Paranoia
Friday Catblogging

How Many People Should Be Working in America?

Jim Hamilton of UCSD--one of the greatest of the masters of econometric time-series analysis--writes:

Econbrowser: How many people should be working in America?: Quite a few commentators have suggested that the labor force participation rate is a much better indicator of the health of the U.S. labor market than is the unemployment rate. I feel that quite a few commentators have this wrong.... The issue... is the divergent impressions one would draw about the current U.S. labor market depending on which statistic you look at. The unemployment rate, which reflects the number of people who say they're looking for work but are unable to find it, suggests that the U.S. economy is currently in significantly better shape than it's typically been.... The labor force participation rate, which is the number of people who are either working or looking for work as a fraction of the population, suggests that far fewer people are working, relative to the size of the potential labor force, than has typically been the case.... The difference between the two measures has to do with people who aren't working and who further described themselves to the BLS as not actively looking for a job. Large numbers of such people explain both a low unemployment rate and a low labor force participation rate....

It seems very hard to quarrel with the statement that any given month's value for the labor force participation represents the confluence of different factors, some resulting from trends that are in all probability quite benign, and some representing cyclical economic swings. Whenever somebody looks at such a statistic and claims to have inferred what it is saying about purely cyclical forces, they must have used some method for distinguishing between these two kinds of forces.

In my opinion, no such reliable method exists. One idea you might try would be to fit a linear time trend to the data, calling any deviation from that trend line the "cyclical" factor.... [But t]here's no good reason to assume that the factors that are contributing to the trend are in fact evolving in a linear, purely deterministic way....

The method that Bradbury used in order to arrive at her lowest estimate, 1.6 million, of the number of missing jobs... amounts to assuming that the slope of a linear trend fit from 1960-1994 could be extrapolated to 2001-2005 to identify the magnitude that we should normally be expecting for that figure. In the case of mature men, that's maybe not such a bad assumption.... On the other hand, for women aged 35-44, this amounts to assuming that the increase in women's labor force participation rates between 1960 and 1994 should have continued to climb upward, and, since it has not, Bradbury finds 1.1 million "missing" jobs in this group alone....

My point is not that there's a right way and a wrong way to control for trends, but rather that there are some fundamental problems with any way you choose to do it, and any conclusions you draw from these exercises need to be very carefully qualified...

One way to gain more information about what is "trend" and what is "cycle" is to take a look at other time series indicators that we believe have a similar cyclical component. When the labor market is cyclically weak, we believe that the unemployment rate will be higher than trend, that the employment-to-population ratio will be lower than trend, that average hours worked will be relatively low (since firms are likely to cut back both on bodies and on hours when their demand for labor is weak), and that the average duration of unemployment will be relatively high (because more of the fluctuations in quits, firings, and hires that drive the employment side of the business cycle are on the hires side). What do these series look like? Here they are:

The issue at stake is essentially whether the past five years have seen a stable value for the trend unemployment rate and a fall in the underlying trend for the employment-to-population ratio (meaning that our current employment-to-population ratio is actually close to, not far below, the long-run trend) or whether the past five years have seen a stable value for the trend employment-to-population ratio and a fall in the underlying trend value for the unemployment rate. The weekly hours series and the unemployment spell duration series seem to vote with the employment-to-population ratio: three series seem to say that the current cycle component is large, that there has been only a smell recovery from the business cycle trough levels, and that we are still pretty far away from full employment. Only one indicator--the unemployment rate--seems to say that recovery is well advanced and that the cyclical component has substantially shrunk.

This, however, doesn't resolve the mystery: why is the indicator that is the unemployment rate giving a different signal? What has happened to keep workers whom we would have expected--given the behavior of unemployment spell duration, hours, and the employment-to-population ratio--to say that they are unemployed from saying so when the CPS interviewers come to call?

Any Berkeley seniors who know (or are confident they can quickly learn) some time series econometrics and are interested in writing a macro-labor thesis, drop me a note. Some serious quantitative work on this would be genuinely useful...

Comments