Monday, February 25, 2013

Job flows, industry composition, and the changing US economy

Net job numbers hide a lot of underlying "churning" or reallocation. It would be interesting to divide this reallocation into (a) the part that is necessary for, or aligned with, net job numbers and (b) "excess reallocation"--job flows that occur in excess of the net figure. In other words, how much shrinking and growing do establishments do that isn't necessary for net aggregate employment growth?

Predictably, this has been done already (skip this paragraph if it looks boring). This must-read paper describes a measure of "excess reallocation," originally defined (I think) by Davis, Haltiwanger, and Schuh (1996). Simply put, the measure is job creation (employment growth at expanding establishments) plus job destruction (employment reduction at shrinking establishments) minus net employment growth (JC + JD - |JC-JD|, where JD is recorded as a positive number). To get a rate, divide this by employment. This is the excess reallocation rate, and it gives us an idea of what portion of jobs are moving around between establishments in a given year.

Excess reallocation rates vary widely across industries--some industries are more volatile than others in terms of job flows. Figure 1 plots annual excess reallocation rates for a few selected sectors* (click for larger image). "FIRE" stands for finance, insurance, and real estate.

Figure 1

Note the following: Construction has typically had the highest rate of excess reallocation. Between March 1979 and March 1980, about half of construction jobs were reallocated--in excess of the net change in construction employment. Manufacturing is consistently the least volatile sector, by a nontrivial margin. This suggests that establishment size and job matches are relatively stable in manufacturing when compared with construction (or any other sector). Presumably there are a lot of factors that determine how volatile a sector's employment is; you can probably imagine some of them. Some sectors have seen declines in these rates, with construction falling by about 20 percent of employment over the last 30 years (but that still leaves almost a third of construction jobs being reallocated). Reallocation rates vary less across industries now than they did 30 years ago--are we seeing convergence? There appear to be both cyclical and secular forces at work.

People preparing to enter the job market may want to keep these job volatility data in mind.

When looking at sector-level data, it's good to be aware of how large a role each sector plays in the national job economy. Figure 2 plots sector employment as a percent of overall private/nonfarm employment for the same sectors (click for larger image). I include only even years to keep the graph readable.

Figure 2

Most readers are probably not surprised by much on this chart. Manufacturing's employment share has steadily declined, while the share of employees working in the services sector has dramatically increased from a quarter to nearly half of the private/nonfarm economy. The other sectors haven't changed a lot (but keep in mind that this is just employment--a chart of sector share of output would look different).

Note two facts from the charts above: First, services experiences higher employment volatility than manufacturing. Second, the share of employment in services has grown dramatically while the share of employment in manufacturing has declined. Is it safe to assume, then, that overall economywide employment volatility has grown? No. Figure 3 plots excess reallocation for the entire economy (click for larger image).

Figure 3

There is a secular decline in excess reallocation rates for the broad economy, and this trend is acting in the opposite direction from the secular trend in industry composition (i.e., if industry composition had stayed constant, reallocation rates may have fallen even more). It turns out that looking at simple job creation and destruction rates reveals a similar trend. In other data sources, the trend can be seen going back even further in time. These trends have already been documented in the academic literature and other places (e.g., here or here, with citations). It's difficult to explain the trend by appealing to composition effects (though changes in the firm age distribution help). It's a bit of a puzzle, and it's the subject of some of my current research (with coauthors who know far more about it than I do).

Broadly speaking, job market dynamism has been declining. The amount of churning that goes on behind net employment growth has been falling. It's not clear whether this is a good thing or a bad thing--reallocation imposes costs on workers (and firms), but it is also one of the key mechanisms behind productivity growth. Without knowing what's driving the trend, we can't know whether it has any policy implications.

*The BDS uses SIC sector definitions rather than the newer NAICS standard; and in general these categories don't line up exactly with the industry classes used in the popular payroll surveys (i.e., CES). I am not sure of the method used to roll SIC codes forward past 2002. Also, my understanding is that the FIRE category may be somewhat less reliable prior to the early 1990s. If you want to become utterly cynical about economic data, ask someone who compiles and releases survey or administrative data how they feel about industry codes.


  1. I truly appreciate this post. I’ve been looking everywhere for this!
    manufacturing industries

  2. What about government services? Included with the other services? I bet state and local government services have expanded, and federal US services have stayed constant.

    1. Hi Ray--these data only cover the private sector. Interesting question, though.