[This is another of a series of editorials by Executive Editor Andrew Karolyi at the Review of Financial Studies featuring recently published papers at the journal. This editorial features “The Effect of Institutional Ownership on Payout Policy: Evidence from Index Thresholds” by Rice University’s Alan Crane and James Weston and DePaul University’s Sébastien Michenaud, an article in Issue 29(6) for June 2016. It was selected as an Editor’s Choice article on on the Oxford University Press web site for RFS.]
“Correlation is not causation” is a mantra drilled military-school style into every budding financial economist. I do not quite remember my graduate school days standing at a blackboard writing the four words repeatedly over and over, but it surely felt like it in retrospect. Now, I am not saying that every study in our midst is subject to potential criticism about endogeneity, reverse-causality, or omitted-variable problems, but sometimes, just sometimes, it’s worth pausing to recognize one that seems to have come upon a setting in which we have confidence in a robust causal finding.
Such is the case, I think, with the study by Crane, Michenaud, and Weston, which tries to break the pernicious, correlative connection between ownership decisions and corporate policy choices. They show that higher institutional ownership causes firms to pay more dividends. They force them to do so. And it is not a small effect. A one-percentage point increase in institutional ownership causes an 8% increase in dividends, worth about $7 million for the typical firm. The impact is even larger for firms that would be perceived to have higher expected agency costs. The authors follow up the headline findings with an analysis of shareholder proposals and voting patterns that suggest even non-activist, passive institutions play an important role in reducing wasteful spending that managers might otherwise allow.
What attracted me and likely others to the paper is the empirical strategy they invoke to break the reverse-correlation problem. They are the first to exploit a pseudo-experimental setting around mechanical definitions of the popular Russell 1000 and 2000 indexes to test their key hypothesis about institutional monitoring driving payout decisions. In a nutshell, the Russell indexes are formed each year (on May 31) on the basis of market capitalization rankings; the largest thousand form the Russell 1000 and the next two thousand, the Russell 2000. At the cutoff between the two, the differences in capitalization are a tiny fraction of the return variation that firms cannot possibly control, so their index assignment on one side or the other of the threshold is as good as random. The weights these stocks enjoy at the top of the Russell 2000 ledger are substantially higher than those at the bottom of the Russell 1000 given the money tracking them, which opens up the possibility for the discontinuity in weights to serve as an instrument for institutional ownership. Voilà! A quasi-natural experiment is had and causal-not-just-correlative findings ensue.
When I contacted the authors to ask them what they think readers may not fully appreciate about their paper’s findings or approach, they mentioned to me that their paper worked hard to unify a number of different econometric approaches to the use of the Russell index thresholds, one of which includes a paper by Ye-Cheng Chang, Harrison Hong, and Inessa Liskovich that the Review published last year (“Regression Discontinuity and the Price Effects of Stock Market Indexing,” January 2015, Volume 28(1)). In a carefully crafted internet appendix available on the Oxford University Press site for Review publications, Crane, Michenaud, and Weston replicate the results of other recent studies and test the sensitivity to methodological differences. They mentioned to me that “subtle biases and small variations in methodology can drive very different results for some outcome variables,” though their dividend result is robust across all approaches.
Diving into this kind of detailed supplementary note may be just the ticket for scholars who want to pursue this particular event involving exogenous variation in institutional ownership. It could be a template for exploring how other pseudo-experimental settings can be exploited to get traction with causal inference. One thing I am 100% certain of is that it is a more useful pursuit than repeating aimlessly a “correlation is not causation” mantra.