Friday, December 13, 2013

Labor force participation trends and assumptions

John Taylor references research by Christopher Erceg and Andrew Levin (this link is to an April 2013 version of the paper Taylor references with a September 2013 date) (HT: Marcus Nunes).  I am disposed to agree with Taylor that the recovery has been slow, partially, due to economic policies.  But, here he is saying that labor force participation (LFP) is low because of these cyclical pressures, and that the economy is much worse than the unemployment rate (UER) would make it seem.  My past reviews of the topic have convinced me that the LFP decline reflects some very typical cyclical movements, and that all of the unusual decline has a demographic source.  So, I took his opening comment: "Research by Christopher Erceg and Andrew Levin is providing solid evidence that the decline in the labor force participation rate since 2007 has been due to cyclical factors" as a challenge.  How could reasonable research coming from the Federal Reserve reach such a different conclusion from what I have found?  Maybe I'm getting it wrong.

I think there are two main factors that are leading to the different conclusions.  My version of the difference is:

1) The study relies on November 2007 projections of LFP from the BLS.  In hindsight, the BLS got it wrong.

2) The minimum wage increases of 2007-2009 pulled a lot of younger workers out of the labor force.  I would agree that this is bad policy, and it led to a kind of cyclical drop in LFP, but the marginal effect of that policy on the rate of change in today's LFP should be small.  The continued downward drift of LFP today is due to demographics.

LFP Trends

Two items from the paper that make this clear are:










First, we can see that the unexpected drop in LFP does not come from workers 55 years and older.  They roughly followed the trend that the BLS forecasted in 2007.

16 to 24 year olds took a huge hit from trend.  We could expect this age group to have more cyclical behavior, but I believe much of this is due to minimum wage increases.

The vast majority of the hit to LFP, in absolute numbers, comes from the 25-54 age group.  The paper acknowledges that there had been a long term downward trend in these age groups.  So, I would expect the actual drop in LFP of -1.5% to reflect the long term trend of -.5%, plus a drop in LFP of about 1%, which would represent about .5% of cyclical movements above and below trend as the economy switched from a very hot labor market with 4.5% UE to a recessionary market with up to 10% Unemployment.  But, strangely, the BLS had projected an increasing LFP for this age group.


Stacking Error on top of Error

Treating the top of the labor market boom of the 2000's as the new normal caused a dual error - forecasts from 2007 started too high, and the trend slopes were adjusted too high.  Here, I will simply compare the least squares linear trend of the long term LFP rates of each age group to the trends set by the BLS in 2007.

Here, we see that, compared to long term trends, the 2007-2012 period looks perfectly normal for 35+ year olds.  The 25-34 year old group has an unusually sharp cyclical deviation, which is now reverting to the mean.  But, the 2007 BLS forecasts all either start above the trend mean or have an inflated slope, or both.  In hindsight, in a data series with such a long pattern of linear behavior with mean reversion tendencies, it was clearly an error to forecast LFP's that would accelerate from a frothy labor market.

The trends in the female LFP's aren't as long, but in 2007, there was a good 10 years of data suggesting a very similar pattern of falling within each age group at a rate of 1% to 1.5% per decade (the coefficients in the trendline equations are based on quarterly units).

Perhaps, due to the shorter history of linear behavior, the BLS can be forgiven for expecting more positive movement from the female series.

In either case, I'm afraid that any analysis that uses the 2007 BLS forecasts as a baseline is simply measuring the understandable failure of the BLS to predict the exact timing of the business cycle.

In hindsight, the naïve linear trends appear to be a much more reasonable baseline than the BLS forecasts.  In every case, the BLS forecasts predict higher LFP's than the naïve trendlines, generally overstating expected levels by 1 to 3 percentage points by 2012, compared to the naïve trends.


Cyclical Decreases in LFP

The Fed paper includes this graph:

The regression produces the following result:

LFP = -0.40 - 0.30*UER

Of which the authors comment:

These regression results provide stark evidence that cyclical factors have been crucial in explaining the recent decline in prime-age LFPR. The coefficient on the lagged change in prime-age unemployment is highly significant (t-statistic of -3.9); that is, the state-level data exhibit a strong negative correlation between changes in LFPR and lagged changes in unemployment for prime-age adults. In contrast, the regression intercept is not statistically significant from zero (t-statistic of -0.97), indicating that the data provides no support whatsoever for structural interpretations of the drop in prime-age LFPR. In effect, the state-level data indicates that the aggregate decline in prime-age LFPR since 2007 can be fully explained by the persistent shortfall in labor demand.

 
That seems like a really strange conclusion to me.

1) The bulk of the long-term trend in declining LFP comes from workers aging out of the prime working age.  Right now, in 2013, there is a downward trend from that effect, and it will be a headwind for the aggregate LFP, regardless of what cyclical issues there are among the prime age group.  These inter-age-group changes cut about 0.65% off of aggregate LFP from 2007-2012.  I don't see how finding cyclical LFP changes within an age group addresses this.

2) Using 25-54 year olds as the prime age group is a problem, because 45-54 year olds have a markedly lower LFP than 25-44 year olds.  Right now, baby boomers are bulking up the 45-54 year old category, so there is a temporary downward trend among this group that amounted to about 0.55% during this period.

3)  They conclude that the intercept, -0.40, is not statistically significant from zero, so they proceed to attribute the entire decline in LFP to cyclical factors.  I would say that -0.40 is not statistically significant from -0.55, which is the secular decline in LFP we would have expected from this age group over this time period.  So, roughly 30% of the LFP decline among 25-54 year olds was the result of long term aging trends.

4) If statistical analysis can come that close to the expected intercept (-0.40 compared to -0.55) and still be interpreted as having an intercept that is not significantly different from zero, then I have doubts about the ability of that analysis to say anything at all.


Taylor attributes this graph to the authors:

This graph looks to me like the Unemployment Rate line assumes that the "normal" LFPR would be practically flat.

On the contrary, the graph below shows a LFP trend line based on long term age-group trends.  This downtrend is accelerating.  Currently, LFP is declining by about 0.1% a year due to long-term trends across the prime age groups, and by about 0.2% a year due to aging baby boomers.


LFP went from about 0.75% above trend in 2007 to about 0.75% below trend at the end of 2012.  This is not outside the range one might expect from a deep recession.  It is about the same as the drop during the 1980-1982 recession, but that recession has a growing LFP trend, due to the entrance of more women into the labor force, so if you don't correct for trend, it looks like there was no LFP retrenchment during that recession.

At the rate the LFP is naturally declining, we will be back to trend by early 2016, even if the Employment to Population Ratio doesn't increase at all.  We are the mirror image of the period of the 1980-82 recession.  At that time, a cyclically rising LFP rose very quickly, and a cyclically dropping LFP looked flat.  Today, a cyclically rising LFP is flat, and a cyclically dropping LFP drops like a stone.  So, comparing the 2007 inflated BLS trends to 1980 is like stacking errors on top of errors on top of errors.

I would agree that policy issues have made this worse, but not in a cyclical way.  JOLTS measures (hires, quits, openings, etc.) are still at low levels, but they have been growing at rates similar to rates in the previous recovery:

I have attributed the entire drop below LFP trend to the minimum wage.  And, the remaining unusual level of unemployment is attributable to the minimum wage and emergency unemployment insurance.  I would call these pro-cyclical structural problems.

Both demand-side Fed solutions and supply-side economic structural solutions can help a little bit, but I think only a little bit.  On the other hand, the pro-cyclical nature of these policies should eventually create some catch-up growth in the labor force and in employment.

No comments:

Post a Comment