The effect of carbon dissemination on cost of equity

Bus Strat Env. 2019;28:1179–1198. Abstract This study examines whether firms can influence their cost of equity (COE) by broadly disseminating their carbon information over Twitter. We study firms' dissemination decisions of carbon information bydeveloping a comprehensivemeasure of carbon information that a firmmakes onTwitter, referred to as iCarbon. Using a sample of 1,737 firm‐ year observations for 584 nonfinancial firmswith aTwitter account and listed on theU.S. NASDAQ stock exchange over the period 2009–2015, we find that iCarbon is significantly and negatively associated with COE. Our results are consistent after determining the effect of Bloomberg's environmental and environmental, social, and governance disclosure. The findings also hold when using alternative measures of COE and iCarbon.

the way firms communicate important information to stakeholders" (Lee, Hutton, & Shu, 2015, p. 368) and can provide positive signals to market participants about a firm's environmental responsibility to respond to the uncertainty of carbon risks and to improve the firm's reputation and image (Barnett & Salomon, 2012). Twitter's design of short messages (tweets) may allow many firms to gain legitimacy among stakeholders and avoid scrutiny by demonstrating that they are environmentally responsible organisations (see Stanny, 2013).
Twitter also allows firms to know the size of their audience and the number of their followers, which may motivate their decision to disseminate to a broader audience, in a much more timely and efficient manner than a corporate website can achieve. Firms can share their news and discuss their performance through the use of a hashtag (#CarbonEmissions or #ClimateChange) to spread their messages to stakeholders who are concerned about global warming issues and threats and to attract the attention of these stakeholders. By retweeting, the recipients of carbon-related tweets can share this information with their followers to expand the information reach to a more diverse audience and to more potential investors. In essence, using Twitter allows firms to reach potential investors directly and prolongedly in a timely manner that can reduce the time, effort, and energy that investors need to spend on finding, searching for, and accessing information (Blankespoor, 2018;Miller & Skinner, 2015).
Our paper makes several contributions to the extant literature.
First, although the extant research (e.g., Balvers, Du, & Zhao, 2017;Chen & Gao, 2011;Gupta, 2018;Jung, Herbohn, & Clarkson, 2018;Kim, An, & Kim, 2015;Lee, Park, & Klassen, 2015;Li, Liu, Tang, & Xiong, 2017;Peng, Sun, & Luo, 2015;Sharfman & Fernando, 2008;Zhou, Zhang, Wen, Zeng, & Chen, 2018) focused on temperature shocks, managing climate/environmental risks and responding to the Carbon Disclosure Project (CDP) survey to examine market responses to firms' voluntary climate change information disclosure or their associations with the cost of debt financing/equity capital, this paper examines the dissemination effect of carbon-related information via Twitter (iCarbon) on the COE. This broader effect is unlike that of disclosure and has its own capital market consequences (Bushee et al., 2010). Corporate disclosures also "often reach only a portion of investors, which results in information asymmetry among investors" (Blankespoor et al., 2014, p. 79). Second, the prior research (Bushee et al., 2010;Li, Ramesh, & Shen, 2011) has paid particular attention to press releases, as an information intermediary, to examine the effect of dissemination on information asymmetry. The press, however, is biased towards the coverage of highly visible firms and often modifies the information released by firms by adding a discussion, providing opinions, and/or summarising the news (Blankespoor et al., 2014). In contrast, tweets disseminated by firms are short and independent of media adjustments, which make them most likely to be used for disseminating purposes rather than for providing comprehensive information.
Prior work shows how firms' dissemination on Twitter improves market liquidity (Blankespoor et al., 2014;Prokofieva, 2015) and attenuates negative market reaction to product recalls (Lee, Hutton, & Shu, 2015) and acquisition announcements (Mazboudi & Khalil, 2017), to the best of our knowledge, no study has examined the effect of the Twitter dissemination of carbon-specific information on the COE.
We employ a sample of 1,737 observations, representing 584 nonfinancial firms with Twitter accounts, listed on the NASDAQ stock exchange for the period 2009-2015. We use the implied COE, which is based on the average of four estimates, as a proxy for the COE, and the number of tweets that relate to carbon information 1 as a proxy for iCarbon. Our findings show that the better dissemination of carbon information reduces a firm's equity financing costs. We also examine the effect of firms' environmental disclosure, using a scoring level, on the association between iCarbon and the COE. Our results report no effect of environmental reporting, whereas iCarbon is negatively related to the COE. Consistently, we find similar results by examining the effect of environmental, social, and governance (ESG) disclosure. Overall, our findings support the legitimacy theory and indicate that firms that voluntarily disseminate more carbon-related information have a lower COE. The results are robust for the alternative specifications of the model. The organisation of the paper is as follows: The literature and hypothesis development are reviewed in Section 2. The data and methodology are presented in Section 3. Section 4 presents the results and discusses the key findings. Section 5 concludes.

| LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT
Organisations operate in a social, political, and economic context (Buhr, 1998) and have obligations to society in general that go beyond their interests and legal responsibilities. As a part of the modern project of justice and progress, organisations establish their legitimacy based on society's perception of their contribution to the public good (Brunsson, 2006). The relationship between organisations and society, then, is viewed as a "social contract" in which their continuing existence relies upon adapting to the social norms, values, and expectations of organisations and their activities. Such a strategy is essential to obtain and preserve social approval or a licence to operate (Schepers, 2010), that is, legitimacy 2 by changing the societal perceptions of social constituencies (Buhr, 1998;Guthrie & Parker, 1989;Oliver, 1996;Patten, 1992;1 We focus on carbon information because U.S. firms that emit at least 25,000 metric tons of CO 2 are mandated to report their emissions, but not on Twitter, which allows us to differentiate the effect of dissemination from that of disclosure decisions. 2 The legitimacy concept is "rooted in neo-institutional social theory…and has branched out from sociology and is commonly used within legal scholarship that examines the connections among legal frameworks, social norms and decision making" (Bowen, 2014, p. 59). Parsons (1960 viewed legitimacy in organisational institutionalism as the sharing of common values between the organisation and the social system in which it exists. Among other institutional theorists, Suchman (1995) provided an in-depth analysis of organisational legitimacy and referred to it as "a generalised perception or assumption that the actions of an entity are desirable, proper, or appropriate within some socially constructed system of norms, values, beliefs, and definitions" (p. 574). Scherer, Palazzo, & Seidl, 2013). As Dowling and Pfeffer (1975) state, "Organisations seek to establish congruence between the social values associated with or implied by their activities and the norms of acceptable behaviour in the larger social system of which they are a part" (p. 122). Undoubtedly, "organisations that … lack acceptable legitimated accounts of their activities … are more vulnerable to claims that they are negligent, irrational or unnecessary" (Meyer & Rowan, 2004, p. 50, cited in Suchman, 1995. We use legitimacy theory (e.g., Adams, Hill, & Roberts, 1998;Campbell, 2000;Deegan & Gordon, 1996;Garriga & Melé, 2004;Guthrie & Parker, 1989;Suchman, 1995;Zhao, 2012) as a positive theory that embraces a system-oriented perspective, which is derived from political economy theory 3 (e.g., Deegan, 2014;Williams, 1999;Woodward, Edwards, & Birkin, 2001), to explain why firms disseminate carbon-related information via Twitter. Much of the prior research drawing on legitimacy theory to explain or predict particular managerial activities claims that environmental disclosures to communicate with society, on whom an organisation depends for its viability, are necessary to gain legitimacy among stakeholders (Deegan, 2014;Zeng, Xu, Yin, & Tam, 2012), improve stakeholders' perceptions of a firm's environmental efforts (Cho & Patten, 2007;Plumlee, Brown, Hayes, & Marshall, 2015), mitigate stock market risk (Bansal & Clelland, 2004;Orlitzky & Benjamin, 2001;Salama, Anderson, & Toms, 2011), reduce the COE capital (Dhaliwal et al., 2011;El Ghoul, Guedhami, Kwok, & Mishra, 2011), improve financial performance (Clarkson, Li, Richardson, & Vasvari, 2011;Margolis & Walsh, 2003;Orlitzky, Schmidt, & Rynes, 2003), and lessen exposures to political and public pressures (Cho, Freedman, & Patten, 2012). Lindblom (1994) identifies four possible paths to legitimisation to respond to such public pressure. The first path is to inform the relevant public about actual changes in activities or intentions to improve performance. The second path is to attempt to alter stakeholders' perceptions of negative events without making any changes to those actions. The third path is to distract attention away from the threatening events by emphasising more positive actions that do not necessarily have to be related. The fourth path is to attempt to influence society's expectations with regard to performance.
It is also pertinent to note that legitimacy is "a multidimensional concept" (Álvarez-García, Maldonado-Erazo, & del Río, 2018, p. 72), which, according to Suchman (1995), is composed of three dimensions that co-exist in most real-world settings: pragmatic, moral, and cognitive legitimacy. Pragmatic legitimacy (Suchman, 1995) emphasises the selfinterested calculations of the particular interests of an organisation's most immediate social actors, through exchange, influence, or disposition. Pragmatic legitimacy occurs when the legitimacy granter fulfils his/her interests, achieving a value contribution, while acquiring specific commitments with the legitimacy seeker (Díez-de-Castro, Peris-Ortiz, & Díez-Martín, 2018). The second dimension of legitimacy is moral (ethical) legitimacy, which "becomes the decisive source of societal acceptance for corporations in an increasing number of situations" (Palazzo & Scherer, 2006, p. 74). Stakeholder pressure reflects moral legitimacy (Salmi, 2008), which rests on judgements about whether a given activity is the right thing to do to promote the social welfare of the actors that surround the organisation, rather than on judgements about whether the evaluated objective benefits a particular set of constituents (Suchman, 1995). Therefore, "moral legitimacy should be achievable by claiming to be ethical and acting accordingly" (Treviño, den Nieuwenboer, Kreiner, & Bishop, 2014, p. 200). An organisation is evaluated as legitimate from a moral point of view when audiences perceive that it defends and pursues principles accepted and valued as socially positive, which are considered more important than private interests by such an organisation (Díez-de-Castro et al., 2018;Miranda, Cruz-Suarez, & Prado-Román, 2018). Maintaining this legitimacy notion leads to competitive advantages, such as enhanced reputation (Schepers, 2010), which emphasises the economic benefits to organisations of being different (Bowen, 2014). Moral legitimacy is usually analysed by evaluating the appropriateness or desirability of the outcomes (consequential legitimacy), procedures (procedural legitimacy), structures (structural legitimacy), and leaders (personal legitimacy) used to achieve the objectives. Unlike moral legitimacy, cognitive legitimacy is established when the techniques and procedures used to achieve an organisation's objectives are perceived to be adequate and accepted without question (Iglesias-Pérez, Blanco-González, & Navalón, 2018;Salmi, 2008). Cognitive legitimacy accentuates that an organisation is granted legitimacy when audiences see its activities as fitting into their beliefs and assumptions or when they cannot imagine that an organisation would not be corresponding to their interests (Treviño et al., 2014).
Cognitive legitimacy, therefore, represents a state of "comprehensibility" or a "taken-for-granted" (inevitability or permanence; Palazzo & Scherer, 2006;Schepers, 2010) and operates at the subconscious level, making it difficult for the organisation to directly and strategically influence perceptions (Suchman, 1995).
Using legitimacy theory as an interpretive lens, Patten (1992) examined the change in the environmental disclosures of annual reports by 21 North American petroleum companies in response to the increased environmental concern resulting from the 1989 Alaskan Exxon Valdez oil spill incident. He argued that if the Alaskan oil spill resulted in a threat to the legitimacy of petroleum firms and not just to Exxon, then legitimacy theory would suggest that companies operating within the petroleum industry would respond by increasing environmental disclosures in their annual reports. Patten's results show that there was a significant increase in the environmental disclosures made by the companies across the petroleum industry for the post-1989 period, even though the incident itself was directly related to one petroleum company. Patten suggested that threats to a firm's legitimacy entice it to include environmental information in its annual report. Deegan and Rankin (1996) also utilised legitimacy theory to explore how organisations altered their environmental reporting practices in their annual reports around the time of environmental prosecutions. The sample consisted of 20 Australian companies, which 3 According to Gray, Owen, and Dams (1996), stakeholder theory and legitimacy theory are both derived from a broader theory that has been called political economy theory. The political economy is "the social, political, and economic framework within which human life takes place" (Gray et al., 1996, p. 47). The viewpoint included is that society, politics, and economics are inseparable, and economic issues cannot be investigated without considerations of the political, social, and institutional framework in which the economic activity takes place.
were subject to successful prosecution by the New South Wales and Victorian Environmental Protection Authorities, during the period 1990-1993. Of those firms that had been prosecuted, 18 provided positive and qualitative environmental news in their annual reports.
Only two of the companies within the sample made any reference to the prosecutions. They found that prosecuted firms disclosed more environmental information (of a positive nature) in the annual report in the year of prosecution than any other year in the sample period.
The prosecuted firms also disclosed more environmental information relative to nonprosecuted firms. The results of the study supported the view that management considered that the prosecutions negatively impacted the community's perception of the organisation, and as a result, management made other affirmative environmental disclosures in the annual report to limit the likely damage to the company's reputation as a result of the prosecutions.
More recently, Cho et al. (2012) examined two competing theories (voluntary disclosure theory 4 and legitimacy theory) to explain why some firms choose to disclose their environmental capital spending, whereas others do not. They found that disclosure does not appear to signal better future environmental performance relative to nondisclosure and that firms with worse environmental performance are more likely to disclose the amount they spend. They concluded that firms use environmental disclosure more as a strategic legitimising resource for reducing their exposures to political and regulatory concerns than as a mechanism for signalling superior environmental performance. Stanny (2013) examined voluntary disclosures concerning greenhouse gas (GHG) emissions by U.S. S&P 500 firms to the CDP from 2006 to 2008 and found that many firms only answer the CDP questionnaire but do not disclose their emission amounts or how they account for them. Consistent with legitimacy theory's predictions, she concluded that firms disclose the minimum necessary to reduce adverse public opinion, avoid scrutiny, and deter the possibility of being targeted by a shareholder resolution. This paper contributes to empirical tests of legitimacy by examining a particular class of voluntary environmental information (iCarbon) and its dissemination impact on the COE. Climate change and its consequences present one of the most persistent threats to global economic stability (Peng et al., 2015) and have the potential to affect firms' costs of equity capital, which is the required rate of return given the market's perception of a firm's riskiness (El Ghoul et al., 2011). The current emergence of investor interest in climaterelated risks calls for a specific type of global data about such risks to support rational investment decisions (The Economist, 2017).
Managers have private information about firms' carbon profiles, including the carbon strategy, carbon emissions, and carbon reduction activities that is not directly accessible by outside stakeholders (Luo & Tang, 2014). Organisations seek to protect (or enhance) past legitimacy accomplishments that they have already acquired by developing "a defensive stockpile of supportive beliefs, attitudes and accounts" (Suchman, 1995, p. 595). Lee, Park and Klassen (2015) provided empirical evidence to support this theoretical supposition. They examined a sample of Korean firms from the CDP and concluded that firms could mitigate the adverse effects of carbon disclosure on shareholder value by communicating their carbon news periodically (i.e., carbon management efforts and performance through the media coverage of global warming in daily newspapers) in advance of its carbon disclosure. It can thus be implied that managers strategically release relevant information to maximise the value of the firm as perceived by capital providers (see Beyer, Cohen, Lys, & Walther, 2010).
Accordingly, iCarbon can be considered a legitimate social contribution made by firms to enhance organisational credibility and legitimacy (see S. Y.  and can be among the various aspects of transparency in environmental reporting to change societal perceptions and to respond to climate change-related political and public pressures. iCarbon is also expected to reduce investors' incentive to acquire private information by improving the broadness of information to a wider reach of investors, reducing information asymmetry, increasing share demand, and thus reducing the COE (Blankespoor et al., 2014;Easley & O'hara, 2004). Correspondingly, using iCarbon enables a firm to transmit carbon-related information at lower acquisition costs, allowing potential investors to gain knowledge about a firm's environmental information and assess carbon-related risks. Such a strategy increases the willingness among those investors to take on a larger portion of a firm's shares, which improves risk diversification (risk sharing) and hence reduces the COE (Heinkel et al., 2001;Merton, 1987).
Legitimacy, then, is a perception resource that organisations manipulate through various communication-related strategies (Aerts & Cormier, 2009;Deegan, 2014;Higgins & larrinaga, 2014) to engage in dialogues with stakeholders, to portray an image that these organisations are trying to convey to the relevant public (Stanny, 2013), and to enhance their reputation (Auger, Devinney, Dowling, Eckert, & Lin, 2013;Beyer et al., 2010;Busch & Hoffmann, 2011;De Villiers & Van Staden, 2006;Hasseldine, Salama, & Toms, 2005;Ullmann, 1985). As an innovative source of information, iCarbon serves as one of the communication channels between a firm and its stakeholders. Legitimacy theory suggests that the need to legitimise business actions will motivate managers to voluntarily disseminate carbon-related information on Twitter. The discussion above leads to the following hypothesis: H1 The dissemination of carbon-related information on Twitter (iCarbon) has a significant and negative association with the cost of equity (COE).

| Sample and data
Our sample comprises all nonfinancial firms with official Twitter accounts that are listed on the U.S. NASDAQ stock exchange for the period from 2009 to 2015. We focus on U.S. firms because foreign firms are exposed to different transparency levels, which influence 4 Voluntary disclosure theory explains the disclosure of both general and financial environmental information (Bewley & Li, 2000). Such theory suggests that companies use the information "to signal an unobservable proactive strategy towards environmental concerns relative to poorer performing firms" (Cho et al., 2012, p. 487). their COE. Additionally, the U.S. Securities and Exchange Commission permits firms to use social media, especially an interactive platform such as Twitter, for disclosing corporate announcements. Many U.S. firms also adopt Twitter and use it for multiple purposes, including corporate announcements (Blankespoor et al., 2014;Jung, Naughton, et al., 2018), which induces an expected coverage during the sample period. We also focus on a single stock exchange to avoid any effect from exchange listing (Bushee et al., 2010). Furthermore, our sample period allows us to mitigate any macroeconomic effects of the financial crisis.
Our data collection starts by identifying whether each firm in the sample has an official Twitter account. We first search firms' websites, including the Investor Relations pages, for any links or mentions of a firm's Twitter account. If a firm has not provided any Twitter account on its websites, we identify all profiles that match their names on Twitter by using the users' search engine. We ensure that only certified accounts, with a blue verified Twitter badge, are considered, assuring that the firms are the main source of carbon-related information. We also use Google's search engine to search for firms' adoption and presence on Twitter.
To measure the implied COE, we require all firms in our sample to have positive median earnings forecasts for 1 and 2 years ahead.
These earnings forecasts are collected in June of each year to ensure that analysts have assimilated all the information from the fiscal year report in their forecasts. We also require firms to have available COE estimates. This procedure retains a full sample of 1,737 observations, representing 584 firms.
To download a firm's tweets, we use two main features that are usually used to aggregate Twitter data. We first use Twitter's application programming interface, which provides up to 3,200 tweets per user. If the number of tweets that the firm posts on Twitter exceeds 3,200, we then use keyword searches using Twitter's advanced search option.
This procedure makes it easier to manually retrieve tweets. We refine our search criteria by using keywords that relate to carbon information (e.g., carbon, climate change, CO 2 , emissions, GHG, global warming, GHG, and pollution). We then merge all firms' tweets from the Twitter application programming interface and advance search under one file.
We use two sources (Bloomberg and DataStream) to collect the data used to estimate the dependent and control variables. We also use LexisNexis to count the number of articles that are disseminated on other communication channels and that are related to carbon information. We allocate these articles by using company identifiers and keyword search features. We use our carbon keyword list, mentioned in Section 3.2.2, to retrieve carbon-related news articles. This procedure allows us to retrieve articles from many sources, such as The Wall Street Journal, USA Today, The Washington Post, and The New York Times. We also Winsorise the amount of carbon news coverage (CD_NEWS), financial leverage (LEV), long-term growth forecast (LTG), beta coefficient (BETA), book-to-market ratio (BTM), earnings surplus (SURP), and the dispersion of analysts' forecasts (DISP) at the 2.5th to 97.5th percentiles to control for outliers. This Winsorising level is also used for the COE to eliminate negative values because we are not expecting investors to require a negative rate of return.

| Cost of equity
Our dependent variable (COE) is based on the implied COE (El Ghoul et al., 2011;Hail & Leuz, 2006), which is measured as the average of four COE estimates: (a) Claus and Thomas' model (Claus & Thomas, 2001), R CT ; (b) Gebhardt, Lee, and Swaminathan's model (Gebhardt, Lee, & Swaminathan, 2001), R GLS ; (c) Ohlson and Juettner-Nauroth's model (Ohlson & Juettner-Nauroth, 2005), R OJ ; and (iv) Easton's model (Easton, 2004), R MPEG . We use the average of these estimates to reduce any estimation error of the COE (Hail & Leuz, 2006). We also use this measure because it enables us to differentiate between the influence of both cash flow and growth from the COE (Chen, Chen, & Wei, 2009). This estimate is useful for time-series variations in the COE (Pástor, Sinha, & Swaminathan, 2008).

| iCarbon
Our independent variable, iCarbon, reflects the number of carbonrelated tweets that are disseminated to the public. We compute this measure by searching for keywords and phrases that relate to carbon-related information. In this regard, we use many keywords that were used in the prior literature and that align with carbon disclosure, reporting and information (e.g., Griffin & Sun, 2013;Hahn, Reimsbach, & Schiemann, 2015;Hsu & Wang, 2013;Lee, Park, & Klassen, 2015;Schmidt, Ivanova, & Schäfer, 2013). We also use the Twitter hashtag key (#), a feature that can be used to broaden climate information and trigger discussions among users about an event or specific topic.
Thus, we include many hashtags that relate to carbon emissions, climate change, and global warnings. In general, we define several keyword lists based on combinations of words and single phrases to identify iCarbon tweets.
After matching firms' tweets with our keyword lists, we count the annual number of tweets that match our keyword lists for each firm or zero otherwise. Appendix A provides some examples of iCarbon tweets.

| Control variables
Our control variables include many variables associated with firm characteristics such as firm size (SIZE), BTM, and LEV (Botosan, 1997;Fama & French, 1992;Hail & Leuz, 2006). Larger firms have a better information environment and thus a lower COE (Gebhardt et al., 2001). The COE increases for undervalued firms that have a greater BTM ratio. Additionally, firms that have high LEV in their capital structure expect to have a higher COE (Cao, Myers, Myers, & Omer, 2015). We also expect a positive association with the DISP, BETA, and LTG. Firms that have a more uncertain information environment, systematic risk or market mispricing would be expected to have a higher COE (Botosan, Plumlee, & Wen, 2011;Cao et al., 2015;El Ghoul et al., 2011;Gebhardt et al., 2001;Gode & Mohanram, 2003). We further control for the availability of information by other intermediaries by including the amount of CD_NEWS and the percentage of institutional holdings (INSTOWN; Cao et al., 2015;Zhou et al., 2018). We expect higher carbon coverage (CD_NEWS) and institutional ownership (INSTOWN) to improve a firm's information environment and thus be associated with a lower COE (Cao et al., 2015;Griffin & Sun, 2013;Li et al., 2017). We also consider the content of firm news by controlling for SURP. Due to the higher uncertainty of future earnings profitability, we expect that firms with negative earnings (LOSS) are difficult to analyse and thus have a higher COE (Orens, Aerts, & Cormier, 2010). Furthermore, we include many variables that determine climate change/carbon information. Additionally, we control for independent directors (BOD_IND), the environmental committee (ENV_COMMITTEE), CDP participation (CDP), firm age (AGE), and whether the firm is subject to the Environmental Protection Agency's (EPA) Mandatory Reporting Rule. Independent board directors play a monitoring role in managerial decisions and activities, which enhances disclosure policy and transparency. An environmental committee plays an advisory role in the better management of emissions and disclosure policy and a motivating role in reporting reliable information. We also include the CDP to control for firms' willingness to report carbon information. This measure represents the firm's ability to identify carbon-related issues and their potential consequences (Jung, Herbohn, & Clarkson, 2018). Aged firms "tend not to choose to operate environmental information disclosure" (Zeng et al., 2012, p. 317). Firms in industries that are more sensitive to carbon information are more inclined to choose greater transparency in the policy of disclosure to avoid the scrutiny of regulators (Deegan & Gordon, 1996). Therefore, we expect firms under EPA regulation to respond more to investor demand and to use iCarbon more. Technology firms are expected to be more inclined towards technology adoption, and thus, we expect them to be more active on Twitter (Blankespoor et al., 2014). The full definition and measurement of our dependent, independent, and control variables are presented in Appendixes B and C.

| Model
To examine the impact of iCarbon on COE, we employ the following Model 1: Our estimation procedures employ pooled Ordinary Least Squares (OLS) regressions with robust standard error clustered at the firm level to control for serial correlation and heteroscedasticity (Cao et al., 2015;El Ghoul et al., 2011;Ferris, Javakhadze, & Rajkovic, 2017;Petersen, 2009). 5 We also utilise a two-stage least squares (2SLS) model as an alternative estimation, clustered at the firm level, to control for any potential endogeneity between iCarbon and the COE (Nikolaev & Van Lent, 2005). In this model, we use both the lagged value of iCarbon and the industry-year iCarbon mean as our instrumental variables. These instruments are more related to a firm's engagement in iCarbon but do not necessarily affect the firm's value or COE (Cheng, Ioannou, & Serafeim, 2014;Schreck, 2011 The partial R 2 is equal to 0.844, with an F statistic higher than the critical value (Staiger & Stock, 1997 indicates that larger firms use iCarbon more, which is consistent with prior findings (Lee, 2012;Weinhofer & Hoffmann, 2010). Our results show that iCarbon is positively correlated with DISP and CD_NEWS.
Firms that have negative earnings are less likely to use iCarbon.
Conversely, higher BOD_INDP leads to increased use of iCarbon.
Consistently, firms that participate in the CDP disseminate more carbon-related information on Twitter. Overall, the correlation matrix and unreported variance inflation factor tests indicate that multicollinearity is not an issue across our empirical models.    greater institutional ownership enhances a firm's information environment, which reduces uncertainty and thus also reduces the COE. Furthermore, the nature of the industry may have a differing effect on the COE (Fama & French, 1997). Our results show that technology firms (TECH_FIRM) tend to have a lower COE. These firms face greater demand for information, which motivates them to provide more information through disclosure (Kothari, 2000). Previous studies have found that firms that belong to this industry and use Twitter to disseminate corporate information reduce information asymmetry and improve market liquidity (Blankespoor et al., 2014)

| The effect of Bloomberg's environmental (ENV) and ESG disclosure
We further address whether a firm's level of environmental disclosure would affect the association between iCarbon and the COE. Firms that are more socially responsible have more incentives to disclose and engage in environmental activities and practices (Harjoto & Jo, 2015). These firms are motivated to maintain and improve their public images by generating positive media coverage, which, in turn, improves firm value and decreases the COE (Cahan, Chen, Chen, & Nguyen, 2015;Fatemi, Glaum, & Kaiser, 2018). That is, investor preference for environmentally friendly firms can lead to a lower investor base that is willing to buy and hold shares in polluting firms. This preference reduces risk sharing and thus increases firms' equity financing, creating environmental costs for firm managers (Chava, 2014;Heinkel et al., 2001;Merton, 1987). Accordingly, poor environmental performance induces lower demand by institutional investors and less "loan syndicate" participation by banks (Chava, 2014;Hsu & Wang, 2013).
These studies show that firms should consider the benefits of environmental information to reduce their equity financing. Accordingly, firms with different levels of environmental performance induce different behaviours towards using communication channels to respond to environmental issues and concerns (de Villiers & Van Staden, 2011).
As such, firms with better environmental performance promote more voluntary climate change disclosure (Dawkins & Fraas, 2011). We therefore expect firms with a higher environmental disclosure score to use iCarbon. Hence, we address whether a firm's disclosure score of environmental reporting would affect our main findings.
We use Bloomberg for firm environmental disclosure (ENV_SCORE). This variable incorporates data from many sources, including annual reports, the CDP, firms' websites, and CSR reports, generating a comprehensive score for firm disclosure. This score is estimated in terms of both industry relevance and data availability, starting from 0.1 for low-disclosing firms and continuing up to 100 for highdisclosing firms. The weighting system takes into account the importance of each category, making a category such as GHG emissions carry greater weight than other disclosure items. Weighting each data point in terms of its importance makes the disclosure score reflect both the quality and quantity of disclosure (Bernardi & Stark, 2018;Qiu, Shaukat, & Tharyan, 2016). We address this issue by including the environmental score (ENV_SCORE) and the interaction between iCarbon and the environmental score (iCarbon * ESG_SCORE) in Model 2 as follows: We also examine a broader aspect of a firm's disclosure than simply environmental reporting by taking into account two components of sustainability reporting in addition to environmental disclosure: social and governance disclosures. In this section, we address whether a firm's disclosure score of ESG disclosure would also influence the association between iCarbon and the COE. The combination of all ESG dimensions enables many investors to evaluate a firm's risks, opportunities, and transparency, which in turn improves firm value and reduces the COE (Ng & Rezaee, 2015;Yu, Guo, & Luu, 2018). Such an effect is more pronounced for lower-ESG-disclosure-performing firms than for higher-ESG-disclosure-performing firms (Crifo, Forget, & Teyssier, 2015).
These firms are likely to disclose their ESG activities and initiatives to signal and differentiate themselves in the capital market from those with lower ESG disclosure ratings (Crifo et al., 2015). We therefore expect firms with better ESG disclosure scores to strategically use iCarbon more than those with lower ESG disclosure scores. 7 Therefore, we investigate whether the ESG disclosure score (ESG_SCORE) would moderate the association between iCarbon and the COE. To examine this influence, we include ESG_SCORE and its interaction with iCarbon (iCarbon * ESG_SCORE) in Model 3: 7 We use the Bloomberg database to obtain the ESG disclosure score, which reflects a firm's social, environmental, and governance data that are available to the public from corporate websites, press releases, annual reports, sustainability reports, and corporate governance reports. The score covers many topics such as board structure and independence, human capital, shareholders' rights, and GHG emissions. Such information is reflected in the ESG index score to reflect both the amount and importance of information. The score ranges from 0.1 to 100, where each data point is weighted in term of its importance and relevance to industry peers.
We employ OLS regression with a robust standard error cluster at the firm level to estimate both Models 2 and 3 and present the results in Table 2. In these models, we have centralised our explanatory variables (i.e., iCarbon, ENV_SCORE, and ESG_SCORE) and their interactions (i.e., iCarbon*ENV_SCORE and iCarbon*ESG_SCORE). The finding from Model 2 shows that ENV_SCORE does not affect the association between iCarbon and the COE, as the interaction between iCarbon and ENV_SCORE has no significant coefficient with the COE. This result means that the number of iCarbon tweets has a direct association with the COE, which is not affected by the environmental disclosure score.
The result from Model 3 shows a similar finding of a negative association for iCarbon on the COE, which is consistent with our main findings. The results also show no significant association for the interaction iCarbon*ESG_SCORE.
Similarly, we found no significant association between ESG disclosure and the COE. Overall, the findings provide evidence that the association between iCarbon and the COE is not affected by either ENV_SCORE or ESG_SCORE. The results support our argument that investors appreciate carbon messages and dissemination, which is different from the reporting score.

| Robustness checks
As a robustness check, we use different measures for the COE and iCarbon and add different sets of control variables to our main Model 1. The results are reported in Table 3. We use R PEG (Easton, 2004) as an alternative measure of the COE. R PEG is considered a reliable measure for the COE and is widely used in the literature (Mangena, Li, & Tauringana, 2016). This measure assumes no dividend payout and is associated "with firm-specific risk characteristics in a theoretically predictable and stable manner" (Botosan et al., 2011(Botosan et al., , p. 1085.
We employ the analysis in our main Model 1 by alternatively using R PEG instead of the COE in Column 1. The results show consistent evidence that iCarbon is negatively associated with the COE, as measured by R PEG .
We also use two alternative measures of iCarbon. First, we use the number of iCarbon tweets that have a hyperlink. Including a hyperlink allows users to acquire more information by following the link (Blankespoor et al., 2014). Second, we use the number of iCarbon tweets that have been retweeted. This measure enhances the size of the audience as users share a firm's iCarbon tweets with their followers through the retweet button (Jung, Naughton, et al., 2018). Cade (2018) claim that retweeted messages are considered more valid by investors. We present the results in Columns 2 and 3 in Table 3.
The results indicate that tweets with a hyperlink to the full information of a press release or news articles (iCarbon_Hyperlink) that are diffused to a larger number of users (iCarbon_Retweet) on Twitter are negatively associated with the COE. This finding is consistent with our main findings.
In Column 4, we control for multiple variables used in the prior lit-    Table 2 presents the effects of environmental and ESG reporting on the association between iCarbon and COE. The sample comprises of nonfinancial NASDAQ firms with Twitter accounts for a period for a period from 2009 to 2015.See Appendixes B and C for variables descriptions and measurements. Model (2) presents the results after adding environmental reporting (ENV) score and its interaction with iCarbon. Model (3) includes environmental, social, and governance (ESG) score and its interaction with iCarbon. The coefficient estimates are results from pooled regression (OLS) clustered at the firm level. In parentheses, robust standard errors are presented.
*10%. **5%. ***1%.    Lee, Hutton, & Shu, 2015). We control for the ratio of advertising expenses to total assets (ADVERTISING) and a dummy variable for whether a firm's CEO is younger than the average (CEOAGE) and the percentage change in sales growth (SALES_GRWOTH). Firms that spend more on advertisements and have younger CEOs and high growth rates in sales are expected to adopt social media, have Twitter accounts, and disclose more announcements on communication channels (Jung, Naughton, et al., 2018;Lee, Hutton, & Shu, 2015). We also expect a firm's valuation to increase by generating high sales growth. Additionally, some industries are subject to different litigation risks and more potential lawsuits. Hence, we include dummy variables (LITI) for firms that operate in high-litigation industries (Dhaliwal et al., 2011). We also control for research and development (R&D) and capital expenditure (CAPX). Although R&D is an expense that a firm pays, this expense might generate value (Servaes & Tamayo, 2013). Furthermore, firms with high growth in sales (SALES_GROWTH, R&D and CAPX) are expected to disclose more environmental information (Dhaliwal et al., 2011;Harjoto & Jo, 2015). The results show negative associations for ADVERTISING and CAPX with the COE. In contrast, DUM_CEO, GROWTH_SALES, R&D, and LITI have no association with the COE. These findings mitigate any concern towards a firm's willingness to adopt Twitter and disclose carbon information.
Finally, we further reestimate our regression model by using the generalised method of moments (GMM). 8 We use the GMM model to address the endogeneity problem that may affect the interpretation of our association between iCarbon and the COE. Our results in  This association holds consistently throughout alternative estimations and is not affected by either environmental or ESG disclosure. Overall, our results suggest that the increase in a firm's dissemination of carbon information improves investor recognition among many potential investors and environmentally concerned groups, reduces information asymmetry between market participants, and enables investors to evaluate firms' potential risk and acquire firm information at lower acquisition costs, which in turn reduces the COE.

| CONCLUSION
This paper provides several implications for market participants, managers, and policymakers about integrating information technology into their strategic voluntary disclosure policy. Our results show the importance of firm managers considering the dissemination of carbon-related information seriously and the benefit to the COE. As Twitter allows market participants to receive firm information in a timely and efficient manner, iCarbon enables many market participants to assess a firm's potential risk and make better investment decisions.
Additionally, firms should consider using iCarbon to address investors' EPS t + 1 = The median forecast of EPS for June next year DPS t + 1 = Dividend per share (DPS) for the next year or 6% of ROA g 2 equals to the growth rate of short-term earnings (EPS t + 2 /EPS t + 1 -1) or long-term consensus analysts' earnings forecasted. This model requires both EPS t + 1 and EPS t + 2 to be positive. g lt equals to 10-year treasury bonds yield minus 3% (2004) cost of equity model

Modified Easton
The model measures EPS for the first 3 years by using analyst earnings forecast. The fourth and fifth earnings forecasted years are measures by multiplying the previous year earnings forecast by long term earnings growth rate (LTG). If the LTG rate is missing, short-term growth rate of FEPS t-2 and FEPS t-2 is used. The model measure g lt as the difference between 10 years Treasury bonds and 3%. The model also assume clean surplus relation to measure future book value (B t + i − 1 = B t + EPS t + 1 − DPS t + 1 ). Future dividend is measured by multiplying EPS by dividend pay-out ratio (DPS t + 1 = EPS t + 1 ⨯ FDIV).

Gebhardt et al. (2001)
The model use analyst forecast to measures future return on equity (FROE) of the first 3 years. Afterward, FROE is measured by using linter interpolation of 10 years historical industry specific ROE median. If industrial ROE is lower the risk-free rate, we use risk free rate to replace industry ROE (Liu, Nissim, & Thomas, 2002). Beyond the 12 year, the model assumes industry ROE to remain constant. Clean surplus is used to measure future book values. Where: B t + i − 1 = B t + EPS t + 1 − DPS t + 1 DPS t + 1 = EPS t + 1 ⨯ FDIV

COE
The average of four cost of equity estimates (R OJ, R MPEG, R CT , and R GLS ) risk free