Littlest recession getting bigger?

A lot being read into an uptick in headline unemployment

Headline unemployment bumped up 0.1 pp — the measurement resolution of the series — and now people are noticing the slow upward (or better to say “not downward”) trend. It has been deviating from the dynamic equilibrium rate of

\(\displaystyle \frac{d}{dt} \log u \simeq - 0.09/\text{y}\)

for a about a year. It’s entering the low end of the FOMC projections from September 2022:

“Core” unemployment¹ has been showing this more clearly for the same amount of time. In fact, the deviation has been sufficient for the recession detection algorithm² to call what I call “the littlest recession”:

While it’s possible that core unemployment could continue to rise, the recent data appears to show “the littlest recession” is already over. There’s no sign in other data that there’s a negative shock to the economy besides the one we have been experiencing since 2022. In fact, initial claims still appear to be at a (hypothetical!) “noise floor”³:

JOLTS measures haven’t shown any major changes — they are still in the latter part of a non-equilibrium shock that’s been underway since 2022. I’ll just show the quits rate as the others show basically the same structure for recent data:

However — based on the past few decades of data, we should be expecting unemployment to continue to rise. It just … hasn’t. At least not as much as would be expected:

There are several possible explanations for this discrepancy:

1. The relationship between JOLTS and the unemployment rate was spurious

2. A relationship between JOLTS and the unemployment rate exists, but is not as regular as it has been for the past few decades — the littlest recession could be all we get.

3. As the rate of change in the JOLTS data (a metric where we have limited data during recessions) for the recent negative shock was higher than has been observed previously, it is possible unemployment just doesn’t react fast enough.

4. Something entirely different is happening

I vacillate between camp #2 and #3. The thing is, the past relationship between JOLTS and unemployment during recessions has been based on shocks not just to these data series but to the matching function itself. If there’s no shock to the matching function, then there’s no big, obvious recession — we get the littlest recession and camp #2. On the other hand, given the recent rise in unemployment is consistent with the slow rise in the early phase of a recession (see the shifted 2000s recession data in the graph above), fast changes in JOLTS manifest as slow changes in unemployment which gives us camp #3. In that case, current projections are for a recession of a similar magnitude to the 2000s recession peaking sometime in 2025⁴.

This is not something that will be resolved with next month’s data point. The Dynamic Information Equilibrium Model (DIEM) recession detection algorithm already called a recession six months ago — one that appears to be very slight so far. However, it’s important to remember that there are more than a few speculative components of these models (noise floor, JOLTS-unemployment nexus) — only the dynamic equilibrium at the top of this post (derived from information equilibrium in my paper) showed empirical success relative to various other models and forecasts for the period from when I first wrote it down to the beginning of the pandemic.

All that model is saying is that we are in the midst of a non-equilibrium shock — and have been pretty much continuously since the pandemic.

1. U3 minus temporary layoffs — I’ve been using this metric instead of headline U3 alone since the pandemic as temporary layoffs when haywire. That giant spike in 2020 in U3? It’s almost all temporary layoffs.

2. More accurately, it is a non-equilibrium shock detection algorithm (and can go both ways, i.e. shocks that increase or decrease unemployment — recessions or e.g. the post-pandemic stimulus).

3. It is unlikely unemployment is at such a noise floor as the dynamic equilibrium path was followed to a lower unemployment rate than today’s prior to the pandemic recession.

4. The typical “width” parameter for a recession shock is 0.3-0.4 year which translates to a 95% width (i.e. 95% of the rise of unemployment) happening over 4-5 quarters.