>

Is There Really A Mystery Why More Jobs Are Not Being Created?

by Guest Author Menzie Chinn, who is Professor of Public Affairs and Economics, Robert M. La Follette School of Public Affairs, University of Wisconsin, Madison, WI – Vita.  This article originally appeared at Econbrowser 05 August 2011.

Professor Scott Sumner says “No more jobs mystery. Period. End of story.“  I’m not so certain. 

From the post:

If I hear one more discussion of the mysterious lack of jobs I’ll explode. The new GDP numbers are the final nail in the coffin. For years I’ve been saying there is no jobs mystery. That any deviation from Okun’s Law was minor compared to the scale of the output collapse. With the new RGDP figures we now know I was right, there isn’t and never was any mystery as to why there are so few jobs. RGDP is very low. Period. End of story.

Professor Sumner couches his discussion in terms of real GDP growth and changes in the unemployment rate in this post. It may be that he is right that there is no mystery once one uses post-revision data in that context *, but I prefer to focus on employment growth, rather than changes in the unemployment rate which is a function of both employment growth and labor force participation rates.

Turning to the idea that lower GDP estimates would then make the lackluster employment growth understandable, I’ll say that this story has a lot of intuitive appeal. But it isn’t so simple once one looks at the data. The relationship between contemporaneous growth rates, using the pre-revision and post-revision data, was illustrated in as in this post. In order to make the strongest case for a shift in the relationship in favor of the Sumner thesis, I ran regressions of the first difference of log employment growth on the lagged first difference of GDP growth, over the recovery period. First, the pre-revision, and second the post-revision GDP data, over the 2009Q3-2011Q2 sample (HAC SE, bold face indicates significance at 5% level).

I plot the data (blue circles post-revision, red + pre-revision) for this sample.

Figure 1 above: Scatterplot of lagged private employment q/q growth (blue, post-revision; red pre-revision) against GDP growth q/q growth (not annualized). Growth rates calculated as log differences. OLS regression lines over 2009Q3-11Q2 sample. Source: BEA, BLS, author’s calculations.

What the regressions and the graph indicate is that with the post-revision data, employment growth is less — not more — positively related to GDP growth than was the case with pre-revision data. This point is illustrated by the flatter slope of the blue line relative to the red line. (It is true that the threshold for positive private employment growth is a little lower, but I’m not sure that saves the story).

By the way, with such a small sample, changing the sample alters the slope coefficient considerably. If I drop the employment growth observation for 2009Q3 (which is plotted against the 2009Q2 GDP growth rate), the slope coefficient becomes negative (albeit not statistically significant). In addition, if I look at contemporaneous changes, the slope coefficient is negative for both pre- and post-revision data(!).

Hence, it’s not clear to me exactly how the data revisions make the job creation puzzle disappear. In fact, once drops nonlinearity, and let the data speak with respect to the relationship between lagged GDP growth and employment growth, the mystery only deepens. I use a nonlinear regression technique, which fits a smooth line to the scatterplot.

Figure 2 above: Scatterplot of lagged private employment q/q growth (blue, post-revision; red pre-revision) against GDP growth q/q growth (not annualized). Growth rates calculated as log differences. Nearest neighbor/Loess regressions, or local linear regression, using a bandwidth of 95% of the sample. The estimates use Tricube weighting, and Cleveland subsampling of the data. Source: BEA, BLS, author’s calculations.

I guess to the extent that a portion of the curve is less negatively sloped, perhaps the mystery is partly resolved. But I think that what Professor Sumner has in mind is that the level of employment should be higher given what we thought GDP was, or GDP was actually lower than what we thought. To answer that sort of question, one has to include levels. I do that by estimating an error correction models over the 2000Q1-09Q2 period, then using those regressions to forecast dynamically out-of-sample. If the Sumner thesis is correct, then the degree of overprediction of employment should disappear with the revised data.

The estimating regressions are (for pre-revision and for post-revision data, respectively):

Both regressions fail to reject Q-statistic (4, 8 lags) tests for serial correlation, and fail to reject tests for no-structural breaks according to 1-step ahead recursive residual, CUSUM and CUSUM-sq tests, at the 5% msl.

The regression results appear plausible. The long run elasticity of private employment with respect to GDP is about 0.424 using pre-revision data. It’s about 0.416 using post-revision data; both a little bit higher than the 0.37 elasticity of total nonfarm payroll I obtained in this post.

Using these equations, I generate dynamic forecasts of (log) private employment.

Figure 3 above: Log private employment (blue line), conditional forecast using pre-revision GDP data (red line) and +/- 1 standard errors (gray lines), conditional forecast using post-revision data (green line). NBER defined recession dates shaded gray. Source: BLS, BEA, NBER and author’s calculations.

The results in the figure highlight that taking into account the levels of GDP as well as the growth rates, the model indicates an overprediction of 0.7% (log terms) in 2011Q1, using pre-revision data. If one uses the post revision data, the overprediction is 1.1%. Whlie this seems like a counter-intuitive result, it’s important to recognize that all of these are within the +/- one standard error band for the pre-revision data-based forecast. That is, the relationships are badly enough estimated that this overprediction could happen just by chance, and might not necessarily indicate a structural change. (Note: The forecasts are conditional on GDP; hence what I have conducted is sometimes called an historical ex post simulation).

One could argue that one should only use expansions. If I use the 2001Q4-05Q4 sample (the expansion after the 2001 recession), then I overpredict by large margins (around 7 percent) and the difference between the forecasts based on pre- and post-revision data is trivial by comparison (around half a percent).

Hence, I think it safe to say we still have something of a mystery why employment growth has been so slow, given the growth rate of GDP. One way to identify where the puzzle originates from, in an almost tautological sense, is to include productivity (Z) in the regression.

Figure 4 above: Log private employment (blue line), conditional forecast using post-revision GDP data (dark red line), and conditional forecast using post-revision data and productivity (green line). Productivity is output per hour in nonfarm business sector. NBER defined recession dates shaded gray. Source: BLS, BEA, NBER and author’s calculations.

The forecast conditioning on output per hour in the differences and level terms results in a much better fit. I don’t want to interpret this in causal terms, but rather in a kind of accounting sense: assuming a constant level of productivity is particularly problematic in the post-2000 sample. Including a productivity variable accounts for some of the misprediction.

Why the apportionment of total output toward output per hour, instead of toward number of workers, has risen is deeper question.

The July Employment Release

Aside from the relief that greeted the news, against the backdrop of other indicators of slowing activity, what else can we glean from the July release? One point is that employment growth remains lackluster, and is decelerating from rates earlier in the year. There were upward revisions in the May and June numbers, but not enough to make a discernable impact in the graph below.

Figure 5: Nonfarm payroll employment (blue), ex-temporary Census workers (red), and civilian employment series adjusted to nonfarm payroll concept (green), in thousands, all seasonally adjusted. NBER defined recession dates shaded gray. Source: BLS via FREDII, BLS, NBER, and author’s calculations.

Interestingly, the research series which adjusts civilian employment from the household survey to conform to the nonfarm payroll concept remains substantially higher than the official nonfarm payroll series.

Aggregate weekly hours is growing faster than private employment, at least on a year-on-year basis.

Figure 6 above: Log private nonfarm payroll employment (dark blue), and log aggregate weekly hours for production and nonsupervisory workers (dark red), all seasonally adjusted, rescaled to 2009M06=0. NBER defined recession dates shaded gray. Source: BLS via FREDII, NBER, and author’s calculations.

Finally, I think it worthwhile to remember that fiscal policy (both at the Federal and the state and local level) is moving toward contraction. That is, while the level of GDP and employment is higher than it otherwise would be, the direction of policy is adding a negative factor to growth (I have to say these things because lots of people have evidenced an inability to differentiate between level and gradient).

Figure 7 above: Change in private nonfarm payroll employment (blue bars), and government employment ex.-Census workers (red bars), all seasonally adjusted, rescaled. Source: BLS via FREDII, and author’s calculations.

That point is highlighted by the fact that in recent months, government employment has deducted from overall employment growth, as state and local governments retrench. Given the nature of the debt ceiling agreement, there is no reason to believe the decrease in government employment will diminish any time soon.

More on the release at [WSJ RTE], [Thoma], [CRE1], [CRE2], [CRE3], [Stone/CBPP], [Rampell/NYT], [Norris/NYT], [Ip/Free Exchange], and [Spencer/Angry Bear].

* The relationship appears to be change in unemployment rate as a function of real growth rate. If I run a regression on q/q data over the 1986-2009Q2 sample, and then forecast out of sample, it makes little difference whether I use pre- or post-revision data. In both cases, I overpredict the unemployment rate.

Related Articles

June 2011 JOLTS Shows Little Growth Pressure on Employment by Steven Hansen

Employment Friday Is Like Waking Up Next to an Ugly Woman by Steven Hansen

July 2011 BLS Employment Better – But Still Not Good by Steven Hansen

July 2011 ADP Employment Data Shows Continued Degradation by Steven Hansen

Dissecting the Employment Numbers by Elliott Morss

Is A Budget Deficit Necessary for an Economy? by Steven Hansen

What’s Up With Employment? by Steven Hansen

Is the Stimulus Effecting Unemployment Claims? by Steven Hansen

College Training for Unemployment by Mike Konczal

Economic Damage Storm Track of The Great Recession by Ted Kavadas

Share this Econintersect Article:
  • Print
  • Digg
  • Facebook
  • Yahoo! Buzz
  • Twitter
  • Google Bookmarks
  • LinkedIn
  • Wikio
  • email
  • RSS
This entry was posted in Employment and tagged , , . Bookmark the permalink.










Make a Comment

Econintersect wants your comments, data and opinion on the articles posted.  As the internet is a "war zone" of trolls, hackers and spammers - Econintersect must balance its defences against ease of commenting.  We have joined with Livefyre to manage our comment streams.

To comment, just click the "Sign In" button at the top-left corner of the comment box below. You can create a commenting account using your favorite social network such as Twitter, Facebook, Google+, LinkedIn or Open ID - or open a Livefyre account using your email address.



2 comments
demand side
demand side

If the debate is whether lower GDP causes unemployment, it is a debate on whether clouds cause the moisture in the air. Does one cause the other? They are the same phenomenon, mitigated only by deficits or private debt. Maybe I don't understand, but trying to draw a causal relationship between two dependent variables is not right.

John Lounsbury
John Lounsbury

demand side - - - There is a chicken and egg element here - something like supply side versus demand side leading the economy. You make a very good point about the co-dependencies. We all see correlations and then try to rationalize cause and effect. I am a master of that trade myself. John Lounsbury