Comments on economics, mystery fiction, drama, and art.

Friday, March 22, 2013

The Choice of Price Indexes Can Matter

I recently ran across a reference to a BLS study looking at the divergence of real wages and labor productivity growth over the past 30 years, and I decided to look at one aspect of it.  The study uses the Consumer Price Index (CPI) to deflate wages and the Non-Farm Business Output Price Deflator (NFBODef) to deflate the value of non-farm business output.  There is an alternative index for consumer prices, the Personal Consumption Expenditure Deflator (PCEDef), and I wondered if we would see (essentially) the same pattern if we used that measure of consumer prices instead of the CPI.  Let me anticipate my conclusions.  Between 1947 and 1978, there was little divergence in the rate of increase in prices regardless of which measure one chose to use.  But from 1978 to 2012, the rate of inflation calculated using the CPI has been noticeable faster, and thus real wage growth has been noticeable slower, when calculated using the CPI than when using either of the other two price indexes.

It took me a while to figure out how to put all three indexes on the same chart.  For one thing, the CPI has a base period of 1982-84, while the PCEDef and the NVBODef share 2005 as a base period.  As a result, they have markedly different starting values in 1947.  What I finally did was start all three indexes at the same value (about 15) at the beginning of 1947 (January 1947 for the CPI and 1947.Q1 for the other two) and then changed then at the period-to-period rate of change calculated from the original indexes.  Then I converted all three index values to natural logarithms.  This gave me the following chart.
Click to enlarge.
Until about 1978 or 1979, the three indexes move almost identically,  But then the CPI begins to increase at a faster rate than either of the other two.  Using it as a deflator, then, will deflate nominal labor compensation more than using the PCEDef, and therefore will show slower growth in real compensation than would using the PCEDef.  And until about 1993, the two deflators continue to move closely (but not as closely) together, but then diverge slightly, with the PCEDef rising (slightly) faster.  The following table shows annual rates of change for all three price indexes for the entire 1947-2012 period, for the 1947-1978 period, for the 1978-1995 period, and for the 1995 – 2012 period.


CPI Inflation

PCE Inflation

NFBO Inflation

















Remember that I scaled all three price indexes so that they begat at the same value.  By the end of 1978 (see the table below), the PCEDef was 3.6% below the NFBODef, while the CPI was 4.9% higher.  But over 31 years, neither of these difference seems particularly notable.  By 1990, the PCEDef was 3.6 higher than the NFBODef—but the CPI was 36.4% higher.  The divergence over that 12-year period was significant.  And by 2012, the CPI was 40.7% higher, while the PCEDef was about 7.9% higher.  So the essential period of divergence was from 1978 to 1990, when the CPI rose (as can be seen in the table above) considerably more rapidly than the other two price indexes. 




End of 1978



End of 1990



End of 2012



The problem, however, is simple.  The degree of divergence between growth in real wages and the growth in the value of real non-farm business output growth will be strikingly different if we use the PCE deflator or the CPI.  The following chart shows median real weekly earnings, with nominal wages deflated by the PCE deflator, on the one hand, and by the CPI on the other.  (The readily available data series on earnings is available from the BLS only back to 1979; however, since the primary divergence of the price indexes dates from 1978, this should not be that much of a problem.)  At the beginning of 1979, real wages calculated by using the CPI were 8% below real wages using the PCE deflator; by 2004, that gap had widened to 20%, and has been roughly unchanged since (at the end of 2012, it was 21%).  To put it another way, between 1979 and 2012, real median weekly wages calculated using the CPI rose by 4%.  Using the PCE deflator, they rose by 15.5%.  (In either case, average annual wage growth has been slow—0.1% per  year if calculated using the CPI and 0.4% using the PCE deflator.)
Click to enlarge
This is not the place (or the time) to get into why the CPI and the PCE deflator have varied so much over the past 35 years.  But they have varied.  And that variation matters quite a bit.

Friday, March 15, 2013

Some Evidence Suggesting that Structural Changes in the Economy May Be Important for Recent Changes in Unemployment

As I began to look more deeply at the data on the duration of unemployment, and on the sources (job leavers, new labor force entrants, labor force re-entrants, and job losers) of unemployment, I found additional evidence that more of what has been happening may be a result of structural changes than  had previously been willing to acknowledge.  This shows up in changes in the details of the duration of unemployment.

The data reported by the BLS allows us to identify four ranges of the (current) duration of unemployment, 0-5 weeks, 6-14 weeks, 15-26 weeks and 27 weeks or longer.  (These are the current duration of existing spells of unemployment, not the final duration of a spell of unemployment.)  In the following chart, I report an “unemployment rate”-like measure for these four duration categories.  The data reported are the number of people reporting that duration of unemployment divided by the number of people in the labor force; available data are monthly from January 1948 through February 2013 and are seasonally adjusted.
(Click to enlarge.)

What seems fairly noticeable is that, beginning about 1982, the “unemployment rate” composed of those with current unemployment spells of less than 5 weeks has trended downward, and, by 2007, was lower (at close to 1.5%) than at almost any time since World War II, lower even than in the low unemployment years of the late 1960s (when, with overall unemployment rates below 4%, this measure of unemployment accounted for about 2% of the labor force).  And, over most of the post-1982 period, the “unemployment rate” composed of those unemployed for 5-14 weeks was slso trending downward, falling to less than 1.5% of the labor force at business cycle peaks (higher than the 1% or so in the late 1960s and the less than 1% in the early 1950s, but lower than at any other time).

Longer-term unemployment, on the other hand, was not trending downward, but bottoming out at about 0.5% of the labor force both for those currently unemployed for 15-26 weeks and for those currently employed 27 weeks of longer.  This was also higher than in the late 1960s and the early 1950s.  Indeed, the “unemployment rates” for these groups at the business cycle peak (2006-2007) was as high as it has ever been at a business cycle peak (about 0.7%).

What all of this suggests is that the share of total unemployment was falling over time for the two groups with the shortest current spells of unemployment, and rising somewhat for those wilt he longest spells.  And, as the following chart shows, that is exactly what happened.  Here, the consistent pattern is that the share of unemployment in the shorter durations falls in recessions and rises in booms.  But the share both at the recession trough and at the business cycle peaks of those whose current spells of unemployment are shorter than 5 weeks has been falling since the 1980s.  (Indeed the share at business cycle peaks has been falling consistently since the end of world War II.)  At the same time, the share of long-term unemployment has been rising consistently since 1970.
(Click to enlarge.)

And in the Great Recession (beginning in late 2007), the “unemployment rate” associated with short-spell unemployment rose by an historically small amount, barely one percentage point, while the “unemployment rate” associated with long-term unemployment (15 weeks or longer) rose from about 1.4% to about 6.3%; it had not exceeded 4.5% in any previous recession, reaching that level in 1983).  And the share of unemployment accounted for by the two shortest duration groups (0-14 weeks) fell to levels not seen since the end of the War—about 13% for those unemployed less than 5 weeks, and about 17% for those unemployed between 5 and 14 weeks—of about 30% of all unemployment.  In no prior recession had the share of unemployment by those unemployed less than 15 weeks fallen below 57%.  And long-term unemployment (15 weeks +) peaked at 70%, when it had never exceeded 43% before. 

So what we observe here is an upward trend in the share of the labor force experiencing long-term unemployment (both at business cycle peaks and at business cycle troughs) and, as a necessary counterpart, a rising share of unemployment concentrated among the long-term unemployed.  If we simply look at the “average” share of unemployment before and after 1980, short-term unemployment (less than 5 weeks) fell from about 48% of total unemployment to about 37%.  Meanwhile, the average share of long-term unemployment (15 weeks or more) rose from about 22% to about 33%.  (The average share of those unemployed 5-14 weeks was about 30% in both periods.) 

It is generally agreed that structural shifts in the economy will lead to increased difficulty among at least some of the unemployed—those disadvantaged by the structural shifts—in finding re-employment.  And that is precisely what has occurred since about 1980.  Let me emphasize that this is consistent with an increase in structural unemployment; it is not a proof that structural unemployment did, in fact, increase.

(All data used here can be found at

Friday, March 08, 2013

Long-Term and Discouraged-Worker Unemployment in the Great Recession

One of the most striking features of the Great Recession, and one it has in common with the Great Depression, has been the extraordinary increase in long-term unemployment.  I want to take a brief look at this occurrence.

The following chart presents the “Long-Term Unemployment Rate” (or LTUR), defined as those unemployed for 27 or more weeks divided by the labor force.  These data are available monthly since 1948.

*(Unemployed 27+ Weeks)/(Labor Force.  Click to enlarge.

Interestingly, while the increases in LTUR were similarly sharp (but relatively small) in the 1990-91 recession, and while the LTUR rose considerably after the formal end of the 2001 recession, the subsequent decline in the LTUR was slower.  In fact, the LTUR took about 8 years after the 1991 recession to return to its previous low, and following the 2001 recession, it did not return to its previous low. 

The LTUR-Overall UR relationship is shown (below) in red for the Great Recession.  It is clearly different from the prior post-World-War II experience.  Had the LTUR increased in line with the recent past, it would have risen from around 1.5% to around 2.5%.  So it has been 1.5 to 2 percentage points higher than our recent experience would suggest.  In general, more severe recessions have resulted in higher LTURs.  In the recession of 1981-83, the LTUR rose above 2.5%, and then fell between 1983 and 1985 to a little over 1%.  Similar (but smaller) sharp rises and declines occurred in the 1974-75 recession and in earlier ones as well.  The decline in the LTUR, however, has been roughly in line with the recoveries from previous recessions—it has dropped by about 1.25 percentage points in the 3 years since it peaked, very comparable to the drop following the 1974-75 recession and just a slightly smaller drop than in the recession of the early 1980s.  Indeed the number of those unemployed 27+ weeks has dropped by about 2 million in the last three years.

Click to enlarge.

In addition to long-term unemployment, it is important, I think, to consider what has happened to the number of discouraged workers—those who have dropped out of the labor force because they believe that there are not jobs available. 

[This unemployment rate is calculated as (DW)/(LF+DW).]  Click to enlarge.

Monthly data on the number of discouraged workers dates from January 1994 (and are available only on a not-seasonally-adjusted basis), so it’s more difficult to get a complete understanding of what happened in earlier, more severe recessions (than the 1990-91 and 2001 recessions).  Clearly, though, the DW unemployment rate rose by a factor of more than 4 from late 2007 to the end of 2011 and has since declined (but not smoothly) and is not around 0.5%.  That’s still more than double what appears to be a typical full-employment level for this rate of about 0.2%.  But, then, the overall economy is far from its apparent full employment rate as well (7.7%, compared with a full-employment unemployment rate of about 5.5%). 

The apparently unusually large increases both in long-term unemployment and in discouraged-worker unemployment are one (very good) reason for being concerned that some of the present unemployment has shifted from being cyclical—a consequence of the recession—to becoming structural—a misfit between the skills of the unemployed/discouraged workers and the current needs of employers.  This may be true, but it’s important to keep in mind that the recovery from the Great Recession has been unusually tepid (with respect to employment) as well.  If we are encountering a growing structural problem, it is, I would suggest a consequence of that slow recovery.

(All data used for this post can be found at




Sunday, March 03, 2013

A new post-retirement, part-time teaching gig

I start my second post-retirement, part-time teaching gig tomorrow.  It's a course called Foundations of Economics, for entering MBA students who have not had an undergraduate introductory economics sequence.  Seven weeks, 2 (2-hour) class meetings a week.

Years ago, when I was at Illinois state University, I had a colleague who asked every job candidate we had on campus for an interview what I thought was a really good, but really hard, question:  You're teaching introductory micro economics (or intro macro), but instead of 15 weeks, you only have 7.  What do you cut out, and what do you leave in?  What I've had to do with this course is take what we usually do across 2 semesters (30 weeks, 2.5 hours of class time a week--75 hours of class time) and reduce it to 28 hours of class time.  It was not a lot of fun to figure out what to do.

We'll see how it goes, I guess