Empirical Evidence on the Role of Proxy Advisors in European Capital Markets

Abstract

Responding to regulators’ requests and filling a void in the academic literature, this paper provides comprehensive empirical evidence on the role of proxy advisors in 14 European countries. Exploiting coverage data and using content analysis of proxy voting reports by Institutional Shareholder Services and Glass Lewis, we provide descriptive analyses of proxy advisors’ firm coverage, the variation and determinants of voting recommendations, the relation between voting recommendations and shareholder voting at annual general meetings, and market reactions to the release of voting recommendations. Overall, our findings suggest an economically important role of proxy advisors in European markets. Throughout our analyses, we document that this role varies with governance- and ownership-related firm characteristics and with country-level measures of institutional strength.

Acknowledgements

We are grateful for valuable comments from David Veenman (the associate editor) and an anonymous reviewer, from Yuping Jia (discussant), Oksana Pryshchepa (discussant), Christophe Volonté, and participants at the 11th Workshop on Corporate Governance in St. Gallen, Switzerland (October 2014), the Annual Meeting of the European Accounting Association in Glasgow (April 2015), and the Annual Meeting of the European Financial Management Association in Breukelen/Amsterdam (June 2015). We thank Franziska Heusel, Marisa Rogge, Thorben Tiedemann, and Simon Wetterau for their excellent research assistance. We also thank Kristof Ho Tiu (Institutional Shareholder Services) and Carla Topino (Glass Lewis) for kindly providing data on ISS coverage and GL voting reports/GL coverage, respectively. All remaining errors are ours.

Notes

1 Ertimur et al. (2015) show that extensive consultations and round tables with PAs’ clients – primarily institutional investors – commonly precede the issue of voting guidelines, thus presumably forming and guiding PAs’ voting recommendations at AGMs.

2 In contrast to other PAs, ISS also markets commercial governance ratings. Introduced in 2002, these ratings cover more than 8000 firms across 31 countries (RiskMetrics Group, 2007).

3 No Europe-wide regulatory measures address PAs. However, on the member state level, some policy recommendations rather indirectly address PAs, e.g. the UK FRC Stewardship Code Principles 1 & 6 of 2012 and the French AMF Recommendation No. 2011-06 (ESMA, 2012). US PAs are commonly regulated under the Investment Adviser Act of 1940 (ESMA, 2012). Under this regulation, PAs have to comply with certain fiduciary obligations and have to meet minimum disclosure standards. However, depending on the services provided, not all advisors are required to register as investment advisors under the Adviser Act. In 2010, the SEC released a concept paper on the US proxy voting system to review the role and influence of PAs in the US and to discuss potential policy options. In June 2014, the SEC released a Staff Legal Bulletin No. 20. The main objective of the bulletin was to provide guidance on investors’ use of PAs (Malenko & Shen, 2016).

4 Untabulated analyses indicate that the random inclusion or exclusion of one sample country (e.g. Greece) does not alter our main findings.

5 Iliev and Lowry (2015) show that US mutual funds that conduct independent research (measured by fund size, fund turnover, and investment size) are less likely to rely on ISS proxy voting.

6 This sample selection restriction is less of a concern because exclusive GL coverage is limited in our underlying sample (Panel B, Table 1).

7 We define a lead analyst as the analyst reported first in the report.

8 The inferences in this section also hold when re-estimating all analyses on a matched sample that only includes firms that are simultaneously covered by both PAs. However, following prior related US research (e.g. Choi et al., 2010; Daines et al., 2010), we consider the PAs’ individual firm universes within our baseline sample.

9 In contrast to our firm-level analyses in Section 3, we exclude financial analyst following from the regression model due to multicollinearity concerns. Also, the findings in Table 4 are not sensitive to any specific combination of our ownership and investor protection variables.

10 Based on up to 80 governance items, ISS provides governance ratings (GRID) for over 8000 firms worldwide. While we are aware of concerns with third-party governance ratings (e.g. Daines et al., 2010; Hitz & Lehmann, 2015), we follow prior cross-country governance research and rely – in the absence of a parsimonious alternative – on ISS governance ratings (Aggarwal, Erel, Ferreira, & Matos, 2011; Bruno & Claessens, 2010; Chung, Elder, & Kim, 2010; Doidge, Karolyi, & Stulz, 2007). However, a common governance perception by ISS across its proxy voting and rating products might compromise the construct validity (e.g. independence) of this variable in our setting. We address this concern by including into our analyses additional governance-related variables (e.g. insider ownership or investor protection) and by reporting comparable results for GL, where the GRID rating represents an external governance measure.

11 These findings potentially point at different market strategies of GL and ISS across markets. In other words, GL might position itself as a more aggressive PA – in terms of rejection rates – to gain market share and expand its business, especially in countries with a more management-friendly ISS voting policy.

12 We estimate a decrease in the likelihood of receiving ‘vote against’ recommendations as the relative change in the predictive probability of ‘vote against’ recommendations between low and high governance quality firms (e.g. 34.4% = (15.4 – 10.1)/15.4).

13 To ease the interpretation of the regression results, we follow Ertimur, Ferri, and Muslu (2011) and use voting results in percentages as the dependent variable. However, as this dependent variable is a percentage with a fixed range between 0 and 100, we re-estimate the regressions with a log-transformed dependent variable. In line with Bethel and Gillan (2002, p. 48), we employ the following transformation: log[% voting result/(100 – % voting results)]. The untabulated results based on the log-transformation are in line with our original findings.

14 Following Ertimur et al. (2011, p. 564), an alternative approach uses the residuals of ISS recommendations (obtained by regressing ‘ISS_AGAINST’ on firm characteristics, which are likely to explain the ISS voting recommendation, i.e. performance, size, ownership, and analyst following). Unless (firm) characteristics, which vary at the AGM voting item level are included, our baseline model with firm-fixed effects will lead to identical estimates. Nevertheless, the untabulated tests employing this abnormal measure confirm our results.

15 Even when we re-estimate our multivariate analyses for board- or compensation-related subsamples, we yield – consistent with our descriptive findings – an economic significance of approximately 11% for ISS ‘vote against’ recommendations and of 7% for GL ‘vote against’ recommendations.

16 Alternatively, ISS might be superior in anticipating shareholders’ voting behaviors.

17 Broadly, this cut-off rule implies that ISS performs deeper analyses of compensation arrangements for only a portion of the covered firms (firms that fall below a certain threshold with respect to total shareholder return). Malenko and Shen (2016) show that firms just below the cut-off have a 15% higher likelihood of receiving ISS ‘vote against’ recommendations compared to firms that are just above the cut-off. In addition, the authors show that firms around this cut-off are locally random, except for their different likelihoods of receiving negative ISS recommendations. The authors argue that any changes in voting dissent around the cut-off can be attributed to ISS voting recommendations rather than to the same underlying information set between ISS and shareholders.

18 Note that ISS reports include the actual publication date (i.e. the release date), whereas GL reports only include information on the issuing date – not on the release date (i.e. GL reports include the latest closing price of the stock and the respective date of the closing price).

19 Untabulated analyses show that the market reaction does not vary with specific management proposal types (e.g. compensation-related proposals such as say-on-pay votes). One might assume higher information content for recommendations on non-routine and more subject voting items. These items are less likely covered by the PA’s general voting agenda which is commonly known to investors prior to the proxy season. Instead, our findings suggest that firm- and country-level characteristics play a more pronounced role with respect to differences in the information content of PAs’ voting recommendations.