Monday, July 22, 2013

Demographic distortions in the Unemployment Rate

There has been a lot of public discussion about the unemployment rate, and how it would be higher, if not for dropping labor force participation.  I will argue here that the unemployment rate is overstated, and that, notwithstanding anemic hiring, the properly adjusted unemployment rate is probably more or less recovered.  We are still 2-3% from what would normally be considered a fully recovered unemployment rate.  I think that a decent portion of that could be explained with the following factors, in reverse order of importance:

I have added some additional analysis on some of these issues here.

1) Federal regulatory obstacles.

     All unemployment is, in the end, a product of social and political frictions and searching costs.  There is a lot of anecdotal evidence that Obamacare has added to the list of uncertainties and frictions firms must address in hiring policy.  I will not assign a weight to this effect.  Even though the anemic hiring rate coincides with the law, I haven't found any statistical smoking gun that would allow us to separate it out from other factors in the economy.  And, this kind of effect probably plays out more slowly through slightly reduced growth in hiring.  The late increase in part time employment, which coincides with the early phases of implementation are reported to be signs of its effects.  (ed.: reports of increased part time employment appear to be overstated.) But, all in all, I think the most we can say is that the effect is probably to slow down recovery and to increase the unemployment rate at full employment, but difficult to measure.  It would explain higher corporate profits and slower growth, as firms would require a higher return on investment in the face of politically imposed uncertainties and liabilities.

2) Minimum Wage Hikes in 2007 -2009

     Political Calculations has done a lot of work on this.  Although, I don't think they have done enough to break out business cycle and cultural effects from minimum wage effects.  Congress has a knack for enacting minimum wage hikes just before an unemployment crisis hits...or is there some causation happening there?  That seems too easy, but I don't know if it's out of the question.

If we go to the bls, we can see the following changes to the minimum wage workforce:

Before the hikes in minimum wages, 2.2% of all workers earned minimum wage or less.  After the hikes, in 2010, that reached 6%.  Most of the change was among younger workers, with 15.2% of workers under 25 years and a full 24.9% of workers under 20 years working at minimum wage or less.  A price floor under a full quarter of the teen workforce is going to create a sizable unemployment problem.  Much of this doesn't appear in the unemployment rate, as younger workers are simply forced out of the labor force, so unemployment understates this problem.

A regression of total employment with minimum wage rates from 1972 to 2013 shows a correlation of a 6.1% reduction in total employment for every doubling of minimum wages.  The three minimum wage hikes were each 10-14% increases, totaling 41% over three years.  In employment terms, we would expect this to reduce the number of minimum wage employees by about .75% each year, for a total loss of about 2.25% of the total workforce.  What we see here among the full population is a rate of 2.3% of the population working at the wage floor in 2007, increasing to 6% in 2010, and then falling to 5.2% and 4.7% in 2011 & 2012.  (It's important to note, this is not an unemployment rate.  This is the percentage of workers who have jobs and whose wages are pinned at or below the wage floor.  So, the change in the percentage is mostly a function of the level of the minimum wage compared to the range of market wages.  Imagine a normal distribution with the far left tail cut off.  The cutoff portion would be the jobs effected by the minimum wage, and the cutoff point would move to the right as the minimum wage was increased relative to market wages.)

So, we would expect a growing economy and wage inflation to slowly reduce the proportion of workers at the minimum wage over time.  The Fed has been doing a poor job of creating inflation, so this effect has not been as great as it could have been, but we would expect this to slowly decline over time, with a stagnant minimum wage, and then to jump sharply with minimum wage increases.  In 2003, 3% of the total workforce was working at minimum wage, and it was declining slowly each year as the general increase in market wages kept moving the minimum wage further to the end of the distribution.  If we don't take this effect into account, we might expect about .75% of the workforce each year when there were hikes to lose their employment.  If we extrapolate from the last two years' trend of decreasing quantities of minimum wage workers over time to estimate .5% to .9% of the minimum wage workforce moving to non-minimum wage jobs each year, we can construct a picture of the minimum wage workforce over the past few years that looks like this:

Again, I'm not aiming for dissertation quality statistical analysis here, but just a ballpark figure of what we might expect from some basic estimates.  (There are some minor adjustments that would need to be made for changes in the labor force, etc., but I think the effects would be minimal for my purposes here.)  If we can avoid further minimum wage hikes, we can expect the percentage of minimum wage workers to decline to an asymptote around 2%.  Considering the huge number of employees caught up in this price floor, attributing the proportion shown in the graph to job losses doesn't seem outrageous to me.

Separately, I have analyzed unemployment rates regressed against minimum wage hikes, going back from 2009 to 1954 for white unemployment and 1972 for black unemployment.  That analysis showed a correlation of unemployment rates to a doubling of the minimum wage of:
     1.7% for white workers
     3.7% for white teen workers
     10.7% for black non-teen men
     18.04% for black teenagers
We can see the strong trend of minimum wage hikes being especially harmful to the most vulnerable population of workers.  If we just use the figure for white workers, we would have expected the recent minimum wage hikes to have created about a .25% shock to unemployment each year, totaling about .75% overall.  Of the 2.25% of minimum wage workers who we assume lost their job, this would imply that 1.5% left the workforce and .75% are now categorized as unemployed.  And, we do find that among workers under 25 years, LFP peaked at the end of 2006 and bottomed at the end of 2009, which coincides with the timing of minimum wage hikes better than with the financial crisis.  The decline in the under 25 labor force over that time amounts to about 1.1% of the total workforce (1.7 million workers).  This fits pretty well with the numbers above, since this age group accounts for about half of minimum wage workers, and takes the brunt of the damage when the wage floor is raised.

For teens, total hourly workers went from 5.7 million in 2006 to 4 million in 2010.  Total teens working at minimum wage went from 436,000 to 994,000 over this time.  Additionally, 1.7 million teens left the labor force as a result of the minimum wage hikes and the recession.

For all workers, total hourly workers went from 76.5 million to 72.9 million workers.  Total workers at or below minimum wage went from 1,692 to 4,361.

So, 2,669,000 workers who retained their jobs were caught up in the rising price floor.  If we apply the 1.7% UE correlation to the total labor force, we would estimate that an additional 1 million workers lost their jobs as a result.  Additionally, 3.6 million workers left the workforce.  We would attribute 2.6 million of workers leaving the workforce to the minimum wage hike.  I will note that for the entire population, during the time when these 3.6 million workers left the hourly workforce, there was no decline in the total labor force.  This is what we might expect to see if minimum wage hikes were to blame for the decrease in hourly workers, since the minimum wage wouldn't effect the employment conditions for non-hourly workers.

     7,961,000     Total Number of workers effected by the minimum wage increase, composed of:
     4,361,000     Number of effected workers who still held their jobs
     2,600,000     Number of workers who lost their jobs & left the workforce as a result of MW hikes
     1,000,000     Number of workers who lost their jobs & are unemployed as a result of MW hikes

So, to get to the number I am estimating for an unemployment effect, these are really very conservative estimates of the effect on employment, considering we have seen a 40% increase in the wage floor.  Only 13% of the effected workers would show up as unemployed.

So, we are left with an estimated shock of about .75% to the unemployment rate, which is declining at some rate over time as these workers find their way back into the workforce.  It's hard to say what amount of this effect still remains, but it would be helped by inflation.  One of the main benefits of Fed accommodation would be to alleviate price and wage stickiness in general, no more so than when our benighted leaders make a legal wage floor.

3) Emergency Unemployment Insurance and Demographics

     The following analysis is not a normative argument for or against EUI.  There are 1.6 million workers currently receiving it, and there are 1.6 million different circumstances.  The following analysis should be taken as an attempt to honestly recognize the effect of the program on stated unemployment, not as a political statement for or against the program.  I suppose this is one of those areas where a speculator can earn profits by noticing the thing that is obviously true but that you're not allowed to believe in public.  EUI will have some negative effect on employment.  Approximately 1.6 million workers are still on EUI.  Here is a GAO report on the long term unemployed.

This figure, from the report, shows that, in an economy near full employment, approximately 1/2 of long-duration unemployed will be employed in the year after insurance runs out, approximately 1/4 will be unemployed, and 1/4 will leave the labor force.  (This is a point in time estimate.  Many of the persons listed as unemployed were employed at some point in the year, temporarily.)  At the height of the crisis, these quantities had changed to about 1/3 employed, 1/5 not in the labor force, and 1/2 unemployed.

As an aside, the narrative of desperate workers leaving the labor force is not supported.  In relative terms, the poor economy did not lead to a higher percentage of long term unemployed leaving the labor force.  (This is covered in more depth in the next section.  There has been no mass flight from the labor force.)  In the trends that have been in place for the past three years, this is what is causing the Non-EUI long term unemployed level to remain level while the EUI level continues to decline.  Some non-EUI unemployed workers are finding jobs, but they are being replaced in that category by unemployed workers who are reaching the end of EUI benefits.

But, as a first blush, if EUI was phased out, we can estimate that after some adjustment period, of the 1.6 people currently on EUI, 500,000 to 900,000 would find employment, 350,000 to 750,000 would be unemployed, and possibly 300,000 will have left the labor force.  Of course, there would be some employment competition with the currently unemployed, which would lead to some more unemployment in the uninsured population.  But, on the other hand, the numbers from the later portion of this survey were from a time where the labor market was much worse than it is now, so that the number of workers who would remain unemployed would probably be lower than it was in January 2010.  This suggests that unemployment is still being inflated by more than .5% by the existence of EUI.

It takes some digging to get the BLS data on unemployment duration.  I haven't gotten the fully detailed data myself, but, fortunately, I came across this item:

Whtat this shows is that older and more educated workers tend to take longer to take new employment.  Even at the site where this is posted, it is attributed to the notion that older, educated workers have a harder time fitting into the labor force.....(in exaggerated whisper) because we're not supposed to say that people who are not employed have any discretion in their life choices, even though we all have close friends and family who have resided in these categories and who had discretion about their working decisions.  Shhhhh.  If you tell anyone I said that, I will deny it.  (clears throat and looks to left and right).  I have color coded the chart to make clear - duration of unemployment is longer for older, more educated workers.  I didn't suggest that this is because they might have their mortgage paid off, or have children that have grown up and moved away, or some savings, or the ability to work temporarily as a consultant in their old industry.  And it's absolutely not because after decades of working and saving, they don't immediately need to work, but simply enjoy working when conditions present them with an especially interesting job.

     As the source for this table put it:
"So what does one do with the over-45s with a BA or higher? ... The current mantra is 'more education is good for you' but this shows that it can, in the long run, hurt you......  It is tough to find a job, especially if you are older and better educated."
I've also noticed that you never see Tom Hanks in bad sitcoms any more....It must be because becoming an international movie star, ironically, makes it harder to get work in TV.  We should warn young actors about this downside!

The GAO report referenced above adds that about 1/4 of unemployed workers who exhausted insurance received some type of social security or disability insurance.  In fact, of the survey respondents who had exhausted UI in 2009, more were on social security and disability than were on SNAP type programs.

Here is a link to the 2012 data on unemployment duration with some more detail.  In line with the other statistics which shall not be spoken in polite company, this shows that in addition to older people, married people remain unemployed longer.  In ethnic terms, white and Hispanic workers have shorter unemployment durations, while black and Asian workers have longer durations.  There appears to be little difference due to gender.

This link to 2006 shows similar patterns that predate the recession.

And, to finish the topic of unemployment duration, here is a graph of unemployment of more than 26 weeks for three age groups.  We can see that while the 25-34 year old group reached a very high level, it is declining very quickly, and is now below the peaks reached in the 1980-82 period.  The 55-64 group and the 65+ group are both at levels simply unrelated to anything that has been recorded before.  The 65+ group, which consists of Medicare and Social Security eligible workers barely appears to be declining.

The combination of recession conditions and aging baby boomers has created the double whammy effect of (1) more people in the age group and (2) higher than normal labor force participation than past generations.  The longer durations of unemployment coming from this doubly inflated age group, further enabled by an unprecedented EUI policy, are creating large numbers of unemployed workers whose employment status is not so easily categorized.  I will cover that in more detail in the following section.

I have one last set of graphs, to compare recessions:

These graphs compare the duration of unemployment of this recession to previous recessions.  In the graph, the levels are shown as a percentage of the level of <5 week unemployment, so that they are normalized to the level of unemployment churn for each time period.  The durations of the current recession are in a different ballpark compared to earlier recessions.  I am not attributing that all to EUI.  In addition to frictions I have mentioned above, we are in a very low inflation environment, which can make it more difficult for wages to adjust.  All the previous recessions were in higher inflation environments, especially the 1980-83 recession, which might otherwise be the most comparable recession to the current one, in terms of severity.  The Fed was restrictive at the time, but inflation rates at the time were very high, so the tight monetary policy would not have created a sticky-price issue.  In any case, there are many factors we can attribute this too, but this is the effect one would expect from a generous EUI policy in any environment.  The unknown is how much to attribute to it.

But, given these caveats, compared to durations three years after the peaks, long duration unemployment in the current cycle is inflated by more than 4 million workers, almost all of them at durations above 26 weeks.  In order to estimate an overstated unemployment rate of .5%, we only need to attribute about 750,000 of the unemployed workers to this issue.  There are that many unemployed workers, just among 55+ year olds who have been unemployed for more than 26 weeks.  There are 1.6 million workers still on EUI and another 2.7 million workers who have been unemployed for longer than 26 weeks, many of them former EUI recipients.

4) Demographics

     Right off the bat, I would like to dismiss the whole notion about the Labor Force Participation Rate falling because of cyclical issues.  Here is a graph of the LFP back to 1948:

That looks really scary, especially if you cut off the chart at 2003 and shrink the scale to 62 to 67.   If we ignore the 20th century, it looks like we have a God-given LFP rate of 66%, and that the world is collapsing beneath us.

Now, let's separate this by age and sex:

Now, we can see that the aggregation creates a false image.  Do you see a crisis in 2008?  I don't.  Maybe a step down among 16-24 year olds related to school trends and minimum wage legislation.  Other than that, we are looking at decades-long trends.  If we look closely enough, we can see a small blip downward.  In fact, to me, it looks like the disaggregated LFP rates turned above trend in 2005-2007, and returned back to trend through the crisis.  It looks like pretty standard cyclical movement around the long term trends, and any recovery in the small cyclical dip that does exist will, at best, flatten the secular decline for a bit, but it is unlikely to cause an uptick in LFP.

The female graph is especially interesting, because we can see the effect of the baby boomer generation - not the population, because this is just a ratio, but the effect of new norms for women in the workplace.  In the 1960's, we see LFP dip as women enter their 20's, then slowly rise as they go through their 30's & 40's, before dipping again in old age.  But, as the baby boomers age, we see a boost in LFP among women in their 20's in the 70's, and as they turn 30 in the 80's, 40 in the 90's, and so on, we see LFP bump up in those age groups.  In fact, the line for 55+ is actually Employment to Population ratio, because the BLS doesn't even have ready data for that age group yet at their main website, as far as I can tell.

Whereas 90% of middle aged men are in the workforce, with a slow long term decline, it looks like women have topped out in the mid-70%s, and are also now in a long term decline.  But, what is really screwing up our unemployment data is the 55+ group.  The behavior in this group is an anomaly.  What appears to have happened is that LFP declined into the 90's, possibly as people were more able to fund retirements, then the trend reversed.  The trend seems to have leveled off right about the time of the recession.

The unemployment rate, as I will show, is unusually high for this group.  Is that because the 15% to 20% of the elderly workforce that is above the previous trend line is a kind of work-for-leisure population that is working for non-economic reasons?  That would mean that unemployment is overstated, and that it might follow cyclical correlations, but that it represents a fairly arbitrary categorization.  These problems are amplified by the fact that in addition to these changes in the rates of workers, the populations in the age groups that present the problem are bulging.

 We can see in this graph how much the age of the workforce has changed over time, especially in the past 20 years.  The behavior of near retirement and retirement aged workers may not tend to increase the unemployment rate at full employment, but the tendency for older workers to have longer durations of unemployment will tend to make the recovery phase of employment appear to be slower.  The effect may be to elevate the unemployment rate until later in the recovery phase of the business cycle.

This graph, which shows the unemployment rate specific to each age group, should be viewed in conjunction with the earlier graph showing labor force participation rates.  (This graph differs from the previous graph, in that the previous graph shows the level of unemployment within each age group as a percentage of the total labor force.  Those percentages would add up to the total unemployment rate, while the weighted average of the percentages in this graph would equal the total unemployment rate.)

We can see how in the earlier years, when LFP among older workers was higher and living standards were lower, unemployment followed a similar trend from 25 years through to retirement age.  As we entered the period from the 1970's into the 1990's, unemployment rates were successively lower as we moved up the age groups, reflecting their relative value and seniority in the labor force, and the gaps between age groups tended to remain through the phases of the business cycle.  This pattern seems to have changed slightly in the last cycle.  The gap was there at the peak of unemployment, but, unemployment among the older age groups seems to be more persistent.  If the now increasing level of labor force participation among retirement age workers is a sign of lifestyle choices, and longevity, we might be seeing the effect of arbitrary classifications from older workers who would like to work, and who might value some additional income, but who do not have immediate monetary needs, and who would have been out of the workforce in an earlier generation.

If we disaggregate the 55+ population, we can see that the dip in LFP there is probably a remnant of the baby boomers moving through the age categories.  The leveling off of the 55-64 age group is likely the result of the densest portions of the baby boomer group moving through the category, so that the average age of the 55-64 group is increasing.  Now that more of these workers are moving into the 65+ category, that category is continuing to show the rising LFP reflective of this generation in general.

Looking back at the age specific unemployment rate as a percentage of the total, the 45+ age group accounts for about 2.5% of the current 7.6% unemployment rate.  As shown earlier, this group has a duration of unemployment about 20% longer than younger workers (about 35 weeks compared to about 29 weeks), which increases among older workers with more education.  If we attribute this added duration to labor flexibility, we can estimate that .5% of the current unemployment rate reflects these demographic factors.  Obviously, this is a very rough outline of the issue.

1) Cause:     General regulatory framework & Obamacare
     Effect:     Undetermined, but probably slows employment growth and lifts level of full employment.

2) Cause:     Minimum wage hikes
     Effect:     up to .75% increase in unemployment at height, diminishing over time, especially if we can get more inflation.  Should not effect full employment level, unless more hikes are enacted.

3) Cause:     Emergency Unemployment Insurance
     Effect:     possibly .5%.  The number of EUI recipients is quickly declining, at about 1% now.  This will naturally decrease over the next year, and eventually the program will be stopped.  This will have no effect on the long term level of full employment, except for some possible effects of hysteresis from incentivizing workers to delay employment.

4) Cause:     Demographics
     Effect:     .5%.  This probably doesn't effect the level of peak unemployment significantly or the level of unemployment at full employment, but it might create an inflated level of reported unemployment during the recovery period.

In total, factor #1 might mean that full employment is above 5%.  Most analysts have raised their estimate of unemployment at full employment levels.  And, if we removed factors 2-4, we could already be at an unemployment rate reflecting true cyclical effects at somewhere around 6%.

I suppose these effects should cause us to expect the unemployment rate to continue declining without deceleration over the next one to two years, but for the unemployment rate to still appear to be higher than normal when we enter a period where employment gains might be reflecting unsustainable levels of growth.  Possibly, this could lead to inflationary pressures before the Fed recognizes them, although I don't believe that cost-push inflation is a proper explanation for inflation.  So, to be honest, despite all the effort I just put into this, I am not clear on the implications.  There are obvious correlations between above-trend employment, inflationary pressures, and subsequent Fed tightening, but I don't have a theoretical framework that pieces that all together with the story I've concocted above to my complete satisfaction.

Short version - Issues 1, 2 & 3 are probably hurting economic growth. #4 hurts economic growth because of an aging populace, but is mitigated by the tendency of the current aged population to work more than previous generations.  All of them are hurting the headline unemployment numbers.  And, while I'm not sure of this, this might lead to an inflationary period as unemployment drops toward 6%.


  1. I think the LFP is the key indicator. In a sustained robust economy, it will settle around 67 percent (it could fall a little as our pop ages).

    I do not think 5 percent of the population decided to stop working in 2008, for any reason another than they could not get jobs. If you tried looking for work in those years, you might know what I am talking about.

    Addtionally, there cannot be more people employed than jobs offered. Were US employers offering more, or less, jobs starting 2008 (the greatest contraction since the Great Depression).

    That said, I think Obamacare is a bad idea, and the minimum wage increases should have been put on hold, and UI should halt after 26 weeks.

    But! And this is a big but---you gotta have a very aggressive monetary growth regime in place.

    1. This comment has been removed by the author.

  2. Thanks for the comment, Benjamin.

    I think we mostly agree. Monetary policy is very important. There was a legitimate severe labor problem in 2009 & 2010.

    I'm trying to figure out the labor market of 2013 & after. I'm not that concerned about what Tyler Cowen would call mood affiliation. I'm just trying to figure out where to place my bets.

    Maybe it will all wash out, because the Fed is being too hawkish and isn't accounting for the low natural real interest rate, so if they overestimate the unemployment rate, maybe that will just pull them back into a more reasonable accommodative stance.

    On the other hand, we could see another round of overheating and then a hawkish overreaction, because I think some of these issues were in play in the last cycle when that happened.

    For instance, I really think it is useful to look at the LFPRs for each gender and age group. Once you account for aging demographics, you see that there are very long term stable trends of decreasing LFP plus a tremendous demographic effect going on. The LFPR of 66% in 2007 was way above trend once you factor in the demographics. I don't think we'll ever see 65% again, let alone 67%.

    Nobody will believe that because everybody is convinced that growth was slow in the oughts for structural reasons, when it was mostly demographics. I agree that there weren't jobs in 2009. But, when we're looking at 2006 or past 2013, I think there is a tendency to understate the labor pool's effect on the labor market.

    Sorry to go on so long. I'd love to hear your feedback.