Are corporate climate efforts genuine? An empirical analysis of the climate ‘ talk – walk ’ hypothesis

This study conducts machine-aided textual analysis on 725 corporate sustainability reports and empirically tests whether climate ‘ talk ’ within the sampled reports trans-lates into performance ‘ walk ’ , proxied by changes in greenhouse gas emissions over a 10-year period. We find mixed results for the ‘ talk – walk ’ hypothesis, depending on the type of talk and the associated climate change actors involved. Indeed, our empirical models show that while some climate commitments are genuine, many constitute little more than ‘ greenwashing ’ — producing symbolic rather than substantive action. We attribute this result to false signalling of climate transitioning in order to mislead due to misaligned incentives. An unexpected positive finding of the study is that talk about operational improvements is a significant predictor of climate performance improvement. On the other hand, reactive strategies are consistent with poor climate performance. Our findings highlight the significance of corporate climate strategies other than emissions reductions in assessing the effective contribution of business to the climate transition.

annually, companies can commit to climate action and, ideally, follow through on commitments. Alternatively, recent research has focused on the effect of firms' participation in 'Corporate Climate Initiatives' (CCIs) (Coen et al., 2021). CCIs, such as the Science Based Targets Initiative (SBTi) and the CDP (formerly the Carbon Disclosure Project), undertake various activities to enhance corporate climate performance, including third-party verification of emissions reporting and the development of scientifically informed climate targets. A small but growing body of scholarship explores the mitigation potential of such climate initiatives, including how normative commitments to corporate climate action can be reinforced. Participation in CCIs is, therefore, a plausible proxy for assessing corporate commitment to climate transitioning (Schneider, 2020).
We build upon the research into CCIs and sustainability reporting more generally to address a central question: when it comes to climate action, firms increasingly 'talk the talk' but do they 'walk the walk' (Backman et al., 2017, p. 569)? In other words, do firms that publicly commit to the climate transition (vis-a-vis sustainability reporting and engaging with CCIs) follow through with climate action, the most visible of such activities being an improvement in emissions performance over time?
Prior research highlights the perennial problems with private sector climate data, not least the extensive number of variables available from various data platforms (including Thomson Reuters, Bloomberg and Sustainanalytics) (Busch et al., 2020;Callery & Perkins, 2021).
Ongoing concerns with data anomalies and inconsistencies surrounding corporate climate metrics necessitated the original approach to data extraction, transformation and modelling techniques employed in this paper. Our approach exploits machine-aided content analysis to build 12 explanatory variables. In addition, due to the high degree of missing observations for firm-level climate and environmental data, we rely on firm-level emissions changes as our proxy for 'climate walk'. This methodological approach constitutes an original contribution to tackling this vexing problem and we hope will encourage scholars to further advance the state-of-the-art on modelling corporate climate metrics.
Our empirical results deliver some unexpected findings with regard to the talk-walk hypothesis. In line with related research (Damert et al., 2017;Diouf & Boiral, 2017;Doda et al., 2016), it appears that while some commitments to climate transitioning are strongly linked to substantive climate action in the form of GHG performance, a larger proportion of corporate climate talk does not translate into climate walk. Overall, we conclude that positive communication on corporate climate strategies and commitments are largely symbolic in nature (Bowen, 2014;Delmas & Montes-Sancho, 2010;Doda et al., 2016). Our results support the contention that voluntary, bottom-up climate governance systems alone are inadequate for dealing with the massive, systemic problem of emissions mitigation in the corporate and private sectors (Berliner & Prakash, 2015;Dingwerth & Eichinger, 2010;Diouf & Boiral, 2017;Marimon et al., 2012). Recent advances in climate regulation are promising. However, concerns persist over whether corporate climate commitments will translate into action absent explicit sanctions for non-compliance. This paper begins by outlining the research gap in empirical methods when it comes to assessing corporate climate walk versus talk. We then locate corporate sustainability reporting within the literature on 'cheap talk' (Farrell & Rabin, 1996), discrete 'greenwashing' activities (Bowen, 2014), as well as 'tick-box' compliance exercises (Marquis et al., 2016). Thereafter, we introduce the private-led climate initiatives which have proliferated in response to demands to align firm's emissions trajectories with climate science. Section 3 presents the data and methodology, paying particular attention to the construction of our explanatory variables based on machine-aided text mining of climate talk. Section 4 discusses the empirical results, exposing an undercurrent of greenwashing in corporate climate transitioning. The paper concludes by highlighting the implications of our findings for corporate climate governance and policy, with particular emphasis on strengthening the emerging EU Taxonomy on Sustainable Investment. 1

| CORPORATE ACTION ON CLIMATE: FROM STRATEGY TO IMPLEMENTATION
It is now well established that steep emissions cuts by the private sector are vital to the success of the Paris Climate Agreement (Walenta, 2020). Yet, despite advances in research, there are few studies research which causally assess the impact of outward-facing corporate climate strategies (such as sustainability reports or 10k filings) on climate performance, as opposed to a more narrow focus on carbon performance alone (see Doda et al., 2016). We respond to this deficit and build upon studies which have leveraged corporate sustainability textual data to ascertain the degree to which corporate climate transition efforts are genuine and substantive (Boiral et al., 2020;Talbot & Boiral, 2015). And, while our modelling approach rests on the climate talk-walk hypothesis (Backman et al., 2017;Green et al., 2021) (Unruh et al., 2016). But first, it is necessary to explain the nuances involved in identifying climate-transition aligned companies, a vexing challenge which we turn to now.  (Abbott & Monsen, 1979), it has increased markedly in recent years due to the growth of mandatory reporting requirements across domestic jurisdictions (Hahn & Kühnen, 2013;Marimon et al., 2012).
With the addition of the EU Taxonomy on Sustainable Investment and delegated acts, 2 regulatory demands for more transparency in the climate transition are set to grow.
Sustainability reports and the information contained thereinnotwithstanding valid concerns flagged in the following section-can provide useful data if properly cleaned and integrated such as through machine-aided textual analysis (Connelly et al., 2011;Dragomir, 2012). Interestingly, although such data can shed light on whether companies are making the substantive efforts required to meet the international target to limit global warming to well below 2 C, rather than following a business-as-usual pathway (Levy, 1997;Milne et al., 2006Milne et al., , 2009Tregidga & Milne, 2006), 'researchers have rarely used the actual data contained in these reports' (Dragomir, 2012, p. 233).
Critical scholarship, however, is dubious of the claim that sustainability reports provide credible signals. So-called 'window dressing' sustainability reporting and the self-serving motivations underlying such practices have been interrogated in detail (Kolk & Perego, 2014).
Indeed, literature focused on corporate 'greenwashing' suggests that public commitment is often not aligned to actual performance (Bowen, 2014;Lyon & Maxwell, 2011). What these practices have in common is the intentional use of environmental 'talk' to camouflage the absence of real behavioural change (Moneva et al., 2006).

| Standardising sustainability reporting
Concerns surrounding corporate self-reporting and conflicts of interest have long been recognised in the Environmental, Social and Governance (ESG) literature (Lindblom, 1994;Milne et al., 2006). This body of scholarship suggests one possible solution lies with non-state climate change actors, such as the Global Reporting Initiative (GRI) and the CDP (formerly the Carbon Disclosure Project), in their mission to enhance the reliability of corporate ESG and climate disclosure (Cho et al., 2012). Below we discuss these actors and weigh up the contribution to corporate disclosure of this largely private climate governance system.

| Non-state actor support of sustainability and climate initiatives
The involvement of the corporate sector in climate change governance has been a feature since the earliest days of the UNFCCC, for good and ill (Newell & Paterson, 1998). Corporations not only have a business interest in shaping the international climate policy agenda but have also benefited from the neoliberal turn which places significant trust in their capacity to act as climate action leaders (Sapinski, 2015).
Reflecting deeper critiques of decades-long neoliberal policy creep towards bureaucratic rationalisation and the privileging of 'efficiency' and 'competition' metrics over state oversight (Chakrabarty & Wang, 2013), some scholars maintain that, while companies might gain legitimacy from self-reporting procedures, such procedures do little to address the underlying environmental challenge (Talbot & Boiral, 2015). Given such concerns, researchers have begun to focus on a new class of non-state climate change actors with an express mandate to ensure consistency in climate transition policies across the private sector. Indeed, the EU's Taxonomy explicitly endorses the Taskforce on Climate-related Financial Disclosures (TCFD), a prominent member of this new generation of non-state climate actors which we integrate into our models below.
During the formative years of the UNFCCC process, sustainability-oriented business groups such as the WBCSD (World Business Council for Sustainable Development) and the CERES (Coalition for Environmentally Responsible Economies) were formed to promote corporate sustainability action and to ensure that business had access to decision-makers in high-level intergovernmental forums (Wright & Nyberg, 2015).
As noted in the introduction, CCIs are voluntary, non-state actors, often spearheaded by longstanding business groups such as CERES and the WBCSD. The remit of CCIs expands beyond emissions mitigation efforts to include internal working procedures within the corporate sector and targeting key decision-making domains at management level. Indeed, while some CCIs focus on corporate emissions improvement specifically (e.g., the CDP), others address corporate climate and environmental reporting (e.g., the GRI).
For instance, the GRI-which was launched by CERES in 1997focuses on sustainability disclosure; it is now the world's largest repository of corporate sustainability reports and has spearheaded transparency and disclosure across the sector. The GRI disclosure framework stipulates that firms should present information 'in a manner that enables stakeholders to analyse changes in the organisation's performance over time, [which] could support analysis relative to other organizations' (Global Reporting Initiative, 2013, p. 14).
Beyond the broader ESG-focused CCIs discussed above, others focus on disclosure of corporate emissions. These CCIs include the CDP and the Carbon Disclosure Standards Board (CDSB), which was initiated to streamline International Accounting Standards (IAS) for carbon reporting. CDP claims to collect and distribute 'high quality emissions information to investors, enterprises and governments to prevent dangerous climate change' and to 'to accelerate solutions to climate change' (CDP, 2021). By 2020, CDP had 560 investor signatories with US$106 trillion in assets disclosing their environmental data through its services; by the end of 2021, nearly 10,000 firms disclosed climate or emissions data to the CDP.
However, in an important study which bears on our research focus, Doda et al. (2016) demonstrate that while the CDP has induced companies to improve their due diligence around reporting emissions, this increase in reporting has not translated into participating firms improving their carbon performance. Based upon this finding, these author conclude that corporate carbon management practices, underscored by participation and disclosure to the CDP, are by and large not exerting a positive measurable effect on carbon emissions performance in the corporate sector.
Another significant non-state actor which we include in our analysis is the CDSB, which works towards making carbon emissions' disclosures more consistent across sectors and companies. Indicative of the influence of private actors over the corporate carbon disclosure ecosystem, the CDSB explicitly endorses the CDP and GHG Protocol-the latter being another emissions measurement tool initiated by the WBCSD. The CDSB has become a key node in the network of corporate carbon disclosure and reporting, as shown in this network analysis of the carbon-based governance regime (see: https://www.globe-project.eu/en/carbon-governance-regime_ 10461). facilitates public companies and other organisations in disclosing climate-related risks, both physical (e.g., stranded assets) and regulatory (e.g., climate-related reporting requirements). Relatedly, the Science-based Targets Initiative (SBTi) requires participating firms to set 'scientifically informed' emissions targets and independently verifies whether targets are 'science-based,' which means aligning with the 'well-below 2 C' as set out in the Paris Agreement (Science Based Taggets, 2017a, 2017b). These two CCIs have made inroads into public climate governance, with the TCFD explicitly integrated within the EU Taxonomy, and the SBTi increasingly prominent in multisectoral efforts to enhance consistency of climate transition targets.
Based on the remit of these CCIs, we expect that if companies participate in these CCIs and 'talk' about the constituent climate alignment frameworks within their sustainability reports, they are more likely to be walking the climate walk. In other words, the expectation is that 'talk' about these CCIs within sustainability reports is indicative of substantive climate action, as opposed to empty rhetoric.
It is important to note that, while we have outlined some of the actors which are integrated into our empirical models, the full list of actors and initiatives empirically examined is specified below (Table 1). More details on these actors can be found on their websites or, alternatively, in the network analysis we conducted. 3

| Research question and hypotheses
As outlined above, an increasingly sophisticated debate centres on the limitations of data derived from corporate sustainability reports (Boiral, 2013;Connelly et al., 2011). Careful attention to debate on the credibility of corporate sustainability reporting guides our empirical modelling technique. In particular, machine-aided content-analysis is widely used in the existing literature to ameliorate such concerns (Boiral et al., 2020;Diouf & Boiral, 2017;Dragomir, 2012;Radu et al., 2020;Rekker et al., 2021;Talbot & Boiral, 2018). In a pioneering study using this method, Dragomir (2012) conducts automated text analysis methods to estimate the extent to which five major European oil and gas companies have adhered to the requirements of the GHG protocol. Her findings indicate that, at least within the sampled sustainability reports, these companies often deliver inconsistent climate and emissions data. From a practical standpoint, this is surprising (and alarming) given that the GHG Protocol, in addition to the GRI-two initiatives which our included in the Dragomir sample-lay down stringent guidelines on how to measure and properly disclose emissions.
Surveying a larger sample of companies, Talbot and Boiral (2018) examine corporate sustainability reports disclosed through the GRI and arrive at similar conclusions. The authors leverage natural language processing (NLP) to identify specific climate strategies contained within the report texts and find that-despite GRI having rigorous standards and guidelines on climate disclosure-reports largely constitute 'impression management' strategies rather than substantive commitments to change. Impression management takes the form of both data distortions and concealment. As such, dispiritingly, Talbot and Boiral (2018, p. 369) conclude that participation by firms in the GRI more closely aligns with a 'logic of public relations than by that of transparency' making it difficult for stakeholders to reasonably assess, monitor and compare companies' climate performance on the basis of these reports. Finally, Diouf and Boiral (2017), employing similar techniques, show that firms publish excessive climate data to conceal poor performance through information overload.
However, even though these researchers have leveraged machine-aided content analysis of sustainability reports, none have taken the extra step of calculating how corporate climate talk-as derived from sustainability reports-relates to climate walk, or changes in emissions performance over time. To the best of our knowledge, this is the first paper to regress corporate climate performance (proxied by emissions) on the textual data mined from sustainability reports.

| Hypothesis
Given the singular focus on emissions as the key performance indicator for corporate climate transitions-indeed, GHG performance is perhaps the 'most important question raised by external stakeholders on firm-level behaviour regarding climate change impact mitigation' (Backman et al., 2017, p. 569)-we expect that changes in emissions over time should be consistent with climate talk, as derived from sustainability reports.
In line with several related empirical analyses of corporate sustainability reports and GHG performance (Backman et al., 2017;Damert et al., 2017;Patten, 2002;Trumpp & Guenther, 2017;Walker & Wan, 2012;Whelan et al., 2019), and consistent with our main research question of whether corporate climate talk leads to climate walk, our central hypothesis can be formulated as follows: corporate climate talk, as published in annual sustainability reports, leads to corporate climate walk or improvements in emissions over time.
T A B L E 1 Dictionary categories and keywords (initiatives and actors) constructed through our own analysis of the carbon governance regime (Coen et al., 2021), and based on prior literature (Huq & Carling, 2020;Lock & Seele, 2016;Radu et al., 2020)

| Sample
We created a sample based on the FTSE-100 (100 companies) and the Dow Jones Industrials (30 companies) from 2010 to 2019. These samples were selected based on data availability, climate disclosure regulations (Robertson & Samy, 2015), and because it is essential that large multinational companies reduce their GHGs (Hahn et al., 2017).

| Content analysis
The aim of automated content analysis and natural language processing (NLP) is to break down textual data through automated and systematic classification techniques (Manning & Schutze, 1999;Mora et al., 2020). This involves coding, processing, and other automated techniques in order to identify key patterns, categories, trends and themes (Boiral et al., 2020;Hsieh & Shannon, 2005). Other advantages include reduction of biases and the enabling of more finegrained empirical analysis such as statistical models. Therefore, because NLP deals quickly and efficiently with large amounts of qualitative data-and can provide statistical inferences from the textual content-these methods are particularly well-suited for corporate reporting (Bowman, 1984;Krippendorff, 2018), and even more for corporate sustainability and ESG reporting (Boiral, 2013;Demaria & Rigot, 2021;Milne & Adler, 1999).

| Empirical steps: Data collation and transformation
Previous content analysis research spells out the specific steps to be carried out during the process of obtaining, cleaning and extracting data (Aykol et al., 2013;Weber, 1990). Below, we list these steps along with the specific research decisions we have taken:

| Explanatory variables
Marking an important methodological contribution to the scholarship, to construct our explanatory variables of interest, we first rely on the dictionary approach used in related literature (Cho et al., 2012;Lock & Seele, 2016;Radu et al., 2020). We then leverage statistics based on the rate of keywords mentioned within sustainability reports per

| The model
After constructing emissions as the dependent variable, a lag can be incorporated into the model to capture delayed effects of climate 'talk' on climate 'walk'. Methodologically speaking, lagging the dependent variable can also 'mitigate potential concerns about the direction of impact between the dependent and independent variables' (Hsueh, 2019, p. 63 • Y is the dependent variable, the natural log of emissions, indexed by i firm, k Scope Emissions, and t year.
• CAT represents the string of 12 explanatory 'climate talk' variables indexed by i firm and t year. Each is expressed as the rate per 10,000 words of any keyword within the category appearing in the sustainability report, for each firm-year observation from 2010 to 2019.
Our method is summarised in Figure 2.

| RESULTS
Overall, we find mixed results for the climate 'talk-walk' hypothesis.
First, we provide a short summary of the findings, and then we go into These unexpectedly strong results should be investigated further.
Below we report the regression results (16 total models) in Tables 2-5, , followed by a summary of the results in Tables 6 and 7.

| Results and discussion
We tested the talk-walk hypothesis on a total of 16 models. We produce significant findings across nearly all explanatory 'talk' variables. We begin first with the variables which lend support to our hypothesis: climate talk is matched by walk. Among the most important findings is that TCFD talk is followed by climate walk: for this variable the tenth and fourteenth models show that the TCFD is significant at the 1% level; in the first, fourth and fifteenth models it is significant at the 5% level, while the third, ninth, eleventh, twelfth and sixteenth models exhibit significance at the 10% level. This shows that TCFD climate talk leads to statistically significant reductions in emissions, confirming that it is an important step forward for climate governance (Demaria & Rigot, 2021). We contend that this is because TCFD is likely to have 'regulatory teeth', as it is already incorporated into the EU's Taxonomy on Sustainable Investment.
Next, the explanatory variable we have designated the 'carbon actors' 5 also provides substantial evidence in favour of the talk-walk hypothesis: three models are significant at the 1% level (models 1, 3 and 4), four models are significant at the 5% level (models 2, 5, 7 and 11), and three models are significant at the 10% level (models 6, 8 and 10)-although model 6 exhibits a positive coefficient. This finding suggests that the 'carbon actor' climate talk does, more often than not, translate into climate walk.
Turning to more troubling findings, the CDP fails to demonstrate any significant effect apart from the fourth and eight models, where it is positive and significant at the 5% level, meaning that climate talk is actually related to rising emissions-climate walk in the wrong direction. Since only two of 16 models exhibit significance for the CDP, this finding deserves further investigation in future research. However, the picture is even more problematic for the other climate actors. The GRI is positive and significant at the 5% level in two (9 and 13), and positive at the 10% level in seven models (2, 5, 10, 11, T A B L E 4 Regression results lagged 2 years (models 9-12) Regression    14, 15 and 16). We interpret this finding, therefore, to indicate that the GRI is providing cover for greenwashing activity; when firms talk more about GRI, their emissions performance tends to deteriorate.
This finding supports the widespread contention that the GRI is principally used by firms for symbolic or branding purposes, leaving false positives unaccountable-firms that commit to climate transitioning with little intention of complying (Dingwerth & Eichinger, 2010;Marimon et al., 2012;Walker & Wan, 2012). More worryingly still is the implication that these bad faith actors are actively maintaining these voluntary elements of the carbon disclosure system to ward off the threat of more robust regulation. Again, this finding has important implications for the pending EU Taxonomy on Sustainable Investment.
The SBTi, which is tasked with tracking corporate climate commitments, is also positive and significant at the 1% level in three models (7, 11 and 16); it is positive and significant at the 5% level in two models (5 and 12), and at the 10% level in model 9. Therefore, even though the SBTi has received much positive attention of late, we find that, based on the available data, it also appears to be doing more harm than good in concealing poor performance in firm-level emissions. However, we caution against drawing firm conclusions from this finding since, among the companies sampled, there has not been much 'SBTi talk' until relatively recently.
The 'climate governors' (e.g., WBCSD, CERES and WWF) perform poorly: this variable is highly significant and positive in the first four models (at 1%), at 5% statistical significance in three models (6, 7 and 11) and at 10% level in the fifteenth model. What this suggests is that large, well-known corporate environmental groups are not helping to drive emissions reductions in the corporate sector; indeed, they may even be helping companies to mask and camouflage their non-climate aligned behaviour. Finally, the 'climate initiatives' are only significant in three models. As such, while there is clearly room for improvement, the results are largely inconclusive for this group.
Moving on to the categories derived from the dictionary methods from prior literature (Dragomir, 2012;Huq & Carling, 2020 strategies tracks with climate walk at the 5% level in three models (4, 8 and 16), and at the 10% level in two models (5 and 12). This suggests that getting firms to articulate precisely how they have improved their GHG performance, and related climate strategies, helps them to improve (Huq & Carling, 2020). However, when firms talk more about environmental partnerships, the findings are mixed; interestingly, this is the only 'talk' variable which shows several negative and positive coefficients; we cannot draw any concrete conclusion, therefore. In three models, talk tracks with walk (1, 5 and 9); yet in five separate models, talk is not matched with climate walk as it has a positive and significant coefficient (2, 6, 10, 14 and 16). Finally, regulatory reporting displays a talk without walk finding in three models (1, 3 and 4) but does not demonstrate any statistical significance elsewhere and is therefore inconclusive.
This confirms the recent findings on sustainability reports using similar dictionaries in Radu et al. (2020). What is more, the coefficients for operational improvements largely exhibit statistical significance at 1% level, with the exception of three models (6, 10 and 12). This can be interpreted to mean that firms which change the way they operate-and indeed take structural actions for the climate transition-are much more likely to walk the climate walk. This contrasts with firms which talk about emissions reductions through climate initiatives such as the CDP, GRI and the SBTi, yet fail to improve on performance. Therefore, regulators and sustainably minded investors should pay closer attention to climate talk in relation to operational improvements, which show that substantive rather than rhetorical action is underway.
Finally, and in sharp contrast to the significant findings of operational improvement talk, when firms talk about 'reactive strategies' (keywords and categories adopted from Radu et al., 2020) they generally do not commit to climate walk. Indeed, the 'reactive strategies' variable is highly significant and positive at the 1% level in six models (3, 4, 6, 7, 8 and 12), significant at the 5% level in 6 models (1, 2, 10, 13, 14 and 15), and at the 10% level in model 5. These significant findings confirm Radu et al. (2020). These last two operationslinked climate talk variables demonstrate the highest overall statistical significance across all models. Certainly, these findings deserve more sustained attention in future research.

| Robustness
We conducted a series of further regression analyses to improve the robustness of this study. First, we specified emissions performance as emissions to sales, a normalisation based on a financial metric, as specified by Weinhofer and Hoffmann (2010). We also tested absolute emissions instead of natural logarithm. Finally, we constructed models with categories from related research. The findings are largely consistent with those reported above, providing further support to our conclusions.

| Limitations and future research
We highlight several limitations of this study. First, using the natural log of emissions for the dependent variable carries with it certain challenges. As others have highlighted, carbon emissions data at the firm level remains problematic and sometimes unreliable (Busch et al., 2018;Callery & Perkins, 2021). As Talbot and Boiral (2018, p. 377) note, 'the comparison of data over time and between companies in the same sector becomes an arduous and approximate task.
Companies' tendency to underestimate their emissions and to provide incomplete information helps create an idealised image of their situation'. We sought to ameliorate these issues by drawing on a larger sample of companies and using a relatively long time-series model, but it is not yet possible to entirely eliminate such concerns (Busch et al., 2020).
Moreover, while emissions performance constitutes an important metric for corporate climate strategies (Hoffman, 2007)-and, hence, climate talk-scholars have found fault in empirical analyses using these modelling specifications (Delmas & Nairn-Birch, 2011;Downie & Stubbs, 2012;Lee, 2012). Because firms tend to practice 'selective disclosure' of Scope emissions, this implies that the data can be inconsistent (Callery & Perkins, 2021;Marquis et al., 2016).
More efforts should be made to construct stronger emissions performance metrics in the future.
The present study could be extended in several directions. First, the occurrence of greenwashing in corporate climate 'talk' remains an important research focus, especially as corporate climate disclosure receives greater attention in regulatory debates worldwide. The capture of the corporate climate disclosure system, which seems to favour corporate interests at the expense of the environment, seems very likely (Cho et al., 2012;Hummel et al., 2019;Moneva et al., 2006;Talbot & Boiral, 2015). In addition, future empirical research could expand on the idea of corporate political activity (Hillman & Hitt, 1999), and how this relates to the corporate climate disclosure system. There is also the question of what institutional redesign of corporate disclosure systems is required to prevent 'dirty firms' from laundering their image through virtually costless climate 'talk' (Lyon & Maxwell, 2011;Pinkse & Kolk, 2009;Clark & Crawford, 2012 Doda et al., 2016;Dragomir, 2012). These weaknesses include the exploitation of schemes such as the GRI for impression management (Dragomir, 2012;Gupta & Mason, 2016;Talbot & Boiral, 2018), the institutionalisation of corporate sustainability (Boiral & Gendron, 2011), compounded by a lack of clear 'scientific' guidance coming from actors such as the SBTi (Walenta, 2020). While schemes such as CDP and GRI may be well-intentioned and could- However, one possibility is that the technical challenge of policing discrete cases of corporate greenwashing exceeds the capacities of these organisations. In addition, political economy factors, including misaligned incentives, are also likely to play a role (Newell & Paterson, 1998). Either way, our findings point to the unsettling possibility that these climate 'governors' have not only enabled companies to talk the climate talk without walking but also served to shield companies, which are walking in the wrong direction, from reputational damage.
Several policy insights stem from our analyses. First, regulators should pay greater attention to preventing private-led climate disclosure systems from mistaking climate talk for walk. Second, sustainable investors could be encouraged to undertake greater due diligence to 'read between the lines' and scrutinise corporate climate actions beyond emissions reductions. Our novel methodology could prove fruitful in this regard. Third, the proposed directive from the EU-the Taxonomy on Sustainable Investment-is a promising regulatory development. However, to fulfil its potential, the legislation will need to integrate a systemic understanding of how climate 'talk', including private-led actors and apparatuses, interrelate and in some ways overlap one another, which may undermine the overarching goals of the system.
There is a glimmer of hope, however, in the TCFD and the Carbon Actors (VER, Gold Standard, VOS, etc.). Our findings suggest that this class of climate actor appears to be positively influencing emissions performance: more talk about these actors, and making use of their frameworks, leads to climate performance improvements. This makes sense since Carbon Actors, which include third-party carbon emissions verification, take pains to ensure companies measure, disclose, verify and claim carbon credits consistently and accurately.
Are firms ready for the energy transition? The results of decades of CSR and now ESG are disappointing, highlighting longstanding problems with transparency-based self-regulation in the absence of explicit sanctions for non-compliance (King & Lenox, 2000). The narrative tropes used by corporations describing sustainability as a 'journey' have too often served as a pretext for not providing concrete data, metrics or 'quantifiable boundaries' and concrete climate actions taken today (Milne et al., 2006, p. 821

ACKNOWLEDGMENTS
We would like to thank especially Adam Holesch, Jared Finnegan, Jacint Jordana, Axel Marx and Angel Saz-Carranza for helpful comments and suggestions. We are also very grateful to the anonymous reviewer and journal editor for their feedback on this piece. This study was supported by the European Commission's Horizon 2020 Program (grant no. 822654).

ENDNOTES
1 The EU has also proposed a Corporate Sustainability Reporting Directive (CSRD) with the aim to make private sector sustainability reporting consistent to drive financial firms and investors. Such legislative initiatives are largely consistent with the recently approved EU Green Deal, with the overall aim to draw in non-state actors such as the TCFD, and to a lesser extent, the SBTi (Science-based Targets Initiative) (COM/2021/188 final). https://ec.europa.eu/info/business-economyeuro/banking-and-finance/sustainable-finance/eu-taxonomysustainable-activities_en 2021, and formally adopted on 4 June 2021 for scrutiny by the co-legislators. A second delegated act for the remaining objectives will be published in 2022. The publication of the first delegated act was accompanied by the adoption of a Commission Communication on 'EU taxonomy, corporate sustainability reporting, sustainability preferences and fiduciary duties: Directing finance towards the European green deal' that aimed at delivering key messages on how the sustainable finance toolbox facilitates access to finance for the transition. This Communication builds on the transition finance report adopted by the Platform on Sustainable Finance in March 2021 (https://ec.europa.eu/info/businesseconomy-euro/banking-and-finance/sustainable-finance/eu-taxonomysustainable-activities_en).