One of my favorite papers at the NBER SI so far: The Macroeconomic Effect of AI: Sizing the Software Engineering Channel by Blumenfeld - @JADHazell - @ChenLian92 - Schaab. (link below) ❓Big question: what are the effect of AI on software engineering productivity and ultimately on GDP? 💡 Key idea: let's infer this from stock prices. Concretely, firms that rely more heavily on software engineers should benefit disproportionately from AI advances. Stock prices should reflect this: firms with higher software engineering intensity should exhibit higher stock return sensitivity to an AI index. The magnitude of this cross-sectional relationship, i.e. how steeply stock return sensitivity to the AI index rises with software engineering intensity, is informative about the size of the underlying AI-induced software engineering productivity growth. The paper carefully develops and implements this idea, using stock prices + Revelio payroll data, along with a model that maps the cross-sectional relationship into SWE productivity and GDP. ➕ 2-and-a-half major advantages of this methodology: (1) Stock prices capitalize the entire expected future path of AI-driven productivity gains - not only the improvements delivered by today's tools. (2) The estimates can be updated in real time, incorporating new AI-related information. (3) The approach side-steps the difficult question of how to aggregate from task-level gains to overall software engineering productivity gains. (Of course, this challenge still looms big in the background--it's just ~left to market participants to solve.) Of course, the approach works only as well as financial markets can price AI-related information - it inherits mis-pricing due to over-reaction, general exuberance, etc. As Joe put in his fantastic talk: "possibly wrong, hopefully useful". Very exciting and, imho, useful work.