Is the Job Vacancy Debate (Beveridge Curve) Off-Target Because The Data Is Skewed by a Move Online? Likely Yes.

Unfilled jobs – the “vacancy rate” – has gotten a lot of attention in the inflation debate. But that debate ignores an obvious discontinuity in the underlying data series. The data have definitely changed drastically around COVID. Some of that may reflect real economic change, but some of it may simply reflect a technological shift to online job listings. If so, forward-looking analysis based on older, pre-online data is likely skewed-to-the-point-of-useless before the conclusion (positive or negative) can even be considered.

There is a vigorous economics debate around the relationship between inflation and unemployment (Phillips Curve) and job vacancies (Beveridge Curve). Larry Summers and Fed Governor Waller got into a spat over it. Different papers (see below) have predicted either doom/gloom or “nothing much at all” from the same data set.

But the debate assumes the data series itself is still reliably reflecting the same trends in the same way. It is obvious the data series changed drastically in the COVID era (see chart below). So what drove that drastic change?

  1. A lot of jobs have gone unfilled (what the data are supposed to measure). Anecdote supports that.

  2. A lot of job listings have gone online (and likely skewed the data series). Anecdote supports that too.

The shift to online has likely driven a permanent, stair-step increase in the vacancy rate. How badly has it skewed the data vs prior history? I have noooo idea. But the reported vacancy rate started creeping up in 2016-2017 and then exploded after COVID. That coincides with smartphone adoption and the rise of regular Joe job sites like Indeed (LinkedIn for the bottom 75%).

The marginal cost of posing a “vacancy” online is near-zero. Businesses aren’t paying to run classified ads in the paper anymore. Even before COVID, that upwards trend from 2016 on reflected solid anecdotal evidence of companies posting “just in case” vacancies for that needle-in-haystack candidate without much expectation of success.

Wanted: Left Handed machine tool operator. Fluent in Japanese. Sight-reading of music and proficiency in the Tuba strongly desired. Must be willing to relocate to Iceland.

At the same time, a lot of high-turnover job vacancies – retail sales, food service, etc – moved online. These vacancies were always out there, but filling by word of mouth or a “Help Wanted” poster in the window. I have no idea how well the JOLTS survey was capturing that data before 2016, but I’d guess a lot weren’t showing up in the actual data series. The data series is better for counting them now, but likely non-comparable to the days when classified ads ruled the land.

So we have likely seen a permanent up-shift in the data series. Some percent of recent vacancies are probably not “new” vacancies. Just “better capture of vacancy data.” This is great from a data collection perspective. But it undermines any statistical analysis that assume the current data are consistent with prior period data.

I do not exactly know how the JOLTS vacancies data is calculated, so I am just guessing at the skew. I do think the post 2016 shift suggests some impact from the shift online. The post-COVID explosion in the data series is harder to guess at but we know a lot of activity was forced online. Meaning it may be technology-driven measurement shift as much as an economy-driven shift. Meaning we maybe can’t rely on any of these papers. Which is particularly scary given that serious people like Larry Summers and Fed Governor Waller seem to put a lot of weight on the vacancy analysis without considering the underlying data.


Some Recent “Vacancy” papers. None of them mention the obvious data skew in the chart above.

Federal Reserve Bank of San Francisco “Finding a Soft Landing along the Beveridge Curve,”

September Brookings paper: “Understanding US inflation during the COVID era,” This paper argues the case that the unemployment rate is a flawed metric to follow as long as the number of job opportunities exceeds the number of job seekers.

This entry was posted in Uncategorized. Bookmark the permalink.

Comments are closed.