Ben Casselman

Signs BlueSky is coming into its own: I have my first impersonator on here. No, @casselmanbe44.bsky.social is not me. This, here, is me. (Thank you to an eagle-eyed observer, whom I won't drag into this, for flagging!)

The folks at @gothamist.com capture it perfectly: "It's Monday in New York City, where we're doing this again."

Trump’s Challenge to Free Market Capitalism

Stakes in private companies. Handshake deals with CEOs. President Trump's economic policies are a long way from the principles that once defined the Republican Party.
I grappled with how we should understand his policies -- and what will come after him.
🎁🔗
www.nytimes.com/2026/02/22/b... #EconSky

I looked at the sub-sector breakdown for another reader, and most of the recent growth has been in hospitals and doctors’s offices. I haven’t looked into the family members piece, but the payroll numbers only include W-2 employees.

Grocery prices were up just 2.1 percent in January from a year earlier, though they're still running hotter than in 2024. But if people don't seem to be celebrating, that may be because grocery prices are up 30 percent since 2020.

Sorry, bad chart labeling on my part. Light blue is core.

One note of caution, though: Core services excluding housing, which should remove most direct tariff effects and the ongoing shutdown-related noise on shelter, has not improved as much lately, and has picked up over the past few months.

Over the past three months, overall prices have risen at a 2.4 percent annual rate. (Core prices at a 2.5% rate.) Given that CPI typically runs slightly hotter than PCE, that's barely above the Fed's long-term 2% inflation target.

Overall consumer prices were up 2.4 percent in January from a year earlier, the slowest year-on-year rate of inflation since last May. Core inflation was the slowest it's been since 2021.

Reminder, though, that methodological quirks related to the shutdown are still artificially dragging down the CPI's measure of year-over-year shelter inflation.

This is the best year-on-year rate of inflation since last May, before the effects of tariffs began to show up in the data. Big drops in gas and used car prices helped offset increases in airfares and other categories.

Consumer Price Index Summary - 2026 M01 Results

U.S. consumer prices rose 0.2 percent in January and were up 2.4 percent from a year earlier. "Core" prices, excluding food and fuel, were up 0.3 percent month-over-month and 2.5 percent year-over-year.
Data: www.bls.gov/news.release... #NumbersDay

Good, explicitly preliminary thread.

Who Is Paying for the 2025 U.S. Tariffs? - Liberty Street Economics Over the course of 2025, the average tariff rate on U.S. imports increased from 2.6 to 13 percent. In this blog post, we ask how much of the tariffs were paid by the U.S., using import data through No...

Who is paying the tariff tab? New York Fed economists: Using import data through November 2025, we find that nearly 90% of the tariffs’ economic burden fell on U.S. firms and consumers. libertystreeteconomics.newyorkfed.org/2026/02/who-...

We knew when it happened in Argentina because the career people they fired blew the whistle. Because other people collecting similar data saw the official data diverge. Because economic data stopped making sense when you looked at them in combination. If it happens here there will be signs

Interesting!

Interesting.
FWIW, ADP says they don't adjust the topline series but rather the individual components, which they then sum them get the total. Leaving aside the merits of that approach, you'd expect it to give a somewhat different result. But you still wouldn't expect a level shift like this.

Perhaps, although "ADP messes up seasonal adjustment on benchmark revisions" is pretty in the weeds for our audience.

If you have an explanation for this chart other than "they messed up the seasonal adjustment," I would love to hear it. (Truly! I would love this.) #EconSky

2. Private data sources are prone to revision, errors, and biases just as government sources are. And often without the level of methodological transparency we expect from the statistical agencies.

But two points I'd emphasize at the end here:
1. "Private" data sources often rely on government data -- if you want good private data, you need good government data to back it up;

To be clear, I don't intend this thread as a criticism of ADP or its econ team. I appreciate them sharing their data publicly, and it's valuable to have private sources at a time when government data faces its own substantial challenges.

My best guess is that this is a mistake, a bug in the code somewhere. If it isn't, then I'm not quite sure how we should interpret their adjusted and unadjusted numbers. I hope ADP will provide a more complete explanation at some point.

I asked ADP about this, and they pointed me to their blog post (linked above), which discusses their overall approach to seasonal adjustment and cites an academic paper. But neither explains why we would get this kind of strange level shift.

Here is what the chart looks like now, post-revision. Notice how the SA line no longer cuts through the middle of the NSA line, smoothing it out. Instead, it now sits at the bottom of the NSA line. In other words, their seasonal adjustment is (basically) only down, never up.

This is what ADP's headline numbers looked like before the recent benchmark revision. To anyone familiar with these sorts of charts, it will look totally normal -- the NSA line rises and falls along a seasonal pattern, and the SA line cuts more or less through the middle of it.

But there's another issue with the ADP revision: There appears to be something wrong with their seasonal adjustment formula.

But the result is that, in this case, the pre-revision data is probably in some sense more accurate than the new, revised data, at least when we're talking about what happened in earlier years.

Five things to know about the January ADP employment report - ADP Research Last week we published the first ADP National Employment Report for 2026. This January release is special because it captures our annual benchmarking, a process that transforms ADP’s expansive payroll...

That's a perfectly defensible decision, the kind of choice that data providers (public and private) face all the time. Whether it was the *right* decision depends on how you're using the data.
(ADP discussed this more in a blog post yesterday:
www.adpresearch.com/five-things-...)

As a result, BLS couldn't use the most granular breakdowns (industry x geography x establishment size) for its benchmarks. It had to use coarser data. And because ADP wanted its data to be fully comparable over time, it re-benchmarked ALL its data in the same way.