Shareholder Activism on Climate Change: Evolution, Determinants, and Consequences

Securities and Exchange Commission (SEC) Rule 14a-8 (SEC, 2020) requires firms to include in their proxy statement (DEF14a) shareholder proposals for voting at the annual general meeting (AGM) unless the shareholder has not complied with the procedural requirements or the proposal meets the SEC’s criteria for exclusion. To be eligible to include a proposal in a firm’s proxy statement, shareholders must meet certain requirements. They must have continuously held one percent of the firm’s outstanding stock or $2,000 in market value for at least one calendar year. In addition, there are some proposal-specific requirements. For example, proposals must be submitted at least 120 days before the proxy materials are mailed to the shareholders as part of a DEF14a filing and cannot exceed 500 words in length. Each shareholder can submit only one proposal per meeting (Byrd & Cooperman, 2014).

As depicted in Fig. 1, after a shareholder submits a proposal, three outcomes are possible, which we denote as omitted, voted at the AGM, and withdrawn (Bauer et al., 2015; Ertimur et al., 2010). First, if the firm disagrees with the shareholder proposal it can send a letter to the SEC including the reasons to exclude the proposal from its proxy materials. SEC Rule 14a-8 (SEC, 2020) specifies the conditions for the omission of a shareholder proposal from a DEF14a filing. Firms can rely on several reasons to appeal to the SEC for the omission of a proposal (Schopohl, 2017). The four most common are (i) the inclusion of a false or misleading statement, (ii) the proposal’s irrelevance (i.e., the proposal is based on issues that account for less than five percent of the total assets related to the firm’s ordinary business operations), (iii) duplication (i.e., another shareholder has filed a similar proposal), and (iv) the targeted firm has already addressed the issue (SEC, 2017). Accordingly, once the SEC receives a no-action request, it evaluates the shareholder proposal and the reasons for exclusion and informs the firm and the filing shareholder of its view on the request by issuing a “no-action” letter. If the proposal does not meet the regulatory requirements to be included in the firm’s DEF14a filing, it will be omitted and, therefore, not disclosed in the firm’s proxy materials.

Second, if the SEC does not agree to exclude the proposal or if the firm agrees to include the proposal in its proxy materials, it will proceed to a vote at the AGM. That is, the shareholder proposal is disclosed in the DEF14a as eligible for voting. We note that an engaged firm is legally entitled not to implement a shareholder proposal, even if it receives more than 50 percent of the votes at the AGM (Ertimur et al., 2010; Flammer, 2015; Michelon & Rodrigue, 2015).

Third, when negotiations ensue after the proposal is included in the proxy statement, the firm must file an amended DEFA14a if the proposal has been withdrawn by the shareholder (again, see Fig. 1). However, if negotiations are successful before the proxy filing, the firm is not obliged to communicate the withdrawal. Because this process is private, it is difficult to know the date of withdrawn proposals. Contrariwise, if the negotiation is unsuccessful and the proposal meets all regulatory requirements to be included in the proxy statement, it will proceed as outlined in Fig. 1 to a vote at the AGM (Bauer et al., 2015; Clark & Crawford, 2011).

While this process yields two dates, namely, the proxy statement filing date and the AGM date, the first formal news to the market arrives in the DEF14a when investors and the public learn that the firm has agreed that a proposal is eligible for a vote (Gillan & Starks, 2007). Knowing the details, investors can then form expectations of the relevant attributes of a proposal, such as whether it will pass, be withdrawn and negotiated, and if passed how it might initiate climate action, such as a reduction in emissions or an increase in climate-related disclosure, and whether such actions are ex-ante costly or beneficial for firm value. Nonetheless, investors may still receive some new information at the later AGM date, such as actual knowledge that a proposal is withdrawn, but this relates mostly to outcomes that have already been anticipated based on earlier expectations, which would be largely unbiased assuming market efficiency, which is a maintained hypothesis for an event study despite puzzling anomalies and evidence of exploitable mispricing (Ball, 1994).

Panel A of Appendix Table 9 illustrates the submission timeline of a 2017 proposal to Exxon Mobil Corporation indicating a shareholder proposal date (before December 15, 2016), a no-action letter date (in January 2017), the proxy filing date (as of April 13, 2017), and the AGM date (as of May 31, 2007). Exxon Mobil filed its 10-K report as of February 22, 2017.

The Ceres sample comprises 944 shareholder proposals submitted to 343 U.S. firms on climate change issues during 2009–2022. The Ceres dataset reports the three possible outcomes, namely, Omitted, Voted at the AGM, and Withdrawn. Of our total sample, 400 proposals receive a vote (Voted at the AGM), 88 proposals are omitted (Omitted), and 456 proposals are withdrawn (Withdrawn).⁸

Because shareholder proposals eligible for a vote must be included in firms' proxy statements, we use Direct EDGAR to collect the dates for our event study of market response. In addition to the shareholder proposal data, we gather firm-level data as follows: (i) financial statements from the Compustat/CRSP merged Fundamentals Annual file, (ii) corporate governance data from Refinitiv, (iii) media attention from RavenPack, (iv) environmental performance ratings from Bloomberg, and (v) daily stock returns from CRSP. We merge these data with the Ceres dataset using PERMNO, CIK, ISIN, and firm name. We checked the merge manually. For some tests, we extract additional data such as whether the firm-year covers a Republican or Democratic presidential period or whether the shareholder represents a UNPRI-aligned institutional investor.

For control firms, we select a matched sample alike in financial characteristics other than having been targeted by a climate proposal. We start with the Compustat universe other than treated firms and propensity score match these firms to control firms based on size (log of total revenues), year, and industry group (two-digit GICS (Global Industry Classification Standard)). We then apply the nearest-neighbor-matching method with replacement (Semenova, 2023; Shipman et al., 2017). As such, the same control firm serves as the closest match firm for one or more targeted/treated firms. Thus, each treated firm has the characteristics of a matched control firm equal to the mean characteristics of its year-industry-size matched portfolio.

To recognize a key distinction between the climate proposals, namely, whether they relate to decision-making by the firm (operations-related) or additional public disclosure (disclosure-related), which could affect the voting, market response, and effect of a proposal on future outcomes, we read each proposal to identify the shareholder’s request. We identify disclosure-related proposals as those that focus on the disclosure of environmental information, for example, they could relate to a response to the CDP questionnaire, the disclosure of emissions (Scope 1 and 2 emissions, primarily), or a request to disclose science-based targets. By contrast, operations-related proposals call for more substantive actions, for example, proposals that press the firm to address the short- and long-term climate risks of a lower carbon economy, those that engage the firm to explain how it will operate in a two-degree scenario, and those that call for a reduction in emissions. Operations-related proposals may also involve more investment by the firm that would otherwise be ill-conceived for firm value maximization since the actions imposed on the firm by outside shareholders would likely be based on incomplete information compared to information internal to the firm, even though such actions could be preferred by ethical shareholders willing to sacrifice higher returns to further a climate or sustainability goal or internalize a negative externality (Baker et al., 2022; Barber et al., 2021; Hong & Shore, 2023).

We use a three-phase process to code our sample of climate-related proposals. To categorize each proposal, we rely on the original content of the proposal (see Appendix Table 9 for examples). Initially, to prevent ambiguous interpretations, two of the authors jointly conducted a pilot test considering 236 proposals (25 percent of the total sample of 944 proposals). They met recurrently to discuss all doubts during and after the pilot test. Next, after revising and establishing a robust coding protocol, the same authors manually coded the remaining proposals. A third author not involved in the initial coding then performed a second coding and ran a Cronbach reliability test of whether a proposal was operations- or disclosure-related, finding a high level of rater reliability based on an alpha level of more than 90 percent.

To understand our first research question of what determines whether activists target a firm with a climate-related proposal, we state the following logistic regression model:

Targeted = 1 if the firm is targeted by a climate-related proposal and zero otherwise (i.e., is a non-targeted control firm as per the previous section). We align the firm-level data to the prior fiscal year closest to the DEF14a filing year. Appendix Table 10 defines the other variables. Note that Eq. (1) is unconditional on a proposal’s outcome, as not all targets receive a vote, and some proposals are withdrawn or omitted from the DEF14a. As such, Eq. (1) addresses the question of what determines whether a firm is targeted and, possibly, whether those determinants vary on the mix of proposals. We, therefore, estimate Eq. (1) for the sample of all targets and the subsamples of operations- and disclosure-related proposals. Prior evidence suggests that firms with ESG shareholder proposals are larger and have poorer levels of environmental, social, and operational performance (Dimson et al., 2015; Ertimur et al., 2010; Michelon & Rodrigue, 2015; Wei, 2020). Given that climate-related proposals are a subset of ESG, we also expect such proposals to engage firms that are larger (Size) and with lower growth opportunities (BTM, Sales_Growth). In addition, we expect targets to spend less on R&D and more on plant and equipment (Capex) compared to non-targets. Equation (1) also includes media attention and governance variables as regressors that might further explain shareholders’ decision to target. We also include year fixed effects and control for industry with a carbon-intensity variable (based on whether a firm is in a carbon-intensive industry). We cluster by firm-year to estimate the standard errors for coefficient significance. Clustering in this way helps control for within-firm correlation among the different proposals since shareholders can target some firms more than once (944 shareholder proposals submitted to 343 firms) and, also, because the control sample could include firms that have been targeted on multiple issues unrelated to climate change.

We use logistic regression to examine the determinants of why some eligible climate-related proposals proceed to a vote and others do not, which is our second research question. As per Eq. (1), we align the firm-level data to the prior fiscal year. Because we examine a subsample of targets, sample selection bias may be introduced into the model (Breen, 1996). To correct for this bias, we run two-stage regressions that calculate and include the Heckman correction factor (HCF) based on the inverse Mill's ratio (Heckman, 1979). This model is

The dependent variable for Eq. (2) equals one if a climate change proposal proceeds to a vote at the AGM and zero if it does not, that is, it is omitted or withdrawn. For firms with proposals eligible for a repeated vote or other subsequent actions, Voted refers to the first eligible proposal. Thus, Eq. (2) counts a proposal only once regardless of the number of subsequent actions associated with that proposal. In addition to the regressors in Eq. (1), Eq. (2) includes proposal-level variables to assess the potential that proposal eligibility changes over time (e.g., before and after Republican presidents), whether voting eligibility is higher for proposals with repeated votes or those calling for emissions reductions, and whether firms’ environmental performance might also explain whether a climate change proposal proceeds to a vote at the AGM. Equation (2) does not include year fixed effects to avoid overlap with Republican, which is also a time-related variable, or industry fixed effects to prevent overlap with Carbon_Intensity, which is also an industry-related variable. Appendix Table 10 defines the other variables.

The model for our third research question of what determines the voting percentage is

The dependent variable (Voting%) is the first-vote percentage if the vote is later repeated. As a sensitivity test, we also use the average vote percentage when the proposal attracts multiple votes. Given that the dependent variable is a proportion (ranging from 0 to 1), we estimate a fractional probit model (Papke & Wooldridge, 1996). We keep the same set of variables and time subscripts as in Eq. (2). In addition, because Eq. (3) focuses on a subsample of proposals that proceed to a vote at the AGM, we estimate Eq. (3) as a two-stage fractional regression model and include the Heckman correction factor (HCF) to control for selection bias. We expect a higher favorable voting percentage for firms with higher capital spending (Capex) and those with less R&D intensity. We also expect higher voting percentages for repeated-vote proposals, potentially reflecting an elevated persistence by shareholders for substantive change. However, unlike Eqs. (1) and (2), it is unclear that larger firms should receive higher voting percentages despite their higher likelihood of being targeted and shareholder success in introducing more eligible proposals [Eq. (2)]. Although larger firms may receive more media attention (Appendix Table 11), such higher attention need not necessarily translate into a higher percentage.

To address our fourth research question, we use an event study. We align each shareholder proposal to an expected price-sensitive date, namely, the DEF14a filing date. We access these dates from direct EDGAR. We estimate firms’ cumulative excess returns around these dates using the market-adjusted model (the market model and a three-factor excess returns model give similar results). The model is

CAR_it refers to a firm i’s cumulative market-adjusted excess return from day t₁ to day t₂. R_it is the common stock return for day t, and RM_t is the expected common stock return on day t calculated as the return on the CSRP market index for day t. While it is possible to estimate CAR around later dates, the proxy filing date is the earliest and forms the basis for our tests (Gillan & Starks, 2007). Different window lengths are analyzed, such as the event windows of days (− 10, 10), (− 5, 5), and (− 1, 1). We assume market efficiency, which is a required maintained hypothesis for an event study (Binder, 1998). We focus on the proxy filing date as the date when stock investors first learn formally that the AGM agenda includes a proposal for voting (and at that point it is not withdrawn or omitted) and form expectations of how the agenda items affect firm value (“The proxy statement filing process” section). We acknowledge, though, that some leakage could occur through the statements of proxy advisory firms (e.g., Institutional Shareholder Services) and pension funds with a history of shareholder activism (e.g., CalPERS). Some eligible proposals in a DEF14a filing are also repeated proposals. We view any earlier or later event dates as weaker tests, however, as less or no significant price reaction about those shareholder proposals eligible for voting should occur around those times.⁹ Also, because the different dates have little overlap, this minimizes the chances that a short window price reaction around an earlier (the annual report date) or later date (e.g., the AGM date) would contaminate the price reaction on the DEF14a date.

We conduct three kinds of tests of the CAR measures from Eq. (4). We first check if the cumulative average mean excess return for the sample with climate-related proposals in their DEF14a (treatment firms) differs from zero or differs from the cumulative average mean excess return for the control sample (control firms) around the different CAR (t₁, t₂) intervals from t₁ ≥ − 10 to t₂ ≤ 10. Our null hypothesis for this test is that CAR (t₁, t₂) = 0 or that CAR (t₁, t₂)_Treatment minus CAR_it (t₁, t₂)_Control = 0. However, because firms’ DEF14a filing contains much other price-sensitive information (e.g., executive compensation, audit fees), this test has low identification.

Accordingly, to increase identification, we test for differences in the mean CAR (t₁, t₂), where the differences vary predictably on the attributes of the shareholder proposals. We test the following five cross-sectional predictions: (i) CAR (t₁, t₂) | repeated-vote proposals < CAR (t₁, t₂) | single-vote proposals, (ii) CAR (t₁, t₂) | withdrawn proposals < CAR (t₁, t₂) | voted proposals, (iii) CAR (t₁, t₂) | GHG-reduction proposals < CAR (t₁, t₂) | other proposals, (iv) CAR (t₁, t₂) | marginally passing proposals < CAR (t₁, t₂) | other proposals, and (v) CAR (t₁, t₂) | UNPRI proposals > CAR (t₁, t₂) | non-UNPRI proposals. As an initial test of hypothesis (H1), we conduct univariate binomial tests of the null hypothesis that the differences in the mean excess returns for days − t to t for a partition are not positive or negative when the expected likelihood of being positive or negative is 0.5. The first four cross-sectional predictions build on the theory that investors react negatively to proposals that impose costs on the firm as a result of expected deviations induced by the proposal from managers’ efficient operating and investment policies to maximize firm value. Assuming targeted firms operate to maximize firm value, the increased cost and higher uncertainty of a shareholder proposal that may succeed should decrease firm value in the short term. Specifically, repeated proposals are likelier to succeed. Withdrawn proposals are likelier to be implemented through negotiation. A track record of non-responsiveness as reflected in repeated-vote proposals may also require more constructive engagement in the future. GHG-reduction proposals are likelier to require additional spending, and marginally passing proposals are likelier to succeed the next time. Additionally, for operations- minus disclosure-related proposals, we also predict negative differential excess returns bolstered by evidence that additional voluntary climate-related disclosure increases firm value (Flammer et al., 2021; Griffin & Sun, 2013). However, if investors interpret such additional disclosure as a form of virtue signaling or the disclosures reveal proprietary information, the differential return (operations-related minus disclosure-related) could be positive or insignificantly different from zero despite an increase in environmental performance and more disclosure. By contrast, UNPRI proposals should generate positive excess returns by improving firms’ environmental performance metrics given the engagement of large sustainability-minded institutions in support of the resolutions. We also explore whether higher levels of media attention generate positive or negative excess returns over the event period. On the one hand, a higher level of media attention could increase firm value by giving a larger voice to the arguments of shareholders in favor of change (Barber & Odean, 2007; Stanny & Ely, 2008). On the other hand, if change is costly and media attention tends to generate controversy, firms with eligible votes that attract more media attention may decrease in value compared to firms with eligible votes and less media attention. Media attention may also increase the likelihood that virtue signaling is exposed, which can sap investor confidence and damage firm reputation leading to further negative price adjustments (Du, 2015).

In addition to the univariate tests, we examine these variables jointly by estimating two forms of multivariate regression model, the first based on treatment and control firms where possible and the second based on treatment firms only. The first model iswhere Partitioning variable = A firm-level dummy variable for high versus low Media_Attention_i = one if the amount of media attention given to each firm-year based on a count of RavenPack documents (all types) is above the median and zero otherwise, and UNPRI_Signatory. For example, if we were to observe a significantly negative coefficient for β₃ for the Media Attention, this would indicate that high media attention more negatively associates with CAR (t₁, t₂) for treatment firms compared to control firms. For the partitions of Withdrawn/Voted, GHG_Reduction, Marginally_Passing_Vote, and Repeated, the treatment and treatment interaction variables drop out of Eq. (5) because there is no equivalent partitioning variable for the control sample.

The second model below applies to the treatment sample only and includes all the variables as regressors to estimate the marginal effects of each partition. The model is

The dummy variable regressors in Eq. (6) are the same as defined for Eq. (5). As in Tables 3, 4, and 5, the regressions control for year and industry fixed effects and use firm-year clustered standard errors. While several of the variables are correlated, we mainly expect the same signs for the coefficients in Eqs. (5) and (6) as in the univariate tests. The significance levels of the tests (and possibly the direction of the univariate relations) may differ, however, due to the correlations among the variables. We include Size and Leverage as firm-level controls in Eqs. (5) and (6), as they should be highly significant in determining whether a vote will occur, which would likely influence investors’ assessment of cost.

Our last research question asks whether climate-related proposals associate with a change in firms’ future environmental performance rating and emissions performance, a topic that has generated mixed results (Christensen et al., 2022) and controversies over whether such scores track real effects (Simpson et al., 2021; Tang et al., 2022). We measure environmental performance using the Bloomberg environmental score, which indicates a firm’s environmental performance based on publicly available and verifiable information on different themes, such as emissions performance and resource use. Given the controversy over the commensurability of outside ratings, we test the robustness of our results using an alternative rating measure (from Refinitiv) (“Additional tests” section). Our goal is to understand which characteristics of shareholder proposals might condition firms’ future environmental and emissions performance.

First, we conduct a univariate difference-in-difference analysis of the two performance measures, namely, ∆Env_Score_Treatment minus ∆Env_Score_Control and ∆Emissions_Treatment minus ∆Emissions_Control. If the proposals generate measurable effects, we expect the mean of the first difference to be positive and significant (environmental performance improves for treatment firms compared to control firms) and the mean of the second difference to be negative and significant (i.e., emissions decline for treatment firms compared to control firms). Also, if the firms targeted also have lower ratings or high emissions to begin with (compared to control firms), this is another reason to expect a general improvement over time compared to control firms.

Second, we conduct a multivariate regression analysis based on treatment observations only, as we have no equivalent data on relevant shareholder proposal characteristics for control firms except for whether a firm has high or low media attention or is in a carbon-intensive industry. Specifically, we assume an autoregressive process and regress Env_Score_t+2 or Emissions_t+2 (Scope 1 plus Scope 2) on Env_Score_t−1 or Emissions_t-1 and interact these measures with the characteristics of a shareholder proposal. The model is

Equation (7) thus regresses Env_Score_t+2 on Env_Score_t−1 or Emissions_t+2 on Emissions_t-1 with controls for firm size, leverage, and fixed effects. The regressions include an interaction variable to test whether the impact of climate-related proposals on environmental performance or emissions differs on that partitioning variable. The partitioning variables are Media_Attention, UNPRI_Signatory, Withdrawn, or voted proposal, whether the proposal relates to GHG_Reduction, proposals with a Marginally Passing Vote (35 ≥ % < 50), and whether the impact varies on the level of Carbon_Intensity. The number of observations for each regression depends on the variables and the partition. Because the regressions require measures of change over three years, the regression observations are smaller than in the earlier tables.

The key test from Eq. (7) is the sign and significance of β₃ for a particular partition. For example, if β₃ > 0 for Env_Score for the media attention partition, this would indicate that after controlling for the predictably positive (and sticky) correlation between average past and future environmental performance, the positive correlation between average past and future environmental performance is incrementally more positive for firms with more media attention versus less media attention, consistent with an improvement. We can also conceptualize 0 > β₁ ≤ 1 in Eq. (7) as the coefficient for a mean-reverting first-order autoregressive process and β₃ as an estimate of how much more of the first-order lag is passed into the series conditional on the partition, assuming 0 > β₁ + β₃ < 1.¹⁰

We also include UNPRI_Signatory as a partitioning variable to explore Kim and Yoon (2023), who find that ESG performance does not improve for firms with UNPRI signatories. Proposals submitted by UNPRI-aligned investors may not necessarily receive more votes in favor, however, as these investors could pressure firm managers directly to carry out more ethical business practices and engage more on environmental issues (Kim & Yoon, 2023). In addition, we distinguish between proposals targeting carbon-intensive firms versus other firms. Carbon-intensive firm proposals can generate significant publicity, yet the firms may be reluctant to improve their environmental or emissions record.

We first provide a breakdown of the sample by year and proposal type (operations- versus disclosure-related). Panel A of Table 1 states the frequency of climate-related proposals by year (cols. 1–5), the number of repeated-vote proposals (col. 8), and their voting percentage (col. 9), differentiating by disclosure- and operations-related proposals (cols. 10–11). First, cols. 2–5 of Panel B indicate a split between voted and withdrawn disclosure-related and voted and withdrawn operations-related proposals. Second, we observe a significant difference in the time trend of the percentage of disclosure- and operations-related proposals. Voting likelihood is higher for operations-related proposals, whereas disclosure-related proposals are likelier to be withdrawn. Based on separate time-series regressions of col. 3 and col. 5, the former (latter) increases (decrease) significantly (p < 0.01) over time. Relatedly, the percentage of votes in favor of operations-related proposals increases after 2015 (after the post-Paris Agreement), potentially an indication that investors now focus more on this type of shareholder proposal. Panels A and B together also reflect a greater preponderance and an increasing trend of negotiated outcomes for disclosure-related proposals (likelier to be withdrawn) compared to operations-related proposals.¹¹ The latter thus reflect more non-negotiated outcomes (since such proposals proceed to a vote). Given the firm’s involvement, a negotiated or withdrawn proposal is also less likely to impose costs on the firm compared to a repeated proposal or a proposal with a high percentage of votes in favor. We test this idea in a later section.

Table 2 summarizes the main statistics for the targets and their matched counterparts in two panels. Panel A summarizes the characteristics of the target and matched firms for the full sample. Panel B compares the summary statistics of the operations- and disclosure-related subsamples.

To create a control sample of firms not targeted by a climate change proposal, we match the propensity score of each firm with a climate-related proposal to a control firm that has not been targeted by size, year, and GICS industry group (“Non-targeted firms” section). As Dimson et al. (2015) note, firms targeted by an ESG shareholder proposal reflect lower levels of Sales_Growth, potentially indicating that climate-related activism can impose a reputational effect on firms’ actions. Size is a significant characteristic of both disclosure-related proposals and operations-related proposals. Firms targeted by an operations-related proposal indicate larger size than firms targeted by disclosure-related proposals. Firms targeted by an operations-related proposal show larger BTM and smaller Tobin_Q ratios than firms targeted by disclosure-related proposals. In addition, firms targeted by operations-related proposals invest less in R&D than firms targeted by disclosure-related proposals. Mean Capex is similar for both subsamples. Regarding capital structure, firms targeted by an operations-related proposal present lower levels of leverage than firms targeted by disclosure-related proposals. Lastly, we note that while the treatment firms and control firms attract similar levels of media attention, firms subject to disclosure-related proposals are far more newsworthy firms in general but have lower environmental performance scores, which is possibly a reason why they are targeted in the first place (Barko et al., 2021). While Table 2 describes the financial characteristics of targeted (treated) versus non-targeted (control) firms, those measures being univariate do not consider that some characteristics are correlated. Appendix Table 11 shows the correlations. In the next section, we apply logit regression analysis to identify those that are the most important conditional on the correlations.

Table 3 summarizes the estimation of Eq. (1). Col. 1 presents the marginal effects of being targeted for the full sample. The results are largely consistent with Table 2. We find that targets are larger and present higher levels of dividend yield, OCF (operating cash flow), Capex, and Tobin_Q relative to matched firms. Moreover, the RD variable has a negative and significant coefficient, suggesting that firms targeted by climate-related proposals spend less on R&D. We also find that firms in high carbon-intensive industries are more likely to be targeted by climate-related proposals. The number of independent board members also positively and significantly influences voting likelihood. In addition, climate-related proposals tend to engage firms with poor environmental performance relative to matched firms. We then split the sample into operations-related and disclosure-related proposals. For the operations-related proposals (more likely to change operations) versus disclosure-related proposals (less likely to change operations), based on the sign of the difference in the coefficients, the former set of firms tends to be smaller in size with higher growth prospects, has more capital spending and less R&D, is more likely to be in a carbon-intensive industry, and has an independent board and a higher environmental rating.

Table 4 reports the results of Eq. (2). Col. 2 reports fewer observations because the regression includes the governance variables and Env_Score. As shown in col. 1, the coefficients for Size and BTM are positive and significant. This suggests that climate-related proposals that engage large firms and low growth potential (BTM) firms are those likelier to receive a vote at the AGM.

Focusing on col. 2, the coefficients indicate that proposals targeting firms with lower levels of R&D are more likely to receive a vote. In addition, climate-related proposals submitted to firms in high carbon-intensive industries are likelier to receive a vote. By contrast, and similar to Table 3, targets' environmental performance negatively and significantly relates to climate proposals’ voting likelihood. In other words, proposals submitted to firms with lower environmental performance scores reflect a higher voting likelihood. Regarding other proposal characteristics, we find that voting likelihood is higher for firms with repeated proposals. In economic terms, an additional earlier proposal increases the probability of being voted on at the shareholders’ meeting by 10.7 percentage points (col. 1) or higher (col. 2). Additionally, the positive and significant coefficient for Operations_Related indicates that shareholder proposals related to climate risks are more likely to be voted on. This finding is consistent with the results in “Firms targeted with a shareholder proposal” section. In addition, proposals relating to Emissions_Reduction are more likely to be voted on. The negative and significant coefficient (p < 0.01) for Republican for climate-related proposals filed during Republican administrations suggests they are less likely to be included in the proxy materials and, thus, being voted on at the AGM. The coefficient for this variable is negative and significant at the maximum level for both models in Table 4. We find that non-profit activist investors associate significantly with lower climate-related proposal voting likelihood.

We estimate Eq. (3) and summarize the results in Table 5. The dependent variable for this model is the percentage of votes in favor that a climate-related proposal receives at the AGM. Thus, the sample comprises only those targets whose proposals receive a vote. The sample also controls for proposals that repeat earlier proposals. Regarding the firm-level characteristics, we find that proposals submitted to firms with higher environmental performance are likelier to receive a higher number of votes in favor at the AGM. Thus, even though targets with lower environmental scores exhibit a higher voting likelihood, those that score better environmentally in fact receive significantly higher vote percentages. By contrast, proposals that engage firms with higher levels of R&D are less likely to obtain a higher percentage of votes in favor at the AGM. This also applies to firms with more media coverage, indicating that how well known a firm is does not necessarily translates into more votes in favor of a resolution.

The UNPRI_Signatory variable relates positively and significantly to climate-related proposals’ voting outcomes. Unlike Kim and Yoon (2023), who find that UNPRI membership did not increase voting in favor of environmental proposals, Table 5 indicates that shareholder proposals with a UNPRI-signatory investor versus a non-UNPRI investor associate with a higher percentage of votes in favor at the AGM. We also find that operations-related proposals are likelier to receive a higher vote percentage in favor than disclosure-related proposals, possibly reflecting a stronger desire by ethical investors for organizational change rather than disclosure, even though the former may be costlier for the firm. Lastly, we find that repeated proposals and those related to emissions reduction are likelier to obtain more votes in favor at the AGM. Repeated proposals may also exert more pressure on firms for constructive change, which could have adverse effects on firm value. We examine this idea further in the next subsection.

A direct measure of how investors evaluate the success of shareholder activism is the market response. Following Gillan and Starks (2000), we select the proxy statement filing date as the date most likely to attribute a market reaction to shareholder activism. On or around this date, the marginal investor first learns that a firm has been engaged formally by a shareholder proposal. We measure the market response over days − 10 to 10 as per Eq. (5)¹² around the DEF14a filing date. Given that firms file the DEF14a electronically, this is the same date as the mailing date as per some earlier studies. On this date, we also assume that the marginal investor unbiasedly estimates that a proposal in a DEF14a will receive a vote at the AGM (although some could be withdrawn) and the percentage of shareholders that will vote in favor of a proposal, and knows whether the vote is the first or follows one or more prior votes or other outcomes, and knows the proposal type based on the DEF14a description. The marginal investor also assumes market efficiency regarding public information about the firm (e.g., from prior financial statements and prior news media reports that are already impounded into prices) (Prevost et al., 2016).

We report our investor reaction tests in Fig. 2 and Table 6. We first present the results on a univariate basis by testing whether the differential cumulative excess return over DEF14a days − 10 to 10 for the different partitions of the sample differs from zero based on a binomial test. Each of the plots in Fig. 2 states the probability that the differences in the mean excess returns for each of days − 10 to 10 for a partition are not significantly positive or negative when the expected likelihood of each being positive or negative is 0.5. In the same order as the columns in Panel A of Table 6, Fig. 2 documents five key results consistent with our cost hypothesis (H1): (i) Proposals with more versus less media attention generate negative excess returns (significant, p < 0.05) (Fig. 2a), (ii) withdrawn minus voted proposals generate negative excess returns (significant, p < 0.05) (Fig. 2b), and (iii) GHG-reduction proposals compared to other proposals generate negative excess returns (significant, p < 0.01) (Fig. 2c). In addition, proposals targeting high minus low carbon-intensity firms generate negative excess returns (p < 0.10) (Fig. 2d), and UNPRI-signatory proposals compared to other proposals generate positive excess returns (p < 0.05) (Fig. 2e). The other plots do not indicate a significant differential price response over days − 10 to 10 different from zero suggesting that factors at play in addition to cost are driving the results. Not significantly different from zero are the excess returns of operations- minus disclosure-related proposals (Fig. 2f), repeated minus single-vote proposals (Fig. 2g), and marginally passing proposals (Fig. 2h). Note that other than the results for the UNPRI-signatory partition (Fig. 2e), none of the tests indicates a significantly positive differential price reaction over days − 10 to 10, which would be a finding contradicting our main contention that investors are reacting to a particular climate proposal type based on expected cost to the firm.

To address the above findings as a multivariate analysis, Table 6 summarizes the regressions defined by Eqs. (5) and (6). These models regress CAR (t₁, t₂) for return intervals of days − 1 to 10 (Panel A) and − 1 to 10, − 1 to 1, − 10 to 10, − 5 to 5, − 10 to 0, and − 5 to 0 (Panel B) on the partitioning variables in two ways, namely, one at a time [Panel A summarizes Eq. (5)] and together [Panel B summarizes Eq. (6)]. In Panel B, we examine the consistency of the results across multiple cumulation periods. The results from Panel A are consistent with Fig. 2 for the following partitions. (i) Media Attention has a negative impact on excess returns compared to control firms (significant in Fig. 2a). (ii) Firms with withdrawn proposals experience more negative stock returns (significant in Fig. 2b), and (iii) UNPRI-signatory firms experience positive stock returns compared to control firms (significant in Fig. 2e). In addition, Panel A shows firms with marginally passing votes (35 < % ≤ 50) experience more negative stock returns (not significant in Fig. 2c). The marginal effects of a partition in Panel B are significant as follows: (i) GHG-Reduction proposals generate more negative returns (also significant in Fig. 2c) and (ii) UNPRI_Signatory proposals generate more positive returns (also significant in Fig. 2e). The other partitioning variables, while significant in Panel A and for some shorter CAR intervals in Panel B are not significant after controlling for the effects of the other characteristics of the climate-related proposals.

Tying the results in Fig. 2 and Table 6 together allows us to draw the following overall conclusion—climate-related proposals expected to impose costs on the firm assuming implementation are also those that prompt a negative price reaction by investors (in line with H1). Such negative price reaction is also consistent with ethical investors who engage the firm for change on climate issues as accepting lower returns as a consequence of the negative price reactions. With one exception, we also find no clear evidence that a particular proposal type generates positive excess returns. Even withdrawn proposals that may be negotiated with firm managers or repeated in the future generate negative excess returns. Nonetheless, proposals supported by a UNPRI-signatory associate with positive excess returns, consistent with the idea that large institutional investors can more easily engage the firm directly on climate issues and produce positive future outcomes for ethical investors such as a decrease in emissions (Azar et al., 2021). Overall, these results support our hypothesis (H1) that costlier climate-related proposals (e.g., GHG-reduction proposals) generate negative returns for investors versus the alternative that less costly proposals (e.g., UNPRI-signatory proposals) generate less negative or positive returns.¹³

Similar to the previous section, we conduct univariate and multivariate tests to explore whether environmental performance or emissions change from before to after a climate-related proposal. Should the proposals induce constructive change in the ways intended, we would expect the mean of the difference from year t−1 to t + 2 to be positive (environmental performance improves for treatment firms compared to control firms) and the mean of the difference in emissions to be negative (i.e., emissions decline more for treatment firms compared to control firms). We evaluate the significance of the difference in difference as a two-sample t test. Table 7 summarizes the analysis and Fig. 3 plots the results. We first show that the environmental performance of treated firms improves more than for control firms by 19.48 percent (50.41%–30.93%) over three years), but that the reduction in emissions is insignificantly different from zero (8.07%–7.06%) over three years (Fig. 3a and Rows 1–4 of Table 7). We next show that low carbon-intensity treatment firms improve their environmental performance more than high carbon-intensity treatment firms compared to control firms (by 27.43% = 31.62%–4.19%) (Fig. 3b and Rows 7 and 8 of Table 7). The treatment less control firm differences in emissions are not significant, however, for low (2.46%) or high carbon-intensity firms (− 6.96%) compared to control firms (Rows 11 and 12 of Table 7). Lastly, we show that firms targeted with an emissions reduction proposal versus other proposal types increase their environmental performance by 3.63 percent (37.32%–33.69%) and decrease their emissions slightly more than firms targeted with other proposals, by 4.53 percent (10.69%–6.16%), but the difference in difference is not significant (Rows 13–16 of Table 7). These results are univariate, that is they do not control for other factors not incorporated into the propensity score matching procedure that might also explain the differences.

Table 8 summarizes the regressions defined by Eq. (7). The results in cols. 1, 6, and 7 include observations for the treatment and control samples. The results in the other columns are relative only to firms in the treatment sample, as there is no equivalent partition for these columns for control firms. The coefficient for the interactions of Partitioning variable x Pre-proposal rating for environmental performance and emissions is our main focus. If positive, this indicates that the observations within the partition exhibit a higher growth in environmental performance or emissions from t−1 to t + 2, that is, an additional portion of the Pre-proposal variable is passed into future values of the series conditional on a partition.

The main results from Panel A of Table 8 are (i) that environmental performance growth is lower for withdrawn versus voted proposals (col. 3) and GHG-reduction versus non-GHG-reduction proposals (col. 4) and (ii) that environmental performance growth is no different for proposals involving UNPRI-signatory firms (col. 2) and marginally passing vote proposals (col. 5), and does not differ based on the level of media attention (col. 1) or carbon intensity (cols. 6 and 7). The main results from Panel B are (i) that emissions growth is higher for firms with high versus low media attention (col.1) and lower for firms with GHG-reduction proposals (col. 4), marginally passing vote proposals (col. 5), and low carbon intensity (col. 6) and (ii) that emissions growth is no different for the other partitions (cols. 2, 3, and 7) conditional on controls and fixed effects.

The results in Table 8 also align well with those in Table 6 and Fig. 2, in that several of the results reinforce each other intuitively. First, the negative excess returns for withdrawn proposals (Fig. 2b) and GHG-reduction proposals (Fig. 2c) versus other proposals associate with a drop in environmental performance (cols. 3 and 4 of Table 8). Second, as expected, GHG-reduction proposals (Fig. 2c) compared to other proposals result in a reduction in emissions (col. 4 of Table 8). Third, consistent with a negative price reaction for high minus low carbon-intensive firms (Fig. 2d), emissions do not drop for high carbon-intensity firms (col. 7) but drop for low carbon-intensity firms (col. 6). (iv) Proposals with marginally passing votes whose excess returns tend to be positive (Fig. 2h) associate with lower emissions (col. 5). Table 8 also aligns with Kim and Yoon (2023), who find that UNPRI-aligned investors do not significantly improve firms’ ESG performance following their decision to adhere to the principles. However, we caution against a direct comparison as Kim and Yoon (2023) posit that PRI investors use “cheap talk” and greenwashing to attract funds to increase their assets under management, which is fundamentally different from the conceptual underpinnings of this paper, where shareholder voting on climate proposals reflects the expected costs to the firm of the outcomes of the different types of climate proposal.