Sentimental Sustainability: Does What Companies Say Tell More Than What Companies Do?

Ignatov, Konstantin; Rudolf, Markus Article — Published Version Sentimental Sustainability: Does What Companies Say Tell More Than What Companies Do? Financial Markets, Institutions & Instruments Provided in Cooperation with: John Wiley & Sons Suggested Citation: Ignatov, Konstantin; Rudolf, Markus (2023) : Sentimental Sustainability: Does What Companies Say Tell More Than What Companies Do?, Financial Markets, Institutions & Instruments, ISSN 1468-0416, Wiley, Hoboken, NJ, Vol. 32, Iss. 4, pp. 221-252, https://doi.org/10.1111/fmii.12181 This Version is available at: https://hdl.handle.net/10419/288197 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ DOI: 10.1111/fmii.12181 ORIGINAL ARTICLE Sentimental Sustainability: Does What Companies Say Tell More Than What Companies Do? Konstantin Ignatov1Markus Rudolf1,2 1Allianz Endowed Chair of Finance, WHU-Otto Beisheim School of Management, Vallendar, Germany 2Chairholder, Allianz Endowed Chair of Finance, WHU-Otto Beisheim School of Management, Vallendar, Germany Correspondence Konstantin Ignatov, Allianz Endowed Chair of Finance, WHU – Otto Beisheim School of Management, Burgplatz 2, 56179 Vallendar, Germany. Email: k[email protected] December 2021 (this version: June 2023) *Konstantin Ignatov and Prof. Dr. Markus Rudolf are with Allianz Endowed Chair of Finance, WHU - Otto Beisheim School of Management. Correspondence: Konstantin Ignatov, Allianz Endowed Chair of Finance, WHU - Otto Beisheim School of Management, Burgplatz 2, 56179 Vallendar, Germany. Email: k[email protected]. Abstract Based on a sample of more than eleven thousand unique 10-K reports of US companies filed with SEC in period 2013 to 2018, this study examines the relationship between actual sustainability performance of companies, evaluated by MSCI ESG performance scores, and the extent and the scope of environmental, social, and governance information disclosure in their annual reports. The study shows empirical evidence supporting the signalling theory view of voluntary disclosure of ESG information in annual reports for most industries, while environmentally unfriendly companies belonging to the Mining industry division show excessive reporting behavior favoring environmental topics, which is consistent with incentives to improve public image and mitigate social, political, and legal risks in line with the legitimacy theory of information disclosure. When differentiating between forward-looking and non-forward- looking ESG statements, the study shows that companies providing more forward-looking ESG information in annual reports show better next-year ESG performance. This study implements established content analysis techniques with focus on ESG reporting and performance, building up on the study of Baier, Berninger, and Kiesel (2020) that proposed an ESG-tailored dictionary for textual analysis purposes. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2023 New York University Salomon Center. Financial Markets, Inst. & Inst. 2023;32:221–252. wileyonlinelibrary.com/journal/fmii 221 222 IGNATOV AND RUDOLF 1INTRODUCTION In recent years, the topic of sustainable economic activity and corporate social responsibility has become firmly rooted in the daily agenda of companies from all industries, including the investment and asset management universe. The main aspect of these concepts implies the reassessment of economic behavior of companies and other economic actors in terms of their impact on the environment and society as a whole, thus urging the integration of environmental, social, and governance (ESG) principles into operational and strategic choices of companies and government entities (Liang and Renneboog (2020)). Following the global economic sustainability trend, the investment management industry has been vigorously shifting its focus towards sustainable investing, with currently more than 80% of institutional investors adapting their investment decisions based on ESG factors (Morgan Stanley (2020)). While the integration of sustainable investment principles into investment processes continues to gain momentum year after year, with the global adoption rate of ESG principles among asset owners growing from 70% to 80% between 2017 and 2019, the investors are looking for better methodologies and more profound data to measure the impact of sustainability policies and activities of their investment targets and thus the success prospects of their investment strategies. The recent studies indicate that almost a third of institutional investors are not satisfied with available ESG data and assessment methodologies, citing this factor as one of the most crucial challenges for sustainable investing (Morgan Stanley (2020)). Besides difficulties in assessment of actual ESG performance of companies, both the research community and professional investors struggle with an exact definition of sustainability and its factors, which complicates the consideration and subsequent implementation of ESG policies in companies as well as their accurate performance assessment, thus leaving companies and rating agencies a lot of leeway in interpretation of sustainability criteria (Kotsantonis and Serafeim (2019), Trahan and Jantz (2023)). This vagueness also exacerbates the agency problems between the investment managers and the beneficiaries they represent, which ignites debates about the usefulness of imperative disclosure requirements of ESG practices that could be introduced by regulatory agencies, such as the Securities and Exchange Commission (SEC) (Mahoney and Mahoney (2021)). While the implementation of ESG policies in business operations entails many difficulties for company’s management, the communication with investors and reporting of sustainability topics to a wide circle of stakeholders constitutes an equally challenging task (Aluchna, Hussain, and Roszkowska-Menkes (2019)). Especially in the US market, the investors have been putting pressure on companies to provide them with more information regarding sustainability efforts and their impact on businesses, leading to increased adaptation of such disclosure practices in corporate America (KPMG (2017)). As an example of investors’ interest and activism in disclosure of ESG information by companies serves the appeal of a number of asset management companies (with more than 5 trillion dollars of managed assets) to the SEC in year 2018, in which they advocated for the need of reconsideration of official disclosure regulations towards a comprehensive integration of sustainability-related information into reporting practices of companies listed on US stock exchanges (Ho (2020)). However, a simple disclosure of separate sustainability reports without inclusion of relevant financial information doesn’t meet all the informational needs of investors, since the detachment of financial figures can lead to distortion of the whole business picture and thus contribute to a misjudgment of the role of ESG-driven achievements in the overall performance results of a company (Aluchna et al. (2019)). To avoid possible misinterpretation of advances in sustainable business development that investors may get, companies embraced new forms of delivering relevant information to the stakeholders by integrating ESG-related information into established financial reporting practices, thus paving the way for new “integrated reporting” standards. With annual report being one of the most important communication channels with investors, companies have been incorporating relevant qualitative ESG information into their 10-K1forms along with standard financial and performance figures (KPMG (2017)). While the disclosure of financial information largely constitutes a highly standardized and regulated process, the reporting of sustainable issues still rests essentially upon incentives and willingness of companies’ management to IGNATOV AND RUDOLF 223 bring this information to the stakeholders (Cannon, Ling, Wang, and Watanabe (2020)). Therefore, with enough room to manoeuvre within publication of ESG-related materials, the management can adapt the degree of transparency of sustainability disclosure according to company’s strategic interests or for their personal advantages (McBrayer (2018)). The diversity of communication strategies being at management’s disposal, ranging from purposeful vocabulary and semantic manipulation to limitation of information disclosure2(Fyodorova, Sayakhov, Demin, and Afanasyev (2019)), is largely covered in the research under two fundamental theories of information disclosure, which embrace predominantly opposing set of behavior incentives influencing the managers of exchange-listed companies. While the signalling theory implies that managers of successful companies are encouraged to disseminate as much information as possible to inform their investors of accomplished work and future perspectives and thus increase company’s valuation on the market, the legitimacy theory postulates, in contrary, that predominantly the management of bad performing companies is motivated to cultivate public opinion through extensive reporting and thus improve company’s reputation and smooth out negative impressions of failures and adverse events by investors (Cannon et al. (2020)). The latter behavior of managers is characterized in the literature under the term “corporate impression management”, which describes the divergence between actual activities of a company and their depiction and characterization in communication with various stakeholders, with some papers arguing that such strategies can bring in comparable gains in terms of investors’ satisfaction as in cases with real, unembellished adjustments in company’s actions (Roman, Mocanu, and Hoinaru (2019)). The proponents of the legitimacy theory readily use this argument in determination of the root causes of consistently expanding sustainability disclosure levels of public companies over the past forty years (O’Donovan (2002)). One of the key instruments in the arsenal of managers seeking to impress investors is company’s own annual report, which describes firm’s activities and performance throughout the year and thus constitutes a perfect opportunity for the management to shape company’s image in the eyes of its stakeholders (Roman et al. (2019)). While the annual report serves as a source of key performance indicators and general financial figures, the overwhelming part of it consists of verbal messages prepared by the management, with a high degree of leeway in formulation of the narrative (Ben-Amar and Belgacem (2018); Lo, Ramos, and Rogo (2017)). However, the empirical studies do not provide overwhelming support for the legitimacy theory in the field of sustainability disclosure and reporting practices, recording rather mixed evidence in terms of dependency between the actual ESG performance and the extent of companies’ sustainability disclosure (Nazari, Hrazdil, and Mahmoudian (2017)). Depending on the sample, methodology, as well as the period of time investigated, several researchers document positive relationship between those factors (i.a. Clarkson, Fang, Li, and Richardson (2013); Lys, Naughton, and Wang (2015); Nazari et al. (2017); Plumlee, Brown, Hayes, and Marshall (2015)), whereas other studies (i.a. Cho and Patten (2007); Cho, Guidry, Hageman, and Patten (2012); Muslu, Mutlu, Radhakrishnan, and Tsang (2017)) predominantly find support for the legitimacy theory. The latest research studies in the ESG field indicate the significance of more extensive examination of semantic and linguistic formats in reports prepared by the management (Nazari et al. (2017)). While the analysis of ESG activities and performance of companies is routinely conducted by rating agencies and data providers based on various metrics determined and compiled by them3, the content analysis of companies’ ESG disclosure has long been burdened by a lack of a standardized linguistic toolkit that could enable the researchers to make a highly granular examination of ESG reporting practices on large data sets4(Baier et al. (2020)). However, the recent publication of a comprehensive ESG dictionary, with more than four hundred words and thirty-four subcategories by Baier et al. (2020), is determined to mitigate described difficulties within research on textual analysis and enables to implement established techniques of natural language processing (NLP) in the context of sustainability disclosure and reporting practices on large data samples. This paper aims to re-examine the fundamental theories of information disclosure by public companies in the ESG context, namely the signalling and the legitimacy theories, using an automated processing of annual reports of US companies based on recently proposed dictionary of ESG terms by Baier et al. (2020) and thus provide new evidence in support of one of the theories in an objective and standardized framework on a large data sample. Analyzing more than eleven thousand unique 10-K reports of US companies filed with SEC in period 2013 to 2018, this study examines the relationship between actual ESG performance of companies evaluated by means of MSCI ESG Research 224 IGNATOV AND RUDOLF performance scores5, and the extent and the scope of their sustainability reporting assessed by textual disclosure scores based on a term weighting scheme following Loughran and McDonald (2011). The regression analysis shows support for the signalling theory for the majority of industries according to Standard Industrial Classification (SIC) division structure, whereas companies belonging to the Mining industry division (with rather bad environmental performance) favor extensive reporting of the environment-related topics, which is consistent with the reasoning underlying the legitimacy theory. Besides the examination of the overall disclosure of ESG information, the implemented framework allows to investigate each ESG pillar category separately in terms of the relationship between actual performance levels and the extent of information disclosure. Thus, this study shows evidence for a positive relationship of environmental, social, and governance disclosure scopes with their respective actual ESG pillar performance categories. By differentiating ESG sentences from annual reports between forward-looking and nonforward-looking statements using the forward-looking word list proposed by Li (2010), this study also finds evidence for a better next-year ESG performance of companies that provide more forward-looking ESG information in annual reports in comparison to companies with lower levels of forward-looking ESG disclosure. This paper contributes to the literature in the fields of information disclosure of public companies, ESG performance and reporting, as well as textual and sentiment analyses. Apart from providing empirical evidence on theoretical concepts of information disclosure in the context of sustainability, the framework and results presented in this paper can also find useful applications in the asset and investment management industries, along with the regulatory environment. The inclusion of ESG disclosure models together with standard ESG performance scores from established data providers into investment decision-making could enhance investors’ stance with regards to data availability and reduce boundaries to sustainable investing (Morgan Stanley (2020); Van Duuren, Plantinga, and Scholtens (2016)), especially in investment cases where ESG performance scores are not available for particular companies, thus mitigating the possibility of a selection bias (Baier et al. (2020)). Besides that, the regulatory agencies could enhance their frameworks by paying closer attention to environmentally unfriendly industries, companies in which could tend to polish up relatively bad performance by means of extensive reporting. The remainder of this paper is structured as follows. Section II discusses the theoretical framework this paper is based upon, along with an overview of the related literature. Section III presents methodology of the empirical part of this paper. Results with robustness analysis are presented in Section IV.SectionVdiscusses the results and provides limitations of this study. Finally, Section VI concludes the paper and provides further thoughts on future research. 2THEORETICAL FRAMEWORK AND HYPOTHESES The relationship between actual performance and reporting of related information and thus management’s motivation for disclosure can be generally summarized in the finance literature under two streams of research (Nazari et al. (2017)). While voluntary disclosure assumptions imply that managers disclose as much information as possible for the benefit of the company and its stakeholders, the legitimacy theory alleges that managers adapt their disclosure behavior and the use of language according to their own opportunistic interests. By voluntary informing the shareholders and other stakeholders about company’s activities via extensive reporting, the management contributes to the reduction of information asymmetry and thus agency costs (Shehata (2014); Zhang, Shan, and Chang (2021)), gaining additional trust of investors (Clarkson, Ponn, Richardson, Rudzici, Tsang, and Wang (2020))6and encouraging new investments into the company (Mittelbach-Hörmanseder, Hummel, and Rammerstorfer (2021); Shehata (2014); Verrecchia (1983)). Companies with good performance face lower operational costs of additional information disclosure in comparison with bad-performing firms, which serves as additional stimulus for managers to report more and differentiate themselves from bad-performers (Lopez-de Silanes, McCahery, and Pudschedl (2019)). As a result of increased trust of investors and broadening of the investor base, company could also benefit from lower7cost of capital (Aluchna et al. (2019); Dhaliwal, Li, Tsang, and Yang (2011); Lambert, Leuz, and Verrecchia (2007); Shehata (2014)). In addition, extensive disclosure of information during good times can further strengthen company’s reputation, creating a IGNATOV AND RUDOLF 225 reputational buffer that could serve the company during a crisis8and mitigate the public relations damage from it (e.g., Zhang et al. (2021)). The recent research also shows that higher degree of sustainable information disclosure favourably affects company-specific crash risk (Da Silva (2021)), providing further evidence for the insurance characteristic of information disclosure practices (Darnell (2021)). Altogether, all these factors should bring additional value for a company and thus enhance its market valuation, which is also beneficial for company’s management, especially in case of a performance-based compensation (Brogi and Lagasio (2019); Mittelbach-Hörmanseder et al. (2021)), and should motivate good-performing companies to disclose more information. In contrast to the signalling theory9, the incentives underlying the legitimacy theory of information disclosure imply that the management acts predominantly opportunistic in their own interest10 and uses information disclosure as an instrument for impacting shareholders’ perception of their actual performance (e.g., Cho, Roberts, and Patten (2010); Cho, Mittelbach-Hörmanseder, Hummel, and Matten (2019); Deegan (2002)). Apart from shareholders, company’s management also seeks to influence other stakeholders and general public through tailored disclosure of non-financial narratives in order to restore company’s legitimacy and thus oppose the negative news sentiment (Cho et al. (2019); Hummel and Szekely (2021); Nakao, Kokubu, and Nishitani (2019)). The concept of legitimacy implies that society generally approves the existence and operations of the company within socially admissible limits, thus allowing it to exist and function on a “legit” basis (O’Donovan (2002); Roman et al. (2019)). Thus, managers can adapt the tone and the narrative of disclosed information for opportunistic reasons of legitimatization of company’s poor financial or ESG-related results (Cannon et al. (2020); Nazari et al. (2017)). In order to achieve this effect, the management can use plenty of communication and lexical strategies, such as intentional complication11 of lexical constructions that can impress or confuse investors. Apart from the intentional difficulty of text comprehension, Fyodorova et al. (2019) provides general overview of other communication and reporting strategies at management’s disposal, among them rhetorical manipulation, visual and thematic distortion, biased choice of benchmarks for comparison purposes, as well as deliberate inclusion or exclusion of certain information and performance indicators. Jin, Shi, and Zhang (2019), for example, discuss reporting practices of several Chinese meat producers that omitted critical information in their sustainability reports about the use of chemical substances in meat production and the subsequent investigation by the Chinese authorities in 2010s. Managers can turn to such tactics since stakeholders find it often difficult or are unable to get to the bottom of things through distorted information – an action that also induces additional costs of information processing (Beretta, Demartini, and Trucco (2019); Clatworthy and Jones (2001)). The recent research also shows that elevated reporting of ESG-related topics serves as one of the strategies to repair company’s legitimacy and improve its image in the eyes of stakeholders. Zhang et al. (2021), for example, find that companies focusing on corporate social responsibility (CSR) reporting reduce the value losses associated with disclosure of financial restatements, thus providing evidence for legitimacy function of sustainability disclosure. Roman et al. (2019) argue that such behavior suits worse-performing companies more than their more successful rivals, managers in which may try to take advantage of their information handicap as opposed to investors and other stakeholders and, therefore, make a subjective image of their company that deviates from the objective state of things. Feng and Gao (2020) provide several examples12 of studies that present evidence for more extensive environmental disclosure by companies with rather bad environmental performance, pointing to support of the legitimacy theory in terms of the negative relationship between disclosure and actual performance.13 However, the empirical research also provides contradictory14 evidence in favor of the signalling theory, which implies the opposite (positive) relationship between ESG disclosure and actual performance levels. For example, Ajina and Bacha (2019) show positive connection of annual reports’ readability measures with actual CSR performance of French companies using computational linguistics measures. Hummel and Szekely (2021) provide further evidence from Europe, showing that more sustainable companies from the STOXX Europe 600 index are inclined to dedicate more attention to topics related to sustainable development goals in their annual reports. Beretta et al. (2019)also find evidence in support of the voluntary disclosure based on integrated annual reports of European publiccompanies. In Canada, Ben-Amar and Belgacem (2018) find that management discussion and analysis sections of public companies with better sustainability performance are longer in comparison with companies falling behind. Lopez-de Silanes et al. 226 IGNATOV AND RUDOLF (2019) report cross-country (both the US and international) evidence for positive relationship between ESG scores and disclosure levels by examining companies with equity market values exceeding $0.7 bn. Adding to the US evidence, Nazari et al. (2017) also find positive link between performance and disclosure by focusing on CSR disclosure documents of S&P 50015 companies. In addition, Clarkson et al. (2020) explore CSR disclosure of US companies by implementing linguistics features analysis on their sustainability reports and find evidence for elevated disclosure levels by more sustainability-oriented firms. In contrast to these studies, Patten (2002) documents negative link between disclosure levels and environmental performance while analyzing annual reports of 131 US companies from the Toxic Release Inventory list of 1988. Also adding to the US evidence, Crowley, Huang, Lu, and Luo (2019) find that companies with bad sustainability performance publish more ESG-related messages on Twitter than their better-performing peers, which is consistent with the legitimacy theory of information disclosure. Thus, the empirical research to date has presented mostly mixed data with regard to the evidence in support of the signalling and the legitimacy theories in the realm of ESG disclosure and performance. Such divergence in results can be justified by several factors, including deviations in samples and time periods examined16, differences in study designs and methodologies17, and disagreements in definitions of disclosure metrics and thus the interpretation of findings18. Nevertheless, the mentioned studies all share in common the lack of a comprehensive and unified ESG corporate lexicon in their methodology19 that could be used for determination of ESG disclosure levels in reports issued by companies. In order to close this research gap, Baier et al. (2020) developed a broad, granular dictionary with ESG terms based on a sample of annual reports of companies from the S&P 100 index, providing an opportunity to explore the ESG disclosure of public companies in more detail on a large scale. This paper implements the dictionary from Baier et al. (2020), along with established textual analysis and NLP techniques, to examine the theories of information disclosure in the ESG context and thus to provide empirical evidence using a large sample of companies’ documents. Thus, this paper offers an empirical framework for analysis of the ESG disclosure of companies that can be implemented in a standardized automated way for both practical and academic purposes. Consistent with the legitimacy and the signalling theories, this paper examines the relationship between ESG performance and ESG disclosure extent of companies, which leads to the following hypotheses20: Hypothesis 1: Companies with good ESG performance have high ESG disclosure levels, consistent with the signalling theory and implying a positive relationship between performance and disclosure. Besides the analysis of the overall ESG performance, the framework implemented in this paper allows the analysis to be extended to each category of ESG performance by focusing on environmental, social and governance dimensions of sustainability separately. Thus, the hypotheses with respect to the relation between the ESG performance and disclosure are extended as follows: Hypothesis 2: Companies with good environmental performance have high environmental disclosure levels, consistent with the signalling theory and implying a positive relationship between performance and disclosure. Hypothesis 3: Companies with good social performance have high social disclosure levels, consistent with the signalling theory and implying a positive relationship between performance and disclosure. Hypothesis 4: Companies with good governance performance have high governance disclosure levels, consistent with the signalling theory and implying a positive relationship between performance and disclosure. Some research articles argue that linguistic features of companies’ ESG disclosure can have predictive power for ESG performance21, which provides further motivation for analysis of the relationship between forward-looking ESG information disclosure and the next-period ESG performance. Thus, the following hypothesis is proposed: IGNATOV AND RUDOLF 227 Hypothesis 5: Companies with high ratio of forward-looking ESG information in their disclosure have higher nextyear ESG performance. 3METHODOLOGY AND DATA SAMPLE 3.1 ESG disclosure Companies generally have at their disposal many instruments for communication with investors and other stakeholders, including i.a. press releases, investor presentations and conferences, sustainability reports, and conference calls. However, the annual report remains the major in-depth source of information for markets and the public about the past activities and future developments of a company (Hummel and Szekely (2021); Mittelbach-Hörmanseder et al. (2021)). Besides general financial key figures and performance commentaries, the annual report also serves as the most informative disclosure of ESG-related information about company’s actions and implemented sustainability policies (Baier et al. (2020)). Thus, this study focuses on annual reports as a source of ESG information disclosure of companies. In order to assess the amount and the extent of ESG information disclosed in annual reports, the qualitative data in textual form should be quantified by means of textual analysis techniques from linguistics and computer sciences that have found application in the field of financial research. This study implements an established bag of words, dictionarybased framework with term weighting scheme following Loughran and McDonald (2011). In this way, the text content is transformed into a matrix consisting of vectors of term counts, where the terms represent words from the text that are also included in a pre-specified dictionary22. The weighting scheme ensures that the textual analysis model accounts for differences in relative importance of words in both single text and the entire lexicon used in the sample of documents, while also normalizing the weight of each word according to the document’s length. By using the term document matrix with words’ frequencies and their estimated weights, one can calculate the disclosure score for each annual report, thus quantifying the scope and the extent of information disclosure. Jegadeesh and Wu (2013) formally summarize the model used by Loughran and McDonald (2011)), which is also implemented in this study, as follows: widf j=log N dfj (1) where widf jdenotes the inverse document frequency (idf)weightofawordj,Nis the overall number of annual reports in the study sample, and dfjis the number of reports where term jappears at least once. The idf weighting model ensures that larger weights are not assigned to the words which are used in many documents too often and thus do not have much informational value. Since the weighting scheme is applied to the words from a dictionary, the word weights are adjusted as follows: wtf.idf i,j ={1+log (tfi,j)widf jif tfi,j >0, 0otherwise, (2) where tfi,j denotes the frequency of appearance of term jfrom a dictionary in annual report i,wtf.idf i,j is the weight of a word from the dictionary. Using this weighting scheme, the ESG disclosure score of annual report is calculated as follows: ESGDisclosureScoretf.idf i=1 (1+log ai) J ∑ j=1 wtf.idf i,j (3) 228 IGNATOV AND RUDOLF where ESGDisclosureScoretf.idf iis the ESG disclosure score of annual report ibased on idf weighting scheme, aiis the total count of words in annual report i,andJis the total number of words in the ESG dictionary. Since the focus of this paper lies on the disclosure of ESG-related information in annual reports, the dictionary applied in this framework should consist of sustainability lexicon used by companies in formal23 reports. Baier et al. (2020) recently proposed a comprehensive, manually24 compiled ESG dictionary, using a broad sample of annual reports of companies-members of the S&P 100 index, that offers a broad classification of fourty ESG subcategories in addition to general categorization into environmental, social, and governance pillars of sustainability. With 482 words25, the dictionary allows a precise quantification of broad ESG topics disclosed by companies in their annual reports and thus the assessment of their disclosure efforts in the ESG context. Tables A1, A2,andA3 show all the three ESG dictionary subcategories proposed by Baier et al. (2020). While we rely on a third-party lexicon in our analysis, this ESG dictionary was derived from annual reports of US companies and has been peer-reviewed and recognized by independent researchers, which lessens the chance of a misidentification of ESG terms used by companies in our sample. Hence, the ESG disclosure score serves as an estimate of the extent and the scope of companies’ ESG disclosure based on their annual reports. 3.2 ESG performance In order to juxtapose the ESG disclosure with actual performance of companies and thus examine our hypotheses with regards to the signalling and the legitimacy theories, an assessment of companies’ ESG activities and initiatives is necessary. Since the ESG agenda has taken on great importance in the economy and financial markets in the last decades, the rating agencies and financial data providers have established various methodologies of consistent assessment of companies in terms of their ESG performance, which allowed their ratings to become the benchmark for determination of companies’ sustainability performance in both academia and industry settings (Escrig-Olmedo, Fernández-Izquierdo, Ferrero-Ferrero, Rivera-Lirio, and Munoz-Torres (2019)). Although the definition and especially quantification of sustainability factors are a source of controversy, with different methodologies of performance assessment leading to sometimes completely different outcomes on both the industry and firm levels (Dimson, Marsh, and Staunton (2020), Billio, Costola, Hristova, Latino, and Pelizzon (2021)), the necessity to assess companies’ ESG actions both internally and externally created a competitive environment for various analysis techniques, thus giving a choice to the markets and regulatory agencies to assess the rating agencies and their methodologies in the search of the most comprehensive one. Besides divergent methods of estimation and dissimilarity in the definition of ESG criteria with an imperfect transparency of choices (Kotsantonis and Serafeim (2019)), some ratings can be also affected by conflict of interests with companies they assess due to the nature of other branches of rating agencies’ business, such as consulting and other paid services for the rated companies.26 However, the comparison of various companies in different industries in terms of their ESG performance on a large scale is hardly manageable without involvement of rating agencies and data providers specialized in the ESG area. While there is a plenty of ESG data providers, the vast majority of asset managers and academic research studies refer to the MSCI as their primary source of ESG performance data, which is considered to be the largest vendor of ESG performance metrics for financial markets and academic institutions (Christensen, Serafeim, and Sikochi (2021); Serafeim (2020)). MSCI ESG Research implements a variety of methodologies, ranging from simple assessment of corporate data and news media to advanced analytical techniques from machine learning and artificial intelligence fields, in order to evaluate sustainability performance of companies in environmental, social, and governance areas, taking into account the specifics of all industries27 to compile the overall annual ESG performance score of a company (Pastor, Stambaugh, and Taylor (2021); Glück, Hübel, and Scholz (2021)). Thus, the MSCI ESG Research performance scores are used in this study as an estimator of actual annual ESG performance of companies and are put in comparison to estimated ESG disclosure scores of annual reports to analyze the relationship between the real performance and the level of disclosure in annual reports. IGNATOV AND RUDOLF 235 TABLE 3 ESG performance regressions. (1) (2) (3) (4) (5) (6) (7) (8) (9)†(10)†(11)†(12)†(13)†(14)†(15)†(16)† ESGDisc 0.1927** 0.2281*** −2.0587*** −0.2677 (0.095) (0.059) (0.474) (0.294) EDisc 3.1345*** 0.8768*** −1.0831** −0.2001 (0.176) (0.051) (0.474) (0.176) SDisc 1.7599*** 0.5950*** 0.3433 0.4099 (0.107) (0.043) (0.684) (0.283) GDisc 0.2697** -0.0964** −0.2223 −0.0317 (0.117) (0.044) (0.706) (0.238) ESGPerf(t-1) 6.5954*** 7.5961*** (0.070) (0.297) EPerf(t-1) 5.5304*** 5.4722*** (0.055) (0.327) SPerf(t-1) 5.0568*** 4.5593*** (0.055) (0.353) GPerf(t-1) 3.5886*** 3.4551 (0.034) (0.154) (Continues) 236 IGNATOV AND RUDOLF TABLE 3 (Continued) (1) (2) (3) (4) (5) (6) (7) (8) (9)†(10)†(11)†(12)†(13)†(14)†(15)†(16)† Size 0.1450** 0.0592 −0.0788 −0.0963*** 0.0368 −0.2082 −0.0963 0.2648** (0.05) (0.038) (0.049) (0.035) (0.260) (0.222) (0.159) (0.107) BM −1.3032*** −0.6776*** 0.4453 −0.8523*** 0.1754 −0.7165 0.2059 0.8962 (0.327) (0.221) (0.320) (0.241) (1.101) (0.597) (0.633) (0.645) AssetProd 2.6806 6.4317** −1.2977 3.9240 48.0566 59.7428**−3.3242 −13.7980 (5.801) (3.110) (5.093) (4.122) (33.044) (22.452) (20.965) (20.094) Lev 2.9771 −0.4926 4.2660***−1.0897*** 0.4052*** 0.1612 0.6234***−0.1075** (2.103) (0.529) (0.556) (0.265) (0.117) (0.101) (0.093) (0.042) Observations 11393 11391 11392 11391 8742 8742 8742 8741 400 400 400 400 297 297 297 297 This table presents the regression estimates of MSCI ESG performance scores on 10-K disclosure scores with various control variables. ESGDisc,EDisc,SDisc,andGDisc are the ESG, environmental, social, and governance disclosure scores of annual reports estimated using the Eq. (3) based on the ESG topic dictionary. The regression models with “t” sign denote the regressions run only on the Mining division companies according to the SIC. See Eqs. (5), (11), (12), and (13) for the definition of control variables. All independent variables are standardized to a mean of 0 and a standard deviation of 1. The coefficients’ estimates are not affected by the presence of outliers in the control variables. All regression models are estimated using year dummy variables and standard errors clustered on the firm level. The number of observations changes throughout the models due to the availability of control variables for each regression model. “***” denotes the 1% significance level, “**” the 5%, and “*” the 10% level, respectively. The values in parantheses report the standard errors of estimated coefficients. IGNATOV AND RUDOLF 237 FIGURE 6 ESG pillars performance. The graphs represent annual average Environmental, Social, and Governance MSCI performance scores of companies for each industry division in the sample according to the SIC. The respective pillar scores are weighted according to the MSCI ESG weightings for each pillar category within the overall ESG score. “Communications” category represents the “Transportation, Communications, Electric, Gas and Sanitary service” division. [Color figure can be viewed at wileyonlinelibrary.com] significant at the 1% level and thus confirming initial descriptive observations. To control for other company-specific effects that can influence the level of ESG performance according to the literature47, we include in our regression control variables, thus extending the regression to ESGPerfi=a+b×ESGDisci+c×ESGPerfi,t−1+d×Sizei+e×BMi +f×AssetProdi+g×Levi+𝜀 i(5) where ESGPerfi,t−1is the previous-year MSCI ESG performance score of company i,Sizeiis the natural logarithm of the market capitalization of company at the end of the quarter before annual report filing, BMiis the ratio of the book value of equity to the market value of equity at the end of the quarter before annual report filing, AssetProdiis the ratio of earnings before interest and taxes (EBIT) to the total assets at the end of the quarter before annual report filing, and Leviis the ratio of the market value of equity to the total liabilities of company at the end of the quarter before annual report filing. Model 5 in Table 3also confirms the positive relationship between performance and disclosure observed in Model 1, with coefficient of 0.2281 significant at the 1% level. Furthermore, the regression model indicates positive effect of companies’ size on ESG performance. Having more financial and operational resources, larger firms show significantly higher levels of ESG performance in line with expectations and empirical evidence48.Furthermore, the book-to-market ratio shows significant negative slope coefficient, implying positive relationship between expectations of future growth, which is reflected in higher market valuations and thus lower book-to-market ratios, and ESG performance of companies. 238 IGNATOV AND RUDOLF TABLE 4 ESG industry-adjusted performance regressions. Models (1) (2) (3)†(4)† ESGDisc 0.6738** 0.3220*** −1.1584 −0.2788 (0.338) (0.111) (1.265) (0.431) ESGAdjPerf(t-1) 8.2685*** 8.5037*** (0.063) (0.295) Size 0.6601*** 0.6988 (0.094) (0.434) BM −2.9347*** 1.1753 (0.661) (1.619) AssetProd −1.3735 57.1217 (9.841) (48.203) Lev 1.8179*** 0.3500** (0.574) (0.171) Observations 11393 8742 400 297 This table presents the regression estimates of MSCI ESG industry-adjusted performance scores on 10-K ESG disclosure scores with various control variables. Industry-adjusted ESG scores are ESG performance scores of companies normalized relative to their industry peers. ESGDisc is the ESG disclosure score of annual report estimated using the Eq. (3)basedonthe ESG topic dictionary. The regression models with “t” sign denote the regressions run only on the Mining division companies according to the SIC. See Eqs. (5)and(7) for the definition of control variables. All independent variables are standardized to a mean of 0 and a standard deviation of 1. The coefficients’ estimates are not affected by the presence of outliers in the control variables. All regression models are estimated using year dummy variables and standard errors clustered on the firm level. The number of observations changes throughout the models due to the availability of control variables for each regression model. “***” denotes the 1% significance level, “**” the 5%, and “*” the 10% level, respectively. The values in parantheses report the standard errors of estimated coefficients. We consider the industry effects to check the robustness of the results by using MSCI industry-adjusted ESG performance scores in regressions presented above.49 Thus, Eqs. (4)and(5) are adjusted to ESGAdjPerfi=a+b×ESGDisci+𝜀 i(6) ESGAdjPerfi=a+b×ESGDisci+c×ESGAdjPerfi,t−1+d×Sizei+e×BMi +f×AssetProdi+g×Levi+𝜀 i(7) where ESGAdjPerfiis the MSCI industry-adjusted ESG performance score of company i. Table IV summarizes the regression results for industry-adjusted ESG scores. Models 1 and 2 also show a positive relationship between performance and disclosure in the main sample, with coefficients 0.6738 and 0.3220 significant at the 5% and 1% levels, respectively. The coefficients of size and book-to-market ratio remain significant with unchanged sign of the relationship with ESG performance. Leverage, in turn, has significant positive slope coefficient in both the main and the Mining (Model 4) division samples. As expected, the regressions within the Mining division show negative coefficients of ESG disclosure in both Model 3 and Model 4 of Table 4; however, the coefficients are not statistically significant. To further explore the drivers of the overall relationship between ESG performance and disclosure, we look at singular pillar components of the ESG performance and the respective disclosure levels of companies by IGNATOV AND RUDOLF 239 differentiating between environmental, social, and governance areas of sustainability. Using the MSCI Environmental, Social, and Governance weighted pillar scores and the disclosure scores estimated based on the three main categories of the ESG dictionary, the regressions (4) and (5) are adjusted to EPerfi=a+b×EDisci+𝜀 i(8) SPerfi=a+b×SDisci+𝜀 i(9) GPerfi=a+b×GDisci+𝜀 i(10) where EPerfi,SPerfi,andGPerfiare the Environmental, Social, and Governance MSCI weighted performance scores of company iin the year in which the annual report is released, and EDisci,SDisci,andGDisciare the estimated environmental, social, and governance disclosure scores of the annual report. Models 2, 3, and 4 in Table 3show the results of the regressions run on the main sample, with all coefficients being significantly positive and implying a positive relationship between performance and disclosure levels in respective sustainability topics. The estimation results in the Mining division sample show significant negative coefficient by the environmental disclosure score, consistent with the descriptive evidence. To control for firm-specific effects, we include the control variables and extend the regressions to EPerfi=a+b×EDisci+c×EPerfi,t−1+d×Sizei+e×BMi +f×AssetProdi+g×Levi+𝜀 i(11) SPerfi=a+b×SDisci+c×SPerfi,t−1+d×Sizei+e×BMi +f×AssetProdi+g×Levi+𝜀 i(12) GPerfi=a+b×GDisci+c×GPerfi,t−1+d×Sizei+e×BMi +f×AssetProdi+g×Levi+𝜀 i(13) where EPerfi,t−1,SPerfi,t−1,andGPerfi,t−1are the previous-year Environmental, Social, and Governance MSCI weighted performance scores of company i. The regression estimates in Models (6), (7), and (8) confirm the results obtained without consideration of control variables, with significantly positive coefficients for environmental and social disclosure scores providing evidence for a positive relationship between performance and disclosure. The coefficient for the governance disclosure remains significant, however, with a changed sign which is caused by the influence of previousyear governance performance score.50 Furthermore, the negative coefficient of the environmental disclosure score in the Mining division loses its significance, probably due to a smaller sample size caused by inclusion of control variables. In the last step, we analyze the relationship between forward-looking ESG statements and the next-year sustainability performance of companies. Using both the forward-looking and the ESG dictionaries, the proportion of forward-looking ESG statements is estimated for each annual report in the entire sample, including the Mining industry since there is no evidence in descriptive analysis for elevated levels of future-related disclosure by companies belonging to this division. The impact of future-related ESG information on performance is examined using the 240 IGNATOV AND RUDOLF regressions51 ESGPerfi,t+1=a+b×ESGForwDisci+𝜀 i(14) ESGPerfi,t+1=a+b×ESGForwDisci+c×ESGPerfi,t−1+d×Sizei+e×BMi +f×AssetProdi+g×Levi+𝜀 i(15) where ESGPerfi,t+1is the next-year MSCI ESG performance score of company ifollowing annual report filing with SEC, and ESGForwDisciis the ratio of forward-looking to non-forward-looking ESG statements contained in annual report of company i. In addition, to control for industry effects and check the robustness of results, we use the industry-adjusted MSCI ESG performance scores in the regressions ESGAdjPerfi,t+1=a+b×ESGForwDisci+𝜀 i(16) ESGAdjPerfi,t+1=a+b×ESGForwDisci+c×ESGAdjPerfi,t−1+d×Sizei+e×BMi +f×AssetProdi+g×Levi+𝜀 i(17) where ESGAdjPerfi,t+1is the next-year industry-adjusted MSCI ESG performance score of company ifollowing annual report filing with SEC. Table 5reports the estimation results of regressions. Both (14) and (15) regression estimations presented in Models 1 and 2 in Table 5show significantly positive coefficients of forward-looking ESG disclosure, implying that companies with a higher ratio of forward-looking statements in annual reports show better next-year ESG performance. The coefficient of ESG forward-looking disclosure of 0.2309 in Model 2 is significant at the 1% level. The results of regression (17) shown in Model 4 with industry-adjusted performance scores confirm the robustness of results, with ESG forward-looking disclosure coefficient of 0.3429 significant at the 10% level. While the slope coefficients of size and book-to-market ratio show results similar to the regressions with same-year ESG scores, the leverage variable shows significant negative coefficients in both Model 2 (−0.4584, 5% level) and Model 4 (−1.3540, 1% level) of Table 5, implying a negative relationship between the level of corporate debt burden and the next-year ESG performance. This outcome implies that heavily indebted companies can face difficulties in directing enough resources to sustainability-related projects, which in turn hurts their yearly improvement in ESG performance. 5DISCUSSION AND LIMITATIONS The results of regressions (4), (5), (6), and (7) provide evidence for a positive relationship between the extent of ESG disclosure and the level of actual ESG performance, supporting the Hypothesis 1 and thus the signalling theory. This result contributes to the stream of literature that finds empirical evidence for a signalling behavior of companies with good ESG performance, thus supporting the view that the management of good companies tends to pursue extensive high-quality disclosure in annual reports to inform the stakeholders about their real sustainability-oriented activities and achievements. However, the examination of the Mining division sample also provides contrary evidence against the Hypothesis 1, implying a negative relationship between the ESG performance and the extent of ESG reporting in line with the legitimacy theory. The analysis shows that this relation is mostly driven by the environmental pillar of sustainability, consistent with other research studies on environmentally sensitive sectors. Roman et al. (2019)and Feng and Gao (2020) argue that language of companies belonging to the environmentally unfriendly industries can IGNATOV AND RUDOLF 241 TABLE 5 ESG next-year performance regressions. Models (1) (2) (3)‡(4)‡ ESGForwDisc 0.2921*** 0.2309*** 0.6281*** 0.3429* (0.095) (0.086) (0.207) (0.176) ESGPerf(t-1) 5.3378*** (0.078) ESGAdjPerf(t-1) 7.1884*** (0.082) Size 0.2458*** 1.1453*** (0.066) (0.135) BM −1.6134*** −4.4675*** (0.425) (0.813) AssetProd 0.2980 6.7365 (8.979) (18.563) Lev −0.4584** −1.3540*** (0.177) (0.162) Observations 9044 6611 9044 6611 This table presents the regression estimates of the next-year (t+1) MSCI ESG performance scores on 10-K ESG forwardlooking disclosure scores with various control variables. ESGForwDisc is the ratio of forward-looking to non-forward-looking ESG sentences contained in annual reports based on the ESG and the forward-looking topic dictionaries. The regression models with “+” sign denote the regressions run with MSCI industry-adjusted ESG performance scores. Industry-adjusted ESG scores are ESG performance scores of companies normalized relative to their industry peers. See Eqs. (5)and(7)for the definition of control variables. All independent variables are standardized to a mean of 0 and a standard deviation of 1. The coefficients’ estimates are not affected by the presence of outliers in the control variables. All regression models are estimated using year dummy variables and heteroskedasticity-robust (MacKinnon and White (1985)) standard errors. The number of observations changes throughout the models due to the availability of control variables for each regression model. “***” denotes the 1% significance level, “**” the 5%, and “*” the 10% level, respectively. The values in parantheses report the standard errors of estimated coefficients. be biased towards manipulation because of management’s incentives; however, other researchers emphasize external public pressure on companies to release more information as the main reason for such reporting practices (Clarkson, Overell, and Chapple (2011); Hummel and Szekely (2021)).52 While the Mining division is a clear outlier in our sample and thus this result cannot be generalized and applied to companies from other industries, the investors and regulatory agencies should still pay a close attention to information disclosure in environmentally harmful industries, companies in which have incentives to polish up relatively bad performance with extensive reporting. Overall, our results predominantly show the support for the signalling theory, confirming the Hypothesis 1. With regards to the singular pillar analysis of sustainability reporting practices, the regressions (8) to (13) also provide evidence for a positive relationship between actual performance and the extent of disclosure in pillar categories, thus supporting Hypotheses 2, 3, and 4 and hence also the signalling theory. The relationship between environmental disclosure and performance in the Mining division sample deviates from the overall results, showing a negative direction. Thus, the results of the ESG pillars analysis also contribute to the signalling theory view of ESG disclosure practices. The regression analysis of forward-looking ESG statements contained in annual reports, implemented using the regressions (14) to (17), shows significant positive relationship between future-related ESG information and the next-year ESG performance, providing evidence in support of Hypothesis 5. This result serves as further evidence for the importance of linguistic features 242 IGNATOV AND RUDOLF as one of the factors explaining future ESG performance of companies, adding to the recent literature that explores the prediction features of ESG performance (Clarkson et al. (2020)). Nevertheless, this study is also subject to several limitations. We limit our data sample by the year end of 2018 due to the availability of ESG performance data conditioned on the licensing agreement with MSCI ESG Research. In addition, our research focuses on annual reports of companies as the only source of ESG disclosure, while further application of NLP analysis techniques on other documents/sources of ESG information could complete the picture of the overall disclosure practices of companies. Furthermore, the study includes only public US companies with annual reports accessible through the SEC’s EDGAR database, while non-public US firms and other regions of the world are not the focus of this study. The European companies could be the object of particular interest for future studies, since the ESG disclosure in Europe became a higher priority for the economy in recent years in comparison to North America due to political discussions and subsequent regulatory policies, with increasing commitment of companies to disclose ESG-related information (Hummel and Szekely (2021); Nazari et al. (2017)). Also, the ESG disclosure metrics used in this study are constrained by the implemented ESG dictionary, with estimators of annual reports’ disclosure scores being dependent on the composition of word lists and sensitive to the sample of documents used. Besides that, we rely on a third-party sustainability lexicon which could be biased by subjective opinions of authors that derived the ESG terms from other annual reports of US companies, while the implementation of other lexicons could lead to results different from this paper. Last but not least, the use of ESG performance scores from other data providers could serve as validation of empirical results obtained in this study; however, the comparison of results can be complicated by limited transparency with regards to a detailed methodology of ESG performance quantification and a low correlation of ESG scores between different vendors, while the choice of ratings should also account for potential conflicts of interest between data providers and rated companies. 6CONCLUSION This study sheds light on ESG information disclosure behavior of US companies based on annual reports filed with the SEC using NLP techniques and the recently proposed ESG topic dictionary by Baier et al. (2020). The empirical results obtained on a sample of more than eleven thousand unique 10-K reports from 2013 to 2018 show support for the signalling theory of sustainability-related information disclosure, except for the environmentally unfriendly companies belonging to the Mining division that have incentives to improve their public image through extensive reporting and thus legitimize their operations. Furthermore, the scope of forward-looking ESG information contained in annual reports showed positive relation to the next-year ESG performance, while companies from the Mining division do not exhibit higher levels of future-related sustainability information disclosure in comparison to other industries. The results of this study could provide useful input for investment and regulatory decisions in terms of the enhancement of current ESG performance assessment of companies and tailoring of regulatory disclosure requirements to specific industries. The proposed framework could also serve as an indicator of ESG performance of companies not covered by rating agencies and data providers, thus enhancing sustainability-focused investment strategies and portfolio allocation. Thisstudyoffersawide rangeoftopicsforfutureresearchopportunities,includingthevalidationofresultsonother ESG disclosure sources and geographical regions, with focus on divergences between different types of economies (Cho et al. (2019)) and companies, such as business-to-consumer vs business-to-business types (Hummel and Szekely (2021)). Furthermore, it would be interesting to analyse the effect of the gender composition of executive boards on ESG disclosure (Tapver (2019)) as well as the impact of CEO turnover on changes in reporting practices of companies (McBrayer (2018)), with particular attention to the role of short-term incentives for the management (Lin, Wei, Yang, and Zhang (2021)). Last but not least, the extension of ESG performance frameworks that do not take into account linguistic features by the assessment of companies with approach implemented in this study could provide new evidence for already existing studies (Clarkson et al. (2020)). IGNATOV AND RUDOLF 243 ACKNOWLEDGEMENTS We thank Dr. Alisse Brühne, Dr. Marc-Gregor Czaja, Prof. Dr. Mei Wang, and Dr. Harm Schütt for their helpful comments. We also thank Dorottya Bruszt, Stefan Hüttermann, Leonid Potok, and MSCI ESG Research team for the assistance with MSCI ESG scores data. DATA AVAILABILITY The data used in this paper can be provided upon request, except for MSCI ESG ratings which are subject to the licensing agreement with MSCI ESG Research UK Limited. DECLARATIONS OF INTEREST This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. ORCID Konstantin Ignatov https://orcid.org/0000-0003-3992-8028 ENDNOTES 1We use the terms “annual report” and “10-K” interchangeably throughout this article. 2While managers actively use different communication strategies to sway stakeholders’ opinion, they do not generally include illegal deceptive practices of information distortion (Fyodorova et al. (2019)), which can put managers at legal risks of prosecution as in cases of, e.g., Enron Corporation, Volkswagen AG, and Wirecard AG. 3Clarkson et al. (2020) argue that some providers (e.g., ASSET4) do not completely take into account the linguistic characteristics of companies’ reports that have predictive power for future ESG performance, thus providing further motivation for development of linguistic analysis frameworks. 4Baier et al. (2020) also argue that without an appropriate framework that could be used in a standardized procedure on a large number of documents, the researchers’ work is largely bounded by subjectivity and small sample sizes because of manual processing of reports and opinion-based classification of textual tone and linguistic structures. 5The data are used under the license agreement with MSCI ESG Research UK Limited. 6See also Barnett and Solomon (2012), Da Costa, Liu, Rosa, and Tiras (2020), and Sharma and Vredenburg (1998) for further details. 7Yet, some researchers argue that particular kinds of information disclosure can lead to a contrary result (Botosan (2006); Shehata (2014)). 8For example, environmental or social crises (Darnell (2021); Godfrey, Merrill, and Hansen (2009); Sharfman and Fernando (2008)). 9While in some articles (see, e.g., Shehata (2014)) the voluntary disclosure practices are subdivided into several theories, such as i.a. agency, signalling, capital need theories, according to a precise aim the good-performing managers strive for via extensive disclosure practices, this paper groups these categories under the signalling theory concept, since the focus of the paper lies on the general distinction between incentives behind the reporting of good- and bad-performing companies and not on the specific incentives of good-performing managers. 10 Deegan (2002) also emphasizes fundamental differences in motivation of managers who aim to restore the legitimacy and public image, and those who act responsibly and disclose information necessary for stakeholders. 11 However, in some cases the text complexity can be associated with the complexity of the indus-try a company operates in, which forces the management to explain complex business environment and operations and thus unintentionally increase the content difficulty (Loughran and McDonald (2016)). 12 See, e.g., Cho and Patten (2007), Cho et al. (2010), Hughes, Sander, and Reier (2000), Hughes, Anderson, and Golden (2001), Patten (2002). 13 However, some papers (see, e.g., Hummel and Szekely (2021)) argue that such dependencies can be explained by “environmental-related public pressure” that urges environmentally unfriendly companies to report more about sustainability issues. Nevertheless, in both cases the management of a company tries to win back or improve its legitimacy status in the society and the shareholders’ assessment of the company. 14 Nazari et al. (2017) p. 167 and Hummel and Szekely (2021) provide comprehensive overview of empirical studies focusing on the connection between ESG performance and disclosure levels. 15 See, e.g., Dawkins and Fraas (2011) for further evidence that includes companies from the S&P 500 index. 244 IGNATOV AND RUDOLF 16 See, e.g., Patten (2002), Nazari et al. (2017), Clarkson et al. (2020). 17 See, e.g., Feng and Gao (2020), Clarkson et al. (2020), Crowley et al. (2019). 18 See, e.g., Ben-Amar and Belgacem (2018), Cho et al. (2019), Lopez-de Silanes et al. (2019). 19 Some studies, e.g., Nazari et al. (2017), use broadly defined metrics of disclosure (such as general word count) in the ESG context without focusing on an ESG-tailored lexicon. 20 To avoid the inclusion of competing hypotheses, we formulate our hypotheses with accordance to the signalling theory based on a slightly greater weight of evidence found in the literature. 21 See, e.g., Clarkson et al. (2020) and Qian and Schaltegger (2017). 22 Depending on the research objectives and the thematic focus of a study, the choice of a dic-tionary can range between general lexicons, such as the Harvard Dictionary, and domain-specific word lists, such as the one proposed by Loughran and McDonald (2011) for finance and accounting studies. 23 Loughran and McDonald (2020) and Renault (2017) discuss the importance of domain-specific lexicons in application of dictionary-based methods of textual analysis, which are tailored to the type of the disclosure source, specifically emphasizing the difference between formal language used in official reports, such as an annual report, and, for example, informal language in online communication via social media. 24 Loughran and McDonald (2020) argue that manual preparation of dictionaries has advantages over procedures relying on machine learning algorithms since they are based on idiosyncratic characteristics of the sample used, which leads to a bias of “pseudo-dummy variables” in created dictionaries. 25 See the Appendix Afor a complete list of ESG subcategories and words in the dictionary proposed by Baier et al. (2020)and implemented in this study. 26 “Wall Street’s Green Push Exposes New Conflicts of Interest”, The Wall Street Journal, 29.01.2022, (https://www.wsj.com/ articles/wall-streets-green-push-exposes-new-conflicts-of-interest-11643452202). 27 Especially with regards to varying importance of individual factors within each sustainability category (environmental, social, and governance pillars) for different industrial settings. 28 Reproduced by the permission of MSCI ESG Research LLC, @2020 MSCI ESG Research LLC All rights reserved. The ESG data contained herein is the property of MSCI ESG Research LLC (ESG). ESG, its affiliates and information providers make no warranties with respect to any such data. The ESG data contained herein is used under license and may not be further used, distributed or disseminated without the express written consent of ESG. 29 Companies with country domicile in the United States of America. 30 See, e.g., Hummel and Szekely (2021), Morgan Stanley (2020). 31 See Escrig-Olmedo et al. (2019) for further details. 32 The December 2018 sample limitation is due to the ESG data licensing agreement with MSCI ESG Research LLC. 33 See, e.g., Fama (1991). 34 Electronic Data Gathering, Analysis, and Retrieval System; see https://www.sec.gov/edgar/ about for further information. 35 For additional information on the MSCI ESG Research methodology, see https://www. msci.com/documents/1296102/21901542/MSCI+ESG+Ratings+Methodology++Exec+Summary+Nov+2020. pdf. 36 See, e.g., Serafeim (2020) and Pastor et al. (2021). 37 The data from Compustat and CRSP databases are obtained through Wharton Research Data Services. 38 MSCI coverage of companies has increased throughout the observation period from 2013 to 2018. 39 We control for size, book-to-market, asset productivity, and leverage factors in our regressions. 40 We use 2of12inf English dictionary following Jegadeesh and Wu (2013). 41 For example, ESG dictionary words “discriminate”, “discriminated”, “discriminating”, and “discrimination” are transformed to the stem form “discrimin”, leading to a reduction of the number of dictionary words from 482 to 299. 42 According to the SIC, the Mining division includes companies operating in metal mining, coal mining, oil and gas extraction, mining and quarrying of nonmetallic minerals areas. 43 The forward-looking dictionary consists of the following words: “will”, “should”, “can”, “could”, “may”, “might”, “expect”, “anticipate”, “believe”, “plan”, “hope”, “intend”, “seek”, “project”, “forecast”, “objective”, “goal”. See Li (2010) for further information regarding the dictionary compilation. 44 We compute average performance scores for each ESG pillar based on their weightings that are used in calculation of the overall MSCI ESG score, because the importance of certain factors/issues in each industry varies throughout the observation period, which is, in turn, captured by the weightings. 45 We control for the time effect using dummy variables for each year and cluster the errors by firm to eliminate the firm effect inthesample(seePetersen(2009); Loughran and McDonald (2011)). 46 Mining division represents only 3.4% of the whole sample. 47 See, e.g., Ben-Amar and Belgacem (2018), Drempetic, Klein, and Zwergel (2020), Li (2008), and Roman et al. (2019)fora detailed reasoning regarding the choice of control variables. 48 See, e.g., Drempetic et al. (2020). IGNATOV AND RUDOLF 251 TABLE A3 (Continued) Topic Category Subcategory succession, tenure, vacancies, vacancy Remuneration: appreciation, award, awarded, awarding, awards, bonus, bonuses, cd, compensate, compensated, compensates, compensating, compensation, eip, iso, isos, payout, payouts, pension, prsu, prsus, recoupment, remuneration, reward, rewarding, rewards, rsu, rsus, salaries, salary, severance, vest, vested, vesting, vests Topic Category Subcategory Shareholder rights: ballot, ballots, cast, consent, elect, elected, electing, election, elections, elects, nominate, nominated, plurality, proponent, proponents, proposal, proposals, proxies, quorum, vote, voted, votes, voting Transparency: brother, clicking, conflict, conflicts, family, grandchildren, grandparent, grandparents, inform, insider, insiders, inspector, inspectors, interlocks, nephews, nieces, posting, relatives, siblings, sister, son, spousal, spouse, spouses, stepchildren, stepparents, transparency, transparent, visit, visiting, visits, webpage, website Talent: attract, attracting, attracts, incentive, incentives, interview, interviews, motivate, motivated, motivates, motivating, motivation, recruit, recruiting, recruitment, (Continues) 252 IGNATOV AND RUDOLF TABLE A3 (Continued) Topic Category Subcategory retain, retainer, retainers, retaining, retention, talent, talented, talents Business ethics: cobc, ethic, ethical, ethically, ethics, honesty Bribery and corruption: bribery, corrupt, corruption, crimes, embezzlement Political influence: grassroots, influence, influences, influencing, lobbied, lobbies, lobby, lobbying, lobbyist, lobbyists Whistle-blowing system: whistleblower Topic Category Subcategory Sustainability management and reporting: announce, Disclosure and reporting: asc, disclose, disclosed, announced, announcement, announcements, announces, discloses, disclosing, disclosure, disclosures, fasb, gaap, announcing, communicate, communicated, objectivity, press, sarbanes communicates, communicating, erm, fairly, integrity, Stakeholder engagement: engagement, engagements, liaison, presentation, presentations, sustainable feedback, hotline, investor, invite, invited, mail, mailed, mailing, mailings, notice, relations, stakeholder, stakeholders UNGC compliance:compact,ungc AUTHOR BIOGRAPHIES Konstantin Ignatov is research assistant and PhD student at Allianz Endowed Chair of Finance at WHU - Otto Beisheim School of Management. Professor Markus Rudolf is full Allianz Professor of Finance at WHU - Otto Beisheim School of Management since 1998. He also heads WHU’s Center of Asset and Wealth Management. He earned his PhD degree in fall 1994 and his habilitation degree in May 1999, both at the University St. Gallen, Switzerland. His recent publications focus on the Euro and sovereign risk crises, on Banking, on asset and risk management, and on derivatives pricing. He is coeditor of the journal "Financial Markets and Portfolio Management". He was deputy dean from 2009 to 2014 and dean of WHU from January 2015 to April 2023.