Jobless Recovery or Are the Data Wrong?

Do we have a jobless economy problem or just a batch of bad data? Pay attention to this one, folks.

According to Economy.com, the U.S. business establishment survey has a sad tale to tell about the labor market: over 1.1 million jobs have been lost since the recession ended—surely, the worst hiring slump since records began in 1939. However, other economic data are questioning the message from these numbers; previous employment revision patterns also point to a more optimistic story. Perhaps this recovery is not so jobless after all.

However, something doesn’t seem right. The U.S. economy is gaining momentum: GDP grew at a healthy 3.1% in the second quarter and is set to grow at over 4% in the third. And unlike the brief spurts over the past year and a half, growth this time appears to be broad based. Productivity gains are part of the story, and these have been truly remarkable. However, they can’t be the whole story.

True, the labor market lags other parts of the economy and is generally the last to pick up during an upturn. Still, payrolls ought to have stabilized by now. Indeed, important hiring indicators—weekly unemployment claims, consumers’ assessment of hiring conditions, ISM services employment survey—soft as they remain, are pointing to stable employment, not outright declines. The establishment survey, by contrast, has reported job losses during each of the past eight months.

Something is amiss in the establishment survey. I’m convinced that the revision patterns of the early-1990s recovery cycle are set to be repeated. In the first twenty one months following the end of that recession, a total of 1.4 million job gains were originally reported; this number was subsequently revised to 2.9 million. The upward revisions were fairly evenly distributed across the months and continued for sometime. By similar—admittedly very crude—analogy, payrolls would be presently understated by a factor of 2; perhaps later revisions will show roughly flat payrolls from the end of the recession through August 2003, instead of net losses of 1.2 million. That kind of performance would be more consistent with other macro data on the economy.

The heart of the problem is hiring by small businesses, which tend to be under-represented in the establishment survey. To rectify this problem, the Bureau of Labor Statistics assumes a scale factor (net birth/death factor) that attempts to adjust total payrolls in a given month for gains or losses in small business employment. The adjustment factor is based on historical observations and, by the BLS’s own admission, typically misses upturns in hiring.

The problem is that small businesses are also usually the first to begin hiring during an overall labor market upturn. Initial estimates from the establishment survey, therefore, tend to understate employment gains in the early stages of the hiring cycle. In the past, subsequent revisions, based on more comprehensive data, have corrected this problem and lifted the originally reported numbers. I expect this time to be no different.

This is a lot of "tech-speak," but there is a very important point at hand--how good are the data and the estimating techniques we're using? Admittedly, this is no easy job. I have felt for too long that we grossly under-invest in good economic research to support economic development. This is a case in point.

Read more here (if you subscribe to Economy.com).