Thread by @paulgp

Now that this is finally officially out, I can give a very brief history of how this paper started -- in 2012.

paulgp.github.io/papers/bartik_…

For graduate school, Henry and I were both in Cambridge (he was at MIT, I was at Harvard), while Isaac was at Michigan. All three of us were the same year at Swarthmore, and took most of the same classes together there.

Bartik-style instruments had really started to experience a resurgence at this point, and the initial starting point was just trying to understand the underlying economics of the instrument in a spatial sense.

My interest was in econometrics, and I basically did not understand how the Bartik instrument worked. Here's my first set of slides. Also my slides were hideous.

What is horrifying about looking at these slides is that, we basically had our first main result back in May of 2013.

Our first draft came out in Nov. 2016 . So, why didn't we write it up / have a draft earlier?

Well, post-presentations in graduate school (the 2013 presentations I alluded to), Gary Chamberlain (RIP) basically said "You should replicate the results from other papers." And, quite frankly, I was too lazy and not interested enough to do it.

I think we were also afraid that it wasn't a very "hard" problem; it wasn't obvious that metrics folks would care that much, and had less of a sense of the value for an applied audience.

A key moment for the paper is that when Isaac started at Stanford, he had a colleague encourage him to take Bartik "out of the desk drawer" and he convinced me as well. At this point, we had the bandwidth and resources to follow Gary Chamberlain's advice.

We were fortunate enough to present the paper in a number of high-quality places, which netted a lot of feedback (an important aspect for a paper that has so much folk wisdom associated with it).

Finally, a key stepping stone for us was adding the Rotemberg weights analysis, which was a real "aha" moment for us.

This was during a week when I visited Isaac in-person at Stanford (in the fall of 2017) and when we figured it out, we were beyond giddy.

I distinctly remember us knowing how great of a result this was for the paper, and excitedly coding it up on the plane ride back.

From there, it was a matter of writing, data clean-up, and then we submitted. That was roughly in March of 2018 that we started submitting -- nearly 5 years after my first presentation :|. From there, it was a relatively fast publication process.

This paper has been a real project of love; I'm glad it's found a really amazing home. /fin