The other day, I verified Ed Leamer’s finding that there is momentum in payroll employment. This is an extremely important finding, as I will explain below.

Just as a teaser, I think it strongly supports Tyler Cowen’s ZMP story, but it poses problems for the New Keynesian model and its relatives.Suppose we are talking about the growth rate of GDP or the growth rate of employment. Broadly speaking, there are three ways that a macroeconomic variable can behave:

1. Mean reversion–if it was high last period, it tends to be low this period.
2. Random walk–it does not care what it did last period.
3. Momentum–if it was high last period, it tends to be high this period.

This is a short-horizon description. If our time horizon is decades, then all interesting macro variables are mean-reverting.

Ed Leamer’s Macroeconomic Patterns and Stories has a brief chapter 6 that discusses persistence and momentum. He writes,

when you read in the newspaper that GDP growth was the very high number, 5.6, forget it. It doesn’t matter. There is no persistence. But if you read that the unemployment rate jumped up by 0.5% points, that’s really important. That jump in unemployment is not going away any time soon.

Leamer explains the momentum in unemployment as follows:

Firms do not hire or fire all in a single quarter. They make an employment plan and phase it in over several quarters.

That story is lame, for at least three reasons.

1. I am not sure it’s true. I think firms tend to make employment adjustments, particularly downward adjustments, really quickly. You do not want to have people hanging around waiting for the axe to fall. You lay off people as quickly as possible, so that you can move on. In any case, somebody could look at the JOLTS data and see whether it’s true.

2. Even if the story is true for one firm, it does not explain the aggregate data. If firm X is in a hiring mode and firm Y is in a firing mode, they cancel each other out. If firms are independent, the law of large numbers should take any momentum out of the net hiring number. On the other hand, if firms’ hiring plans are highly correlated with one another, then that is the story.

3. The story does not explain why GDP growth and employment growth have such different time series properties. GDP growth is close to a random walk, and employment growth has strong momentum.

The story I would tell is that there are clusters of firms that interact with one another. In an expanding cluster, growth of one firm leads to growth in others. In the 1920’s, as more people were employed in building automobiles, there were bound to be more people employed at gas stations. In a contracting cluster, declines in some firms lead to declines in others. As you get fewer horse-and-buggy drivers, you get fewer horse trainers, fewer horseshoe makers, and fewer manure sweepers.

I would expect to see momentum in these sorts of clusters. Not every firm gets the expansionary or contractionary signal at once. The signals take a while to propagate.

For example, in a classic inventory cycle, cars might pile up on dealer lots. The car manufacturers have layoffs and cut production, and this leads to an inventory pile-up at the plants that manufacture spark plugs, so they have layoffs and cut production, and this leads to an inventory pile-up at the companies that supply the materials to make spark plugs, and so on.

What about the momentum in employment growth relative to GDP growth? We do know that GDP and employment tend to move together, but evidently the relationship is not one-to-one. Instead, it seems that well after GDP growth slows down, the net gains in employment stay high for a while. (Remember that the gross flows into and out of employment are on the order of 4 million a month, while the net gain or loss is only about 150,000 a month.) Similarly, after GDP growth turns up, it takes a while for hiring to pick up.

I think that this cries out for a story of Zero Marginal Product. At the end of the boom, it takes firms a while to realize that they have a problem with ZMP workers. Forty years ago, auto inventories would pile up before manufacturers realized it, and meanwhile their production workers were producing cars that they could not sell. Today, in the Garett Jones economy, most workers are ZMP in the short run, but firms take the short-run view only when they are in financial difficulty.

In 2000, folks suddenly came to terms with the ZMP of workers at many Dotcom darlings. In 2008, we came to terms with the ZMP of many mortgage securities traders and home builders.

I think that the momentum in employment casts doubt on stories that explain employment in terms of real wages. Those stories imply a lot of mean reversion. In fact, the big “contribution” of the New Keynesian model to the older macro models has been to force the modelers to impose strong mean reversion on real wages and employment in order to get past the “microfoundations” police. But the data instead tell us that there is persistence.

I think that the momentum in employment could be consistent with the “general disequilibrium” version of Keynesian economics. (Barro-Grossman, for those who remember. Or Clower.) I think it could be consistent with the PSST model of Austrian economics that I have been pushing. Clusters of interdependent firms are an example of a pattern of sustainable specialization and trade.

I suspect that the JOLTS data could be exploited to learn more about the sources of momentum in employment. I would think that my hypothesis about interdependent clusters ought to be investigated.