3 final things:
1. I will still be affiliated with TBL as a nonresident fellow.
2. More news on next steps in a couple weeks.
3. @riccoja will be taking over TBL's tariff work for the time being! Follow him and read TBL's new tariff update, posted today:
x.com/riccoja/status...
3/3

Major thanks to Danny Yagan, Natasha Sarin, Martha Gimbel, Patrick Oakford, Suzanne Pinto, Kelly Friendly, Sylva Kroeber, Rich Prisinzano, Ken Matheny, John Ricco, Harris Eppsteiner, Josh Kendall, Dylan Saez, Maddie Lee, Sophia Graham, Juan Carlos Gonzalez, Annabelle Xing, & Thomas Triedman.
2/3

Full column here: www.bloomberg.com/op...
9/9

There's still a lot of counter evidence. The NY Fed's Survey of Consumer Expectations shows similar spending growth at the bottom, middle, & top. Meanwhile, the bottom's share of aggregate wages has actually risen over the past 2 years. www.newyorkfed.org/m...
8/9

Partial Truth #3: The K-Shaped Expansion
If the labor market is weakening, this is a valid fear: the bottom typically get pinched more in downturns.
But we need to be extremely cautious about real-time estimates of spending distribution from private data (not a knock on them)
7/9

Moreover, I find that the rise in the unemployment rate over the last 2 years has been most acute for jobs most *& least* exposed to AI. So even if AI is *a* story, it's clearly not the *only* story, & we should be open to the possibility that the story is just broad weakness
6/9

But the evidence is not uniform. E.g. my colleague @marthagimbel.bsky.social & coauthors find that the recent change in the occupational mix is comparable to past technological shocks like the internet & the rise of personal computers.
budgetlab.yale.edu/r...
5/9

Partial Truth #2: AI is weighing on the labor market.
Could be. There's been thoughtful work on this lately, e.g. from @erikbryn.bsky.social & coauthors, who find young workers in AI-exposed fields have seen more job declines than older workers in the same fields. digitaleconomy.stanf...
4/9

If you just looked at the gross effects of software, information equipment, & data centers on GDP, you'd conclude they added 1.3 points to 2025 H1's 1.6% SAAR growth!
Net out imports though, & the contribution falls to ~0.5pp. Still big! But just enough to offset tariffs.
3/9

Partial Truth #1: AI is driving a boom in GDP growth.
There's no question business investment in AI has surged. But on the question of **GDP** effects specifically, it's important to note that a lot of the investment has been imported, which needs to be netted out.
2/9

Full report here: budgetlab.yale.edu/r...
7/7

And of course our approach is symmetric: policies that *reduce* the deficit have *higher* aggregate net benefits under our methodology than under conventional distributional analysis.
6/7

One can think of our "existing fiscal distribution" as basically being the equivalent of assuming proportional spending cuts & tax hikes in the future to pay for a policy. The result is that a policy's aggregate benefit falls & its distributional impact looks quite different.
5/7

There’s no consensus or single way to incorporate deficit-financing into distributional analysis, especially for public goods like defense. We choose an “existing fiscal distribution”—stacking taxes & mandatory spending, & divvying discretionary spending equally per tax unit.
4/7

But of course, this is an illusion. Higher deficits present a host of trade-offs, including interest costs & the fact that higher debt eventually needs to be paid off. This can skew how policymakers weigh the costs & benefits of different financing options.
3/7

The problem in a nutshell: imagine two versions of a $1,000-per-family tax credit. One finances the policy with higher deficits (i.e. no payfors). The other fully pays for the policy with an offsetting tax hike. Version 1 will look like it *clearly* has a greater net benefit.
2/7

Full report here: budgetlab.yale.edu/r...
10/10