Thread by @jasonfurman

My charts on the distribution of real wage growth were flat out wrong, as @IrvingSwisher correctly pointed out. I'm deleting all of the tweets (here's a screenshot of one so you know what I'm talking about). Feel terrible to have put this out.

This 🧵 explains more fully.

The question I was trying to ask was what has happened to wage growth adjusted for inflation by different income groups & how does it compare to pre-COVID.

The broad inference is likely still true but possibly less bad than I showed (it is unclear). The data was wrong.

I still believe the following are true for the period from the pandemic to today:

1. On average nominal wages have increased more slowly than inflation.

2. Faster wage gains for lower-wage workers

3. For most groups any inflation-adjusted gains are smaller than pre-pandemic.

It is also likely that the bottom quartile has (on average) seen real wage gains and that the top quartile has (on average) seen real wage losses.

I do not have a confident view about what has happened to the middle quartiles. And I'm not sure the data can answer the question.

Let me reiterate: I've shown real wage losses for the two middle quartiles. A lot of people noticed that and cared because, if true, it's a big deal. I do not know if that was true or not anymore.

OK, now to explain the error. The Atlanta Fed Wage Tracker provides nominal wage data w/ various breakdowns. I've mostly used their wage growth by quartile. They match the same workers who were employed 12 months apart to address compositional changes.

atlantafed.org/chcs/wage-grow…

When they do the breakdowns of the data the sample size gets very small so what they call "12 month growth" is actually a moving average of the last 12 12 month growth rates. This is effectively a moving average of a moving average of monthly growth rates.

In my effort to show real wage changes I paired it with CPI growth measured between two points in time. So my CPI growth was completely mismatched to the wage growth. That is flat out wrong and a dumb thing to do.

You might think the way to correct this would be to use the same moving average of a moving average for CPI. That would, indeed, give you a coherent answer. This is what that chart looks like. Unfortunately it is not an answer to the question anyone cares about.

The question I was asking is what happened to wages and prices from January 2020 to January 2022, ie during the pandemic and the economic recovery from it.

The chart is internally consistent but doesn't answer that question, in fact it doesn't answer any relevant question.

Specifically, the question one wants is the sum of the last 24 monthly growth rates.

The internally consistent chart de facto draws 27% of its growth from the pre-pandemic period (should be 0%) and only 27% from the last year (should be 50%).

(I might post a technical thread on that later because it's fun, is analogous to the issues with annual average growth vs. Q4/Q4 growth I've written about before.)

https://twitter.com/jasonfurman/status/956891837293309953?s=20&t=ohCmS8nytm83uXRQv4q8zQ

The Atlanta Fed does not publish the data to look just at January 2020 to January 2022 but you can run the programs they provide to get it. The problem is the data is really volatile (why they smooth it) so the answers change a decent amount from month to month.

Every chart in the previous tweet is correct (I hope!). They all tell a broadly similar story (the one I listed above). But they differ a lot from each other, especially in the middle two quartiles). That could reflect true changes but almost certainly reflects data noise.

Thanks to Twitter I've also learned about other ways that the Atlanta Fed Wage Tracker, while excellent, may have limits in answering these questions, especially at such a strange time. These include mean reversion, age composition, etc.

All data has its limits, one does the best one can. One way is also to compare to other data. In particular, I (and many others) find the officially published ECI data to be among the most accurate in dealing with composition.

Here is the growth of real wages by occupations for the pandemic period compared to pre-pandemic ranked by wages.

You can see real wages up at the bottom, down at the top, and generally doing worse than pre-pandemic.

Increases my confidence in the broad story.

I genuinely appreciate @IrvingSwisher letting me know and giving me time to figure this out--we sometimes agree and sometimes disagree on the economics but I'll never make the mistake of arguing data with him again.

Apologies for the error and (yet another) long thread.

A technical footnote to the thread from yesterday.

https://twitter.com/jasonfurman/status/1492486185515687940?s=20&t=zgwxKilES4v6O0z67utz3g