Grad School Applications

Stats and Advice

I will be starting a PhD in economics at Harvard Business School this summer!

This post is a bit of reflection on whether and for whom grad school makes sense post Deep Research and some advice for prospective applicants.

Should you do an Econ PhD? Won’t AI Automate All That?

A few years ago, before LLMs existed outside of AI research labs, econ PhDs were obviously good investments for people interested in research and writing. Salary prospects were good even compared to other STEM fields like biology or mathematics, fully funded PhD positions were common, and the academic job market wasn’t so tight so good students had a good chance of landing a tenure track job and a stable, interesting job for life.

Many of these things are still true but AI advances should make reasonable observers wonder about how true they will be in five years. Deep Research is literally a PhD machine, doing a full day (or weeks!) of work from the median grad student in a few minutes. Both the teaching and the research responsibilities of professors seem harried by AI advances.

I have a couple of responses to this reasonable concern:

First, a grounding in the stability of the academy through several previous technological revolutions, including the internet and computing.

You might have reasonably thought that the internet would up-end the university system. Why would people go to their local college when anyone can take classes from the top professors? The Internet also changes the production and dissemination of research, making each professor much more productive. Is there any demand for all that extra research when most papers aren't read anyways?

In some ways, the Internet and computing did completely change economic research. In other ways, it is exactly the same.

Second, all empirical evidence so far points to AIs being highly complementary to human labor, rather than substitutes.

Source

I collaborate with AIs everyday to research more widely than I ever could before and write more complex code in languages I could have never learned. But the AIs are useless without my input.

Deep Research doesn't close the gap between the productivity of the most skilled academics and the least skilled ones, it widens it. If you have taste, determination, and enough knowledge to evaluate AI outputs, then you can run a one-man research lab. Most people lack these skills. I have a chance to be at the frontier of using AI for economics research.

Third, AI career uncertainty applies pretty broadly to all the jobs I could be doing. There’s not some obviously better alternative that I could easily switch to. Anything where the inputs and outputs happen on a computer seems exposed, but that’s basically all I’m good at. Maybe I should go to carpentry school, but that’s a massive sacrifice of my comparative advantage. Certainty of AI’s future effects is not high enough to justify this yet.

I don’t think that business as usual is the right prediction for the future of economic research. Some big things will obviously change. The variance of this path has gone up, but the expected value hasn’t clearly fallen and may be rising.

Application Advice

The main inputs to an econ grad school application are the GRE, your letters, research experience, and any research outputs that people can read online.

GRE:

• I took the GRE twice and studied for several months leading up to each test. The vast majority of the emphasis is placed on the quant section for econ programs. The quant section severely top-coded and thus extremely sensitive to mistakes. In my GRE cohort, the 170 perfect score was only a 92nd percentile so a single question wrong sends you down nearly 10 percentile points.
  
  The main study advice I have is just to do 1000+ practice questions and lots of timed practice tests. The only additional structure I found useful was tracking every question I got wrong across all practice and then slowly mixing in review from that set until for the final week or two before the test you’re only reviewing the hardest questions and the patterns you consistently mess up.
  
  For the verbal section, Anki flash cards work very well.

Letters + pre-docs/research experience:

• I am grouping these together since all of these things depend on making connections with economics professors. There are three easy ways to do this that most people don’t do:
  
  Send cold emails. If you can express a bit of knowledge about and interest in a professor’s work in an email to them, they are likely to respond and meet with you. This is arguably getting less true with AI assisted research and email writing, but honestly the main constraint on this has always been the initiative on the student’s part rather than any constraint on learning about what a professor does, so I don’t think the power of cold outbound will change much.
  
  Talk to your professors. Stay after class and go to their office hours. Ask them a question about their research and pitch an idea of your own. This will distinguish you from everyone else in your class. During my pre-doc at Dartmouth, I ate lunch in the faculty lounge every day and they told me I was the only pre-doc to ever do this. Even some of the visiting PhD students never went to lunch!
  
  Finally, and most importantly, write online! Having a credible signal of research skill and interest is extremely powerful. This essay I wrote about science funding is the entire reason I landed my pre-doc with Heidi Williams. Again people will say that AI is draining the credibility from online writing but I don’t think that’s true. It doesn’t matter how easy AI makes it to write a blog post, most people simply will not do it so you can still show that you have more interest and initiative by doing this.

Actual Applications:

• I applied to 9 programs at 6 schools (three econ and business school pairs) and I was accepted at 3 programs. My application was very strong, but there is still a lot of randomness and variance. I had relatively strict location constraints since I’d be moving with my wife and strong outside options that shrunk the set of schools worth applying to but I still probably under-applied.

P.S if any of my subscribers are in Boston, please let me know! Shoot me an email or a Substack chat and I'd love to talk.