# A noise floor in initial claims?

### The fluctuations in initial claims may not contain any information

You’ve all heard the term “buried in the noise” — it’s possible that is where the initial claims (ICSA) equilibrium currently is. I noted this possibility several years ago in the context of the unemployment rate (here or here in slide 16), but the recent initial claims data (or rather the claims rate1) along with the data prior to the pandemic shows a remarkable flatness coupled with a high (log) level of month-to-month noise:

The noise term could be the dominant term right now — a combination of measurement error and the fundamental stochastic noise of the series.

Given this is the lowest level of the series ever recorded, the only way to confirm this hypothesis would be if the claims rate fell to this level again after rising and remained flat. There haven’t been any earlier periods where we’ve bumped up against the noise floor.

The downside of this is that we effectively lose this series as a macro indicator until it rises closer to 0.5%. The upside is that we can potentially glean the actual level of measurement + fundamental stochastic noise in the series and use it in our construction of our error bands once we have a longer series (and evidence that this is actually a noise floor).

A Dynamic Information Equilibrium Model is technically written as an information transfer process (information equilibrium) between two process variables: *A* ⇄ *B*. When *A *and *B* are taken to be exponentially growing variables over time (the “dynamic” bit) you end up with the equilibrium being defined in terms of a ratio *A*/*B*

*d*/*dt* log(*A*/*B*) ~ α *t* + *c* + σ

where α is the dynamic equilibrium slope, *c* is a constant, and σ are non-equilibrium shocks. In the initial claims model, we are looking at the ratio ICSA/CLF16OV — initial claims per member of the labor force (employed + unemployed) i.e. the claims rate.