Churning in the Labor Market: An Introduction

 

Churning butter, circa 1935; Shutterstock

I'm posting this as another reaction to the shocking, if unsurprising, firing of the BLS Commissioner after President Trump objected to revised employment numbers. There is much to be said about this, but here we present some data that provides some context. Specifically, we look at "churn" in the labor market, a.k.a. the difference between net and gross employment changes.

In March I dropped a post on macro indicators, that included a large downloadable teaching library in PowerPoint. The post also included a few comments about issues faced in U.S. data collection, and also the now well-known result that data quality suffers in autocracies. 

If you download the library, employment slides start circa slide 920. (It's a library, not a presentation). Circa slide 685, you will find a few comments on data manipulation in Greece and Argentina. If time permits, I might later add some examples of apparent data manipulation from, inter alia, China, Russia, Turkey, Venezuela.

I retired from full time teaching in 2016 so more than a few slides are a bit dated. Today I've updated a few germane slides, from Business Employment Dynamics, and have a few comments.

Net and Gross Employment Changes: Two BLS Sources

The headline BLS employment data are net changes in employment. The scale of these data are usually, at most, plus-or-minus a couple hundred thousand workers, at an annual rate. Remember, total U.S. employment is over 160 million.

There is a lot of “churn” behind these net data. In any given month, quarter, or year, millions of workers quit or are laid off, and millions are hired, at an annual rate. It's the difference between these two large numbers that gives us something between a couple hundred thousand, or so, net gains or losses in any given month (at an annual rate).

BLS produces two series that provide gross as well as net employment data: Business Employment Dynamics (BED) data, and the Job Openings and Labor Turnover Survey (JOLTS). JOLTs is monthly, and BED is quarterly. JOLTs comes out faster than BED; BED is considered to be more accurate. (All such statistics are subject to both sampling and non-sampling errors). I used ChatGPT to help create some tables comparing BED and JOLTS:

We will use the Business Employment Dynamics data today. Qualitative results would be similar if we used JOLTS.

This figure presents quarterly gross employment, seasonally adjusted at an annual rate. Gross job gains are positive numbers, in blue; gross job losses are negative numbers in red. The black line is each quarter's net employment, i.e. the difference between the red bar and the blue bar. This black line is functionally equivalent to the monthly employment released (and later revised) by BLS, although the numbers vary a bit from the headline number, given the different survey used by BED and the headline data sources.

Obviously the black line traces out net changes in employment that are small relative to gross job gains and gross job losses.

Notice the second quarter of 2020, when the COVID pandemic hit -- gross jobs lost at a 20 million per year rate, at the same time gross job gains fell from between 7 and 8 million, to under 6. Net employment was falling at a rate of over 14 million per year, although thankfully that rate did not last very long.

After that quarter gross losses returned to somewhat normal levels, and over the next two years the job market healed.

Let's look at the same chart, but truncate the vertical axis; we can't see the Q2 2020 data in full but we get a better look at the rest of the data:

In Q4 2024, the latest BED data available (and the time of the November 2024 election), gross job gains clocked in at 7.77 million, gross job losses at 7.48 million, and the net job gain was about 287,000.

So what's our point, in the end? Given the large size of the U.S. job market, and the large amount of churn we see at any given month or quarter, it's no surprise that employment data from surveys are imperfect and subject to revision. The recent revisions ran at an annual rate of less than a tenth of a percentage point of total employment, although they were large and worthy of a closer look. There's much more to say about this, of course. A future post?