Economist studying trade policy, finance, and macroeconomic outcomes. Research on tariffs, industrial policy, and global trade.
Atlanta, GA
Joined August 2020
- Tweets1.4K
- Following133
- Followers3.9K
- Likes2.6K
Adrien Matray retweeted
The amount of bad macro data still in circulation is staggering. That's exactly why the work of @KarstenMueIIer and @chenzix on the Global Macro Database is such a valuable public good.
Adrien Matray retweeted
Nobel season! Little experiment - I’ve made a prediction game:
sfuchs-de.github.io/nobel-ex…
Choose a team of economists for monday's announcement by spreading 100 credits across up to 10. You can change them until Sunday, Oct 11, at 8 p.m. Eastern. Just for fun and bragging rights!
(at least one or two year out) so people interested should apply sooner (i.e., in the next few weeks) rather than later. JOE postings:
Macroeconomics: lnkd.in/eGUuagUELabor Economics: lnkd.in/erKsm5PiHousehold Finance: lnkd.in/em-cJtFR
Reminder! The Federal Reserve Bank of Atlanta is hiring economists! We are looking to fill four positions:
• 2 Macroeconomists
• 1 Labor Economist
• 1 Household Finance Economist
We are hiring on a rolling basis, and are particularly looking for seasoned economists
Super excited to announce the @AtlantaFed is starting a new "day visitor program": come spend the day with us the day of the finance seminar, we will get you an office for the day and you can interact with us and the speaker: atlanta-finance-seminar.com/
Adrien Matray retweeted
📘 Très heureux d’annoncer la parution aujourd’hui de notre livre, Les Politiques économiques à l’épreuve des faits !
Fruit de notre collaboration avec @XJaravel, @emm_monnet et Mathieu Peruyero, aux @OdileJacob, avec une préface de @Ph_Aghion.
And this is what makes writing a good econ paper structurally different than solving a software engineering problem. Bottom line: for now the angst of the profession is ahead of the evidence.
It is not possible to just let probabilistic models throw hundreds (millions?) of attempts at a deterministic gate and see what sticks, without getting bored. A paper, or a model, cannot be cut into pieces with a clear success or fail outcome.
(3) More fundamentally, writing a good economic paper is profoundly different than solving a well-defined problem that can be sliced into a lot of small "true-false" questions. This reduces how much you can get out of AI models.
(2) Modern applied work requires administrative data, or data from private providers worried about leakages, and both live in closed environments where AI cannot be run. Unsupervised on public data like the CPS, AI gets you a good senior thesis, not a research paper.
So why the gap? (1) Unless you are willing to produce pure AI-slop, the binding constraint was never production, it was always production *and* verification. We used to verify in real time, while producing. AI speeds up the production and the verification still has to happen.
I also looked directly at whether papers say they used AI to make the paper. I used python to flag every paper naming a generative model, then had AI read the passages around each mention to separate papers that used one from papers that are about AI (ironic I know!).
Maybe fields do not isolate enough pure applied papers. So I classified papers into pure applied (no mention of a model), pure theory (no mention of data), and mix. Here again it is hard to detect any AI effect among the purely applied or the mix.
Split by CEPR and NBER research programs, data intensive fields that would benefit most from agentic AI, labor or economic history, show no acceleration post 2023. Trade jumps, but 31 of its 138 NBER papers in 2025 have tariffs or supply chains in the title, against 10 of 99.
There is indeed a small increase in papers released after 2023, but it is the long trend running since 2010 and it is hard to observe any deviation. The "everybody must write a COVID paper" moment was a much larger increase in papers produced, and we are still under the peak.
Since it does not match my perception of people churning papers, I went through all the CEPR and NBER since 2010 to see what the data say. I excluded ungated platforms like arXiv, where very few economists post, so whatever accumulates there never reaches the median economist.
I keep being told AI is about to bury the journals. Even if we set aside the "AI-written slop," it is possible that AI helps us so much we all become far more productive, which would overflow journals and make evaluation much harder. Long version: tinyurl.com/bdy2f4pw (1/N)
If you have any question about how it is to work at the Fed Atlanta, don't hesitate to DM me and I will be happy to have a chat on Zoom if needed
We are hiring on the seasoned market at the Atlanta Fed:
HH finance: aeaweb.org/joe/listing.php?J…
Labor (even better if you can do both applied with admin data + you know macro-labor) aeaweb.org/joe/listing.php?J…
Macro (and macro-finance, financial intermediation) aeaweb.org/joe/listing.php?J…