Arnold Kling  

Momentum in Employment

Two Provocative Quotes on Educ... Which Jobs to Save?...

According to Ed Leamer's Macroeconomic Patterns and Stories, chapter 4, payroll employment is not a random walk. The change in payroll employment in August will be highly correlated with the average change from May through July. This was 26,667, according to the most current figures on the BLS web site.

As I read Leamer's chart, it would be highly unusual, relative to the period he focused on, for this to be followed by a large gain in employment in August. Only once from 1990 through 2006 was a three-month average gain this low followed by a monthly gain of over 200,000 jobs. The curve of best fit suggests a gain in August of somewhere between 25,000 and 50,000 jobs. Once the three-month average exceeds 150,000 jobs, employment growth tends to rise further, although past an average of 200,000 jobs it tends to rise less than proportionately to the three-month average.

Keep in mind that the payroll employment number is a net number. About four million jobs are created and destroyed each month, with a net increase of 150,000, give or take.

It would be really interesting to know more about the source of momentum in the employment numbers. It would be even more interesting to know more about the source of turning points. In particular, is it possible for government to create a turning point upward by using fiscal or monetary stimulus?

Also, I wonder why Leamer's chart only uses data from 1990 to 2006. The data go back much farther. Do the older data tell a different story? If so, is the current recession more likely to resemble the older data or the data from 1990 through 2006?

Well, I just downloaded the data from 1960 through the latest month and charted it. I changed one data point, though. In 1983 in August, employment falls by 300,000 and then jumps by 1.1 million the next month. I arbitrarily changed it to rise 400,000 each month. The chart looks a lot like Leamer's chart.

Next, I arbitrarily set the following cutoffs:
under 50,000 jobs gained per month = low job growth
between 50,000 and 350,000 jobs a month = normal job growth
over 350,000 jobs a month = high job growth

I have 603 observations of three-month averages followed by one-month values.

327 have medium job growth in the three-month period followed by medium job growth in the next month.
112 have low job growth followed by low job growth
63 have medium followed by low
45 have medium followed by high
33 have low followed by medium
14 have high followed by medium
8 have high followed by high
1 has low followed by high (December 1970, looks fluky)
0 have high followed by low

Folks, that is a lot of evidence for momentum.

Comments and Sharing

COMMENTS (4 to date)
Jody writes:

Mmmm.... hidden Markov models... Why not run the data set through Baum-Welch ( and see what pops out?

Various writes:

Well, accuse me of stating the obvious, but momentum means there is one or more self-reinforcing drivers here. It could be phsycological.

MikeM writes:


The probable reason for Leamer to only use data to 1990 is because of the transition from SIC coded job data to NAICS coded job data. NAICS is the standard going forward but SIC was the old way of doing things. The two are not cross compatible and if you use one to go back previous to 1990, than you must continue to use it going forward.


fundamentalist writes:

Obviously there is strong momentum, but the turning points are more important. When depressions hit, we know that low employment growth will persist for a while. But there will be a turning point and it will be a rare event. Then when growth picks up, we know it will persist for 6-8 years before hitting another wall. Something will break the momentum and cause a turning point. The data won't tell you what causes a turning point. That depends upon sound interpretation of the data.

Comments for this entry have been closed
Return to top