Data

Imputing Proxy Advisor Recommendations

For use by academics, please find below imputed ISS and Glass Lewis recommendations for US companies using mutual fund voting data from 2005 to 2023. The imputed recommendations are indexed by the itemonagendaid field from ISS Voting Analytics. I estimate 96.2% of ISS recommendations with 99.5% accuracy and 92.0% of Glass Lewis recommendations with 98.8% accuracy. I intend to update the data each year. My methodology and accuracy tests are included in the paper, Imputing Proxy Advisor Recommendations (See Tables 1-3).

As an additional resource for academics, I have also created a crosswalk between itemonagendaid and the proposal ID field from Insightia (with methodology and accuracy tests in Appendix A of Imputing Proxy Advisor Recommendations).

If you have questions or recommendations on how to make this information more useful, please feel free to email me at jonathon.zytnick@georgetown.edu.

When using either the imputed recommendations or the ISS Voting Analytics - Insightia crosswalk, please cite to:

Zytnick, Jonathon (2024), Imputing Proxy Advisor Recommendations (Working Paper), available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4878758.

Download links:

Imputing Proxy Advisor Recommendations: SSRN link

Imputed proxy advisor recommendations, 2005 to 2023: .dta   .csv

Crosswalk between ISS Voting Analytics and Insightia: .dta  .csv

Quick Use Guide: .pdf

Quick Use Guide:

Recommendations Variable Guide:

itemonagendaid: ISS Voting Analytics itemonagendaid

prob_gl_1: Likelihood that a given Glass Lewis recommendation is in favor

rec_gl: Imputed Glass Lewis recommendation (1=For, 0=Against, Missing=Undetermined)

prob_iss_1: Likelihood that a given ISS recommendation is in favor

rec_iss: Imputed ISS recommendation (1=For, 0=Against, Missing=Undetermined)

prop_N: Number of distinct mutual fund families that cast a vote on the proposal (with all mutual funds in a family aggregated together)

I include recommendations for any proposals with at least one vote. Researchers also may wish to increase that threshold. Higher thresholds will produce more accurate (but fewer) imputations. See Table 1 from Zytnick (2024), reproduced below.

Researchers may also adjust the probability cutoffs. I use 10% and 90% probability cutoffs to generate recommendations, though few observations are within 1% and 99%. More severe cutoffs produce more accurate imputations.

Crosswalk Variable Guide:

proposal_id: Insightia Proposal ID

itemonagendaid: ISS Voting Analytics itemonagendaid

uncertain: Flag for matches with lower degree of certainty