Friday, August 23, 2013

More on Duration, Demographics, and Emergency Unemployment Insurance.

I've been looking some more at unemployment duration data, to see if we can make any broad estimates of what is going on.  I realize that my back of the envelope estimates aren't academically rigorous, but I hope this provides plausible food for thought, and some grist for speculative decision making.

Postscript: Upon further reflection, I think the approx. .75% of age-related unemployment, anomalous to the recent recession, which I found in the demographics section, is probably closely related to the .7% excess unemployment that I found in the later section on EUI related to excess unemployment duration above 26 weeks.  So, in total, of the approx. 5% in cyclical unemployment that we saw at the peak, I am attributing approx. .5% to age demographics and at least .75% to EUI.  This leaves 3.75% attributable to other factors, although EUI would likely be responsible for some of the remaining 3.75% in ways that I haven't been able to isolate here.

Under the fold.....


Robert Valletta at the Federal Reserve Bank of San Francisco, using data through 2010, found that durations have not increased so much when factoring in changes in data collection in 1994.  When I include a rough adjustment for the change, the relative level of durations from 1970 to 2000 appear to be pretty stable, compared to the unadjusted graph, which appeared to have a slope during this time period.  I think there is still evidence of age-related duration extension during the past decade or two, but any longer term secular trend looks weaker than I had previously thought.

To suss out that age effect, this graph compares the average duration of unemployment, by age (in 12 month moving averages, for clarity):

Here, we can clearly see the steady increase in unemployment duration as workers age.  This is from some combination of searching costs from specialization, greater income support and ability for consumption smoothing, and other factors.  This is unadjusted data, so the data from before 1994 is understated by 1-2 weeks compared to recent data for all age groups, but I am only interested in the relative positions at any point in time.

In particular, we can see the effect of demographics on duration by comparing the 25-34 year old group to the Total group.  The 35-44 age group shows similar levels of change over time.  Here I have isolated the difference between the series.

(1) We can see a long term decline in the age-group durations compared to the total durations, from before the 1980's recession to just before the recent recession.  In the graph above, we can see that at duration low points, the 25-34 group started the time frame just above the total average and slowly moved to a level just below it at the last low point.  This amounts to about 1 week, or if measured in proportional terms, about 5-6% of total trough level duration.

(2) The durations of the older groups also exhibit extra cyclical volatility compared to the younger groups, roughly proportional to the level of unemployment duration for each age group, except that the 65+ group seems especially volatile.  Compared to the early 1980's, the aging labor force added about 6 weeks to the total peak average duration, or about 15% of total peak duration in proportional terms.  We can see in the larger chart that in 1983, at the duration peak, total average duration was about 1 1/2 weeks lower than the 25-34 age group, or about 7 1/2%, and it was about 3 weeks higher than the 25-34 group at the recent peak, or about 7 1/2%.

If we assume that a change in unemployment at a given time would be roughly proportional to a change in duration, age demographics might add .2-.3% to the unemployment rate when it is in the 4-5% range.

At the peak, the cyclical effect might have added 1.25% to the comparable unemployment rate.  Right now, stated Total Duration is 37.6, 25-34 duration is 35.6, and we could estimate adjusted duration at 33.6.  That would suggest that about .8% of the current unemployment rate is inflated from age demographics, compared to 1983.  (Actually, since these are 12 month moving averages, the effect is probably slightly lower by now.  The monthly data is too noisy to use, but appears to continue to settle to less inflated levels.)

(Added: I derived the estimates above by comparing the most recent recession to the 1980's recession.  After doing a more comprehensive review of the behavior of duration among the age groups, it appears that except for the most recent recession, the age group durations increase fairly proportionately through the business cycle.  So, compared to 1980, the higher durations of the older age groups are increasing the number of unemployed workers by about 5% all the way through the business cycle, adding .2-.3% to the Unemployment Rate at boom periods and around .5% during recessionary periods.  The additional cyclical effect is anomalous to this cycle.  The source of that additional age-based unemployment duration could be a number of things, but I think it's plausible to finger EUI for much of it.  As a start, we should expect EUI to have an increasing effect on age group duration in proportion to the number of workers above 26 weeks of unemployment within that age group.)

Note on Interpretation

I am not saying that this unemployment doesn't exist or doesn't matter.  I am simply saying that some of the unusual persistence in unemployment that we are seeing can be explained by normal cyclical behavior of an aging labor force.  This should be factored into our expectations for changes in unemployment as well as for expectations stemming from a particular unemployment level.

Emergency Unemployment Insurance

Valletta also touches on the recent cyclical durations.  He appears to have done some interesting work getting duration details out of the data.  His research was early in the recession, but I think he gives UEI a pass.  In table 4, he compares the duration data of this recession to the 1981-1984 recession.  In the first month of unemployment, this recession shows a significantly stronger relative trend for returning to employment.  For the next few months, duration behavior is similar.  Going from the 2nd quarter to the 3rd quarter, which is when UI would have been expiring in 81-84, the relative rate of exit from unemployment peaks for the 81-84 recession.  The rate of exit peaks for the recent recession in the 2nd year of duration, when the extended UI would be expiring, while the rate of exit in 81-84 remains relatively strong.  For durations of more than two years, the relative rate of exit declines in both recessions, but the rate of exit in the recent recession falls off a cliff.

This poses an important policy question.  This is what we would see if EUI was pushing durations out, deepening the recession, and creating hysteresis among a labor pool that has been incentivized to prolong unemployment.

David Grubb from the OECD suggests that more than half of the excess unemployment at the peak of unemployment was due to UEI.  I can't add too much to the analysis.

Using the BLS historical data for duration, I estimated the duration of unemployment for workers unemployed for less than 27 weeks and for workers unemployed for more than 26 weeks.  I found that the duration under 27 weeks was a good predictor of the duration for the over 26 week group, with a 2 year lag.  Here is a chart with the estimated durations for the under 27 group, the over 26 group, and the forecast for the over 26 group, based on the relationship from 1976 to 2006:

During previous recessions, as described in the Grubb paper, EUI potential duration amounted to an average of 13 to 16 additional weeks in earlier recessions, compared to about 64 weeks in this recession.  With the relationship shown in the graph, we can get an estimate of the demographic effect and the unusual UEI effect for this recession.  The aging effect should inflate both the short term duration and the long term duration proportionately.  Since EUI is only applicable to durations over 26 weeks, it should mainly be manifest in an unusually high average duration for workers who have been unemployed for more than 26 weeks.  Actually, this should understate the effect, since EUI would cause transitions to employment to slow among the recently unemployed also, at least in the early part of the recession.

Durations under 27 weeks peaked at about 9.5 in 1983, and just under 9 in the more recent recessions.  It peaked at about 10.5 in the recent recession.  If we attribute 15% of the additional duration during a labor crisis to the higher volatility of duration among the older age groups, as described above, the age-adjusted duration would have peaked at about 9.  Perhaps, almost all of the non-UEI related excess duration is age related?

(Added:  Further analysis, as mentioned above, suggests that the average age-adjusted duration under 27 weeks for the recent recession would be about 10, which is .5 to 1 weeks more than the previous recessions.  Some of the remaining elevation in the under 27 week duration could be the result of EUI, or it could be the result of other factors specific to this recession.)

Durations under 27 weeks have settled at 7 to 7.5 weeks in recent cycles.  The age factor should cause the under 27 duration to settle .35-.4 weeks higher this cycle.

The Unusual Length of Durations over 26 weeks

The forecast predicts an average duration of over 26 weeks to be 60 weeks, but the actual duration is 80.  There are 4.3 million workers who have been unemployed for more than 26 weeks.  This suggests that EUI has led to nearly 1.1 million excess long-term unemployed, or an excess unemployment rate of about .7%.  As I mentioned in my previous post on the topic, the EUI effect could be heavier among older workers, so some of the age-related volatility I picked up above could be related to the effect of EUI on the duration of unemployment among older workers.  Due to the way I got at this, any effects of the combination of EUI and aging would have been marked here as part of the aging effect above.
(Added: So, the .75% of age related unemployment at the beginning of this post that was anomalous to this recession could be EUI related.  And, some of the extra unemployment reflected in the under 27 week durations could also be EUI related.)

The Good News

If there is any truth in this, and if EUI expires at the end of this year, then an unemployment rate not much over 6% could be attainable in 2014.  This could lead to further rises in interest rates sooner than anticipated.

(Added: In my subsequent analysis, I forecasted the effect of age-related durations in coming years.  I had expected the problem to worsen.  But, it appears that the lower labor force participation of the older groups helps to mitigate this problem.  The data used in the forecasts back tests accurately, but it predicts that the effect will remain roughly at the current level, with unemployment rates elevated by approximately .25% to .5% over the course of the business cycle.  It will remain to be seen if future generous EUI policies confirm my other estimates.  I will note that some of the expected extra duration would have been in place during the recovery period of the mid-2000's, which I count as more evidence that a very strong labor market was masked by demographic factors.  I've already discussed how the demographic adjusted labor force participation rate was higher than it appeared.  I think we can also say that the age adjusted unemployment rate was slightly lower than it appeared at the time.)

These numbers are slightly higher than the broader guesses I made the other day.  I assumed before I ran the numbers that these methods would show a smaller effect, at least on the aging issue.  My working title for this post was "Ok, maybe it's not ALL demographics."  I'm surprised by how strong the effect seems to be.  I think it's easy to forget that in unemployment numbers, we are making a big deal about a couple percentage point changes.  It shouldn't be surprising that grand generational changes might loom pretty large over that.

Here's the kind of reminder that makes Scott Sumner slap his forehead: The EUI was instituted in June 2008 when the unemployment rate was 5.6%.  At the time, the Fed Funds rate was still at 2%.

I'm planning on revisiting minimum wage in a separate post, soon, to see if the clues on that issue are stronger or weaker than my original estimates.


  1. You know your projects stand out of the herd. There is something special about them. It seems to me all of them are really brilliant!
    Insurance leads