Print
E-MailJonathon Zytnick is an Associate Professor of Law at Georgetown University Law Center. This post is based on his recent working paper.
The recommendations of proxy advisors are the subject of extensive academic research. Yet in recent years, proxy advisor recommendations have become largely unavailable for use in academic research. In a recent paper, I use shareholder votes to impute the recommendations of the two major proxy advisors with a high degree of accuracy. I place the imputed recommendations on my website for use by academics.
Two proxy advisors combine for a 97% market share: Institutional Shareholder Services (ISS) and Glass Lewis, with ISS the far larger of the two. There is no widespread access to the recommendations of either. Until recently, academics studying corporate governance have universally relied on ISS’s Voting Analytics database for ISS recommendations, but sometime between 2020 and 2022, ISS removed its recommendations from that dataset, both retroactively and on an ongoing basis. Academics have traditionally had limited or no access to Glass Lewis recommendations. The lack of access to proxy advisor recommendations creates a need for a practical, accurate substitute.
Comparing my imputed recommendations with actual recommendations for the period of 2011 through 2017 suggests that academics should feel confident using them as variables of interest or control variables. For proposals that received at least one vote by a mutual fund, I impute 96.2% of ISS recommendations with a 99.5% accuracy rate and 92.0% of Glass Lewis recommendations with a 98.8% accuracy rate. Coverage and accuracy are better on proposals with more votes from mutual funds: for proposals with at least 20 votes, I impute 100.0% of ISS proposals with 100.0% accuracy, and 99.0% of Glass Lewis proposals with 99.5% accuracy.
My method statistically reverse engineers the proxy advisor recommendations from the votes of the mutual funds that follow those recommendations. The fundamental logic is simple: if you know something, even imperfectly, about how frequently mutual funds agree with a proxy advisor’s recommendation (their “follow rates”), and you observe mutual funds’ votes, you can use their votes to estimate the probability that a given recommendation is for or against. For example, if many mutual funds that follow ISS most of the time vote in favor of a given proposal, ISS likely recommended in favor of that proposal.
I improve the quality of imputations using an iterative approach. After estimating recommendations in the first iteration, I re-calculate each investor’s follow rates. I do not limit myself to the mutual fund’s overall follow rate. Instead, I calculate situation-specific follow rates that vary depending on the recommendations of corporate management, ISS, and Glass Lewis. Thus, for each fund, I calculate its follow rates with ISS and Glass Lewis when management, ISS, and Glass Lewis support the proposal, when management and ISS support the proposal but Glass Lewis opposes the proposal, when ISS and Glass Lewis support the proposal but management opposes the proposal, and so forth.
Calculating situation-specific follow rates in this manner improves the accuracy on the toughest proposals. Reverse-engineering imputations from votes can lead to more errors and non-imputations on proposals that feature heavy disagreement, and such systematically distributed errors and non-imputations would generate artificially high correlations between the recommendations of management, ISS, and Glass Lewis. Instead, I use situation-specific follow rates and produce correlations between the recommendations of those entities that match the true correlations.
The result is a set of proxy advisor recommendations, from 2005 to 2023, that is nearly comprehensive and highly accurate, and that will be updated each year as new voting data comes out. To make the imputed recommendations more useful to researchers, I include on my website the imputed probabilities and the number of mutual funds voting on the recommendations, so researchers can subset to proposals for which there is a higher degree of confidence. I also include a crosswalk between two datasets, ISS Voting Analytics and Insightia, which has Glass Lewis recommendations for some proposals. My aim is to provide, for public use by academics, a nearly complete set of recommendations by the two major proxy advisors, backed by accuracy tests and thorough documentation of methods.
