On Thursday, we brought together an impressive array of scholars to discuss the causes and consequences of, and policy responses to, long term unemployment, including Prakash Loungani, Advisor in the IMF’s Research Department, Kenneth Scheve, Professor of Political Science at Yale, Phillip Swagel, Professor of Public Policy at the University of Maryland, and a former Assistant Secretary of Treasury for Economic Policy, Rob Valletta, Research Advisor at the Federal Reserve Bank of San Francisco, Dan Aaronson, Director of Microeconomic Research at the Chicago Fed, and Kenneth Troske, Professor of Economics from the University of Kentucky. And that was in addition to the researchers from the University of Wisconsin-Madison (more on them below). For me, this was a tremendous learning experience. But like all good conferences, by the end I understood that I knew less than I thought I knew about long term unemployment. In today’s post, I will discuss the presentations and papers by Prakash Loungani and Rob Valletta; in the next post, I’ll cover the findings of Ken Scheve and Phillip Swagel and Ken Troske.
Prakash Loungani’s paper (with Jinzhu Chen, Prakash Kannan, and Bharat Trehan) provided one approach to trying to determine the sources of long term unemployment. They proxy shifts in the structure of the economy with a measure of the dispersion of stock returns (Figure 4 from the paper). They then use a vector autoregression to identify impulse response functions for unemployment at various horizons. Long term unemployment responds positively to this index, as shown in Figure 6.
Figure 4 from Chen, Kannan, Loungani, Trehan (2011).
Figure 6 from Chen, Kannan, Loungani, Trehan (2011).I’m always fascinated by empirical relationships that appear to be robust, and this one, at least so far, does not seem particularly fragile. Loungani observes that the results are robust to the inclusion of an alternative measure of dispersion (Bloom’s measure (Econometrica, 2009)). The results imply the following for the cyclical/structural components at various durations of unemployment (a slide from the morning presentation).
Figure from Loungani presentation.I have two observations here: The first is that the cyclical structural component is larger for those with longer unemployment duration, which is consistent with intuition. The second is that even at the longest duration category, no more than 40 percent is structural.
Dan Aaronson, Director of Microeconomic Research at the Chicago Fed discussed the paper. He noted that it was remarkable that nearly half the rise in long term unemployment was explained by one variable. One of his key concerns was that the outliers in both series associated with the Great Recession naturally made the dispersion variable successful. I also wondered whether the strength of the identified relationship would persist in a sample truncated before the Great Recession. Dr. Loungani observed that the relationship still prevailed in a short subsample, although he had not conducted a formal out-of-sample forecasting exercise.
Dr. Aaronson also observed that in a Mortensen-Pissarides matching framework [lecture notes on MP model], with reasonable calibration, no more than two percentage points of the increase of the unemployment can be structural. He illustrates this point in this figure:
Slide 6, Aaronson discussionHe stressed that something closer to one percentage point was more likely an estimate.
Rob Valletta addressed the question of whether rising unemployment duration in the United States was due to a composition effect, or a change in behavior, over the long term. Carefully addressing the changes in the construction of the survey, he concluded
“…After accounting for changes in the [Current Population Survey] survey and using a more complete and appropriate set of conditioning variables than has been used in past work, the results suggest limited changes in unemployment duration over the past three decades. However, conditional durations have been longer during the recent severe recession and its aftermath than in the similar episode during the early 1980s, primarily due to higher labor force attachment for women and lower unemployment exit rates among the very long-term unemployed.”
This finding is illustrated in the paper’s Figure 1, which plots the unemployment rate against the unadjusted and adjusted duration. The key point is there is no pronounced trend in the adjusted series, until the Great Recession. This suggests the compositional effect dominates.
Figure 1 from Valletta (2011).Dr. Valletta concludes:
The higher durations in the recent recession appear largely due to: (i) increased labor force attachment among women, as reflected in the patterns for female labor force entrants (consistent with the arguments of Abraham and Shimer 2002); and (ii) lower unemployment exit rates among the very long-term unemployed (2 years or more). Both of these groups are generally ineligible for extended UI benefits. These findings suggest that there has been little or no change in the behavior of unemployed individuals over the past three decades, including a limited impact of the historically unprecedented extensions of unemployment insurance benefits over the past 3 years.
The discussant, UW Professor Rasmus Lentz, observed that typically, one would want to estimate the hazard rate of exiting unemployment, which is straightforward with a sample of the flow, whereas the CPS is a sample of the stock; the hazard rate can still be estimated if the proper adjustments are undertaken. The analysis of generated of synthetic cohorts that Valletta implements is another approach to estimating the hazard rate.
He also observed that one could obtain a more structural interpretation of the results by including the vacancy rate as a regressor — that is the structure of the market implies a covariance between job finding and tightness. However, if all one wants to do is to is to estimate how job finding has changed over time, it’s not clear that one needs to include a measure of unemployment or market tightness. [minor edits — mdc 11:40am]
In the discussion, I noted the fact that the mean duration figure did not completely convey the full story. In terms of the topic of the conference, I asked him to interpret further the lower exit rate out of the group unemployed over 99 weeks. Dr. Valletta agreed that this result was consistent with rising structural unemployment for those who have been unemployed for an extended period, but that this was quantitatively a small figure.
How to interpret these findings? I think it’s useful to recall the points that Dr. Valletta made in the morning session. In particular, there are at least three definitions of structural unemployment:
- often equated with persistent (long-term) unemployment; but this can be cyclical (disappears as economy recovers)
- mismatch between skills/location of workers and jobs (common definition)
- sources of unemployment (other than “frictional”) that contribute to a higher equilibrium “natural rate of unemployment” or NAIRU (“nonaccelerating inflation rate of unemployment”).
So, my reading of the views at the meeting were that there was a consensus that most of the unemployment increase since the onset of the Great Recession was cyclical in nature. Even when one takes the higher estimates of structural unemployment, structural unemployment still does not account for more than two percentage points of the increased amount of unemployment. Citing joint research with Mary Daly and Bart Hobijn, Dr. Valletta puts the estimate at 1.25 ppts, while Dr. Aaronson thought 1 ppts was reasonable.
The entire agenda for the conference, organized by myself and Mark Copelovitch, and sponsored by the La Follette School and the UW Center for World Affairs and the Global Economy, is here, while a recent Institute for Research on Poverty conference agenda is here. Recent posts on the subject are here, here, and here. The Economist has a long article on male unemployment in the latest issue. And for a survey of the costs of high unemployment, see this November ILO-IMF report.
This post originally appeared at Econbrowser and is reproduced here with permission.