Abstract

A recent literature uses the annual reconstitution of the Russell 1000 and 2000 Indexes as a source of seemingly exogenous variation in institutional ownership to study the effect of institutional ownership on firm outcomes. We show that lagged institutional ownership measured prior to reconstitution exhibits very similar pre-existing differences at the 1000/2000 cutoff, and thus the results from the most common implementation of this setting (e.g., as in Bird and Karolyi, 2019) reflect selection bias instead of a treatment effect. Additional tests confirm that it is the use of rankings based on Russell’s June index weights that leads to biased results. With an unbiased approach, there is no significant discontinuity in institutional ownership at the 1000/2000 cutoff despite the large difference in index weights.