, most people interested in what is happening to aggregate employment rely on the payroll numbers. They use the household survey for the unemployment rate, demographic breakdowns, and various more arcane measures of labor market developments that only it provides. This paper questions whether the overwhelming preference for the payroll data as the measure of aggregate employment is justified, and concludes it is not.

Table 1.

Explaining Changes in Treasury Yields, January 1994 to June 2004^a

Policy and Market Reactions

What is the evidence that financial markets and policymakers respond to the payroll employment data and ignore the household data? Table 1 analyzes the response of market interest rates to the monthly employment release. The employment data are presented as initially reported rather than the later revisions, which reflect a much larger sample of payrolls and updated seasonal factors. Since market interest rates adjust continuously, they are modeled as responding to the shock in the monthly employment change, measured as the difference between the actual change in the payroll or household data and the expected change in employment as reported in a survey of analysts by Money Market Services.¹

In the first two panels of the table, the employment shocks are used to explain the change in ten- and two-year Treasury yields in the window from five minutes before to twenty-five minutes after the employment data are released.² The regressions are run on data from January 1994 to June 2004, the period for which data were available. The payroll employment shock has a statistically significant and meaningful effect on yields, and the several regressions using it explain between 40 and 45 percent of the variation in interest rates in the thirty-minute window. In the last column, even the change in the unemployment rate has no significant effect when used together with the payroll employment shock. In contrast, the R-squared of the equation is very low when the household employment shock is used alone in the regression, and this variable is insignificant when entered together with the payroll shock. The same regressions found no consistent effects when the window over which these interest rate changes were calculated was extended to the twenty-four hours between the market close the day before and that on the day of the employment release. For these two- and ten-year maturities, the initial reaction appears to be overcome by other developments in the course of the day. Adding the change in the Standard and Poor's (S&P) stock index over the same interval, as an explanatory variable that might capture such developments, did not change this result. However, in regressions over the twenty-four-hour period for yields on shorter maturities (three- and six-month Treasury bill rates, third and fourth panels of table 1), the payroll data again show significant effects on rates and the household data do not. Data for these maturities were not available for the thirty-minute window.

Turning to the response of policymakers, table 2 reports on how policy targets have related to employment changes and changes in unemployment...