Who uses AI agents? How do agents impact output? How might agents change work patterns? New working paper studies usage + impacts of coding agents (1/n)

💡 Why would a firm spend millions training workers it doesn’t even employ? From open-sourcing to certifications, superstar firms often train the workforce of their customers. My JMP explains why. 🧵

For a sense of what I'll be covering, check out the open source course materials. Problem sets quickly get you working with my and @conlon_chris's PyBLP. GitHub:

Next week, I'm teaching another round of my online Mixtape course on BLP-style demand estimation! The sessions are hands-on and last time we had a ton of great questions. Sign up here: https://t.co/Tf1rW4OXdx

@giulia_bran and I are looking for a pre-doc to work with us on topics related to Industrial Organization and Trade @NYUStern Ideally for two years starting in September https://t.co/xVPH1UuyB1 @econ_ra

My job market's done, and I'm heading back to the NYC area! Excited to join @PrincetonEcon as a postdoc next year and @NYUSternEcon as an AP in 2026. In a happy twist of fate, this means @conlon_chris and I will be colleagues exactly 10 years after our first PyBLP commit:

Thanks for the shout out, @marxmatt! We plan to release this data publicly next year. Please reach out if you’re interested in learning more.

7/6 Also: I'm on the job market! If you're interested in the economics of (industrial policy for) open source or want to see a (very simple) example of micro BLP in action, check out my job market paper:

6/6 All this to say: if you're doing methods research, there are some clear private benefits that can come jointly with doing open source work (impact, transparency, networking, etc.) but also others that could be more surprising.

5/6 Having an open source package lets you show students exactly how the method works with real-time examples. I used PyBLP to teach one of @causalinf @kylefbutts's Mixtape sessions ( https://t.co/QWpHCARo53), which would have been far less accessible without it.

4/6 When others rely on your code, you write more tests, which can benefit your (future) self. For every PyBLP version, I run ~2k tests across ~50 functions. These made it easy to broaden the scope of our papers by adding new simulations and empirical examples with confidence.

3/6 For micro BLP, I spent time iterating on PyBLP's interface to balance usability and flexibility. Feedback from users helped improve my design, which mapped directly into our standardized econometric framework, a key contribution of the 2nd paper.

2/6 I expected PyBLP's issue tracker ( https://t.co/0amWXEGcvX) to be mostly bug reports, but most are questions. Our 2nd paper was motivated by these interactions, and many of our explanations were informed by discussions on the tracker and emails from less vocal users.

Happy to share that this paper was just accepted @JEconometrics! It's my and @conlon_chris's 2nd paper tied to our PyBLP software, so here's a thread on surprising (to me) benefits of combining methods research with open source work. 1/6

1/12 I’m excited to share my job market paper, “Delay Your Rivals? Vertical Integration in Securitization and Lending Competition” (with @JosDSalas1). In this 🧵, I’ll explain how vertical integration (VI) impacts competition, loan rates, and origination.

Economic valuations fluctuate in ways empirical research cannot fully explain What information are we missing? Economic theories emphasize the role of hard-to-quantify beliefs and perceptions My job market paper develops algorithms + measurement to quantify perceptions of firms

Excited to share that I'm on the 2024-2025 job market! My job market paper is about failures of contingent thinking -- the act of reasoning about hypothetical events. 1/

18/18 This has been a joy to write and present (not only because I get to use PyBLP screenshots to illustrate US-China collaboration). I'm excited to keep studying the tech sector from an IO perspective, and building econometric tools on the way. Links:

17/ Subsidies seem more effective and generate large innovation spillovers, especially if the US responds in kind. Subsidies can be cheap because OSS investment is rare (weak private benefits), but impactful because OSS use is widespread (strong public benefits).

16/ What if China tightened its restrictions? Benefits don't seem localized enough for a "tax" on foreign collaboration to boost domestic OSS investment. Instead, lost spillovers raise web development costs globally.