The prevailing wisdom within economic circles is that the US is currently enjoying a very strong labor market.

Unfortunately, this prevailing wisdom is more like epic cluelessness, and a paradoxical lack of essential curiosity that is essential for any good analysis. Despite what the BLS said in July’s Employment Situation Summary, or what Jay Powell said in his brief address at Jackson Hole1, the labor market in the US is not strong, but is toxic and particularly vulnerable.

Despite the seeming strength of the headline numbers within the Employment Situation Summary, the reality of the data is that the numbers underneath simply do not add up, not without some truly vaudevillian mental gyrations—hence my naming them Lou Costello Labor Math.

Why The Labor Market Appears Strong

The most obvious reason for declaring the labor market in the US “strong” is simple and straightforward: the unemployment rate is at 3.5%, one of the lowest unemployment marks in the postwar era.

When you look at the civilian labor force data, the labor force appears to be increasing, with the number of persons not in the labor force decreasing.

That the labor market is tight is amply demonstrated by the reality that there are twice as many unfilled job openings as there are currently unemployed persons to fill them.

Could one ask for more from an economy? Well…yes. One could (and should) ask an economy to have a consistent unemployment situation, something the US does not have currently.

Divergence Between Seasonally Adjusted Data And Raw Data

The first indication that something is off kilter within the employment data comes when one looks at the civilian labor force data—both the raw data and the seasonally adjusted data—on the same graph.

Note how the seasonal curve and the raw data curve diverge beginning around March of this year. All things being equal, that should not happen over that long a time frame for seasonally adjusted data.

Seasonal adjustment2 is a technique applied to time-series data to eliminate seasonal shifts within the data due solely to the varying seasons. Eliminating such time-related shifts eliminates statistical "noise", and helps to bring meaningful statistical relationships to the foreground.

The math behind a seasonal adjustment is simple:

1. Compute the average of a complete time series (the long term average).

2. For BLS data which is given monthly, divide each month’s data by the long term average to get the seasonal factor (which can be expressed as a percentage or multiplied by 100 to be expressed as an index).

3. Repeat for each time series available as appropriate.

4. Compute the average for each month’s seasonal factor, dropping the higheset and lowest value.

5. Divide each month’s unadjusted data in each time series by the seasonal factor to derive the seasonally adjusted number for that month.

Because seasonality is built around the average across an entire time series, it is mathematically impossible for the seasonally adjusted data to diverge from the raw data across several months—the seasonal data is not going to increase for several months while the raw data decreases, or vice versa; the seasonal factors invariably pull the data closer to the time series average. Yet that divergence is exactly what we see in the Civilian Labor Force data.

If we index the raw data and the seasonally adjusted data to January of 2020, we see the divergence much more clearly.

We see the same divergence in the data for persons not in the labor force.

Again, indexing the data to January of 2020 makes the divergence much more evident.

If these divergences are not possible, all things being equal, then we are forced to conclude that, beginning in March, all things within the labor force are not equal. “Something” has changed within the data that is overwhelming the seasonal adjustment calculation.

The Labor Force Is Growing, But Slower Than Before

To analyze these divergences, we need to break apart the years and compare the raw data month to month across the years. To eliminate variation due to an expanding labor force, we index each year’s data to January of that year, thereby focusing on the relative magnitude and direction of data shifts.