The Effect of High-Tech Clusters on the Productivity of Top Investors: Comment

Moretti (2021a,b) (henceforth M21) studies agglomeration effects for innovation, testing whether the size of technology clusters causes patenting. Agglomeration effects are important for understanding technological progress (Kerr and Robert-Nicoud, 2020), and are affected by constraints on housing supply (Duranton and Puga, 2020). Using US data on patents filed between 1971 and 2007 (Zucker and Darby, 2014), M21 presents multiple lines of evidence supporting a causal effect of cluster size on patenting. The baseline results are from linear regressions of patents on cluster size, controlling for an extensive set of fixed effects (including inventor effects). The main finding is an elasticity of patenting with respect to cluster size of 0.0676. In this comment, I identify two major problems in M21. First, M21 uses an event study to test for selection bias from promising inventors sorting into large clusters. The event is inventors moving across cities, but M21's specification does not use the variation in cluster size generated by moving. I construct an event study following the literature on 'mover' designs, where the change in average cluster size is interacted with event time indicators. I find a null effect, which is consistent with no causal effect of cluster size, but could also be explained by attenuation bias from moves being misclassified in the data. Second, M21 uses an instrumental variables strategy to control for omitted variable bias from sources like local field-specific subsidies. The instrument is based on changes in the number of inventors in other cities. Due to a coding error, the data is not sorted by city, so the instrument is constructed incorrectly by taking first-differences across cities. When I calculate the first-difference within cities, the estimates are nonsignificant.