Corporate carbon performance data: Quo vadis?

Corporate carbon performance (CCP) has become a central topic in political, financial, and academic domains. At the same time, several characteristics of CCP data, including comparability and consistency, remain unresolved. The literature has extensively covered issues regarding the comparability of CCP data from a firm‐internal perspective. However, it has not yet examined the consistency of CCP data between third‐party data providers. This article investigates the degree of CCP data consistency between third‐party providers according to three dimensions: scope (i.e., direct and indirect emissions), scheme (i.e., mandatory and voluntary reporting schemes), and source (i.e., data stemming from corporate reports and from third‐party estimation methods). The results reveal that data on direct emissions are more consistent than data on indirect emissions, and they are especially inconsistent for Scope 3. Second, mandatory and voluntary reporting schemes do not substantially improve the consistency of CCP data, which is surprising. Third, third‐party estimations are less consistent as compared to data stemming directly from corporate reports; however, the combination of Scopes 1 and 2 third‐party estimated data raises consistency levels. On the basis of these results, we conclude the following key implications: academic researchers must be mindful of the consistency of CCP data, because it can significantly affect empirical results, corporate management should avoid situations where different CCP data are communicated externally, investors should engage firms to follow a standardized approach, data providers should increase the transparency about their estimation methods, and policy makers need to be aware of the importance of a sound and standardized methodology to determine CCP.

terms of comprehensiveness, CCP data are typically available for large listed companies only, neglecting significant economic areas, such as supply chains and small and medium-sized enterprises. In terms of quality, the literature has extensively covered how companies should gather and report CCP data from a firm-internal perspective (Downie & Stubbs, 2013;Hahn et al., 2015), but has assessed only to a limited extent the consistency of CCP data across firms from an external stakeholder perspective (Andrew & Cortese, 2011;Marland, Buchholz, & Kowalczyk, 2013;Matisoff, Noonan, & O'Brien, 2013;Stanny, 2018).
Furthermore, the consistency of CCP data between third-party data providers has not yet been analyzed in a systematic manner. Third-party data providers, including Bloomberg, CDP, ISS Ethix, MSCI, Sustainalytics, Thomson Reuters, and Trucost, have established extensive datasets covering thousands of firms' carbon emissions over many years. We match this provided data over multiple years and address the following research questions: (RQ1) How consistent are CCP data between different third-party data providers in general?; (RQ2) To what extent do mandatory and voluntary reporting schemes affect the consistency of CCP data between third-party providers?; and (RQ3) How consistent are estimation methods between third-party providers?
We applied correlation analyses to measure consistency of CCP data offered by major third-party data providers. The analyses distinguish three dimensions of CCP data: scope (i.e., direct and indirect emissions), scheme (i.e., mandatory and voluntary reporting schemes), and source (i.e., data stemming from corporate reports and from third-party estimations). The results show that the consistency of CCP data is high for direct emissions, and progressively decreasing in Scope 2 to 3 indirect emissions. Subsequently, the results highlight that mandatory and voluntary reporting schemes do not substantially improve the consistency of CCP data between data providers. This is especially surprising for mandatory reporting schemes, that is, the EPA's Greenhouse Gas Reporting Program (GHGRP) and EU Emission Trading Scheme (ETS), as they prescribe the strict adherence to specific accounting and reporting methods. Third, data stemming from corporate reports is more consistent than data generated from third-party estimations; however, consistency of third-party estimations increases significantly when Scopes 1 and 2 are combined. From these results, we derive several implications for academic research, corporate management, financial markets, data providers, and policymaking.
The disclosure of CCP data "potentially reduces the principal-agent problem of asymmetric information by increasing transparency" (Hahn et al., 2015, p. 86). As such, companies disclose CCP data through voluntary channels-for example, sustainability reports or CDP-as a way to signal investors and other stakeholders of their superior firm performance (Clarkson, Li, Richardson, & Vasvari, 2008;Diaz-Rainey, Robertson, & Wilson, 2017). However, several issues remain pertaining to the quality of CCP data.

The quality of CCP data
In an attempt to increase the quality of CCP data, the GHG Protocol was released over 15 years ago as a standardized framework in accounting and reporting GHG emissions in organizations, including entire operations and value chains (WRI/WBCSD, 2004). Since 2008, the CDP encourages all companies to use the GHG Protocol to increase comparability (CDP, 2017). However, the comprehensiveness and quality of CCP data remain a major issue to this day (Matisoff et al., 2013;Stanny, 2018). Hummel and Schlick (2016) describe several desirable characteristics of data quality. These include verifiability, reliability, comparability, and consistency. In terms of data verifiability and reliability (i.e., the data are accurate, fair, and true), the literature has extensively covered how firms internally gather and report CCP data, for example, examining the drivers and motives to report according to guidelines, such as the Greenhouse Gas Protocol (Downie & Stubbs, 2013;Hahn et al., 2015;Schaltegger & Csutora, 2012). Hummel and Schlick (2016) discover an interesting finding when it comes to data reliability. The authors find that superior sustainability performers choose to disclose high-quality, timely sustainability reports, whereas poor sustainability performers disguise their true performance by releasing low-quality, less frequent reports (Hummel & Schlick, 2016). The latter finding was confirmed in a recent study showing how companies conceal information and employ various impression management strategies, including making excuses and delays, to hide poor performance (Talbot & Boiral, 2018).
The literature acknowledges the differences and limitations of CCP data comparability across firms. Andrew and Cortese (2011) examined different accounting methodologies from various reporting schemes, and concluded that CCP data will remain incomparable and ultimately unreliable for an unforeseeable future. Matisoff et al. (2013) studied the degree of data convergence in CDP from 2003 to 2010, examining trends in disclosure, transparency, and comparability of corporate carbon emission data over this period. They concluded as the amount of companies reporting increases, the comparability between disclosures increases in Scopes 1 and 2, but not in Scope 3. However, as more companies disclose CCP data, Matisoff et al. (2013) observed a decrease in the percentage of reports becoming available, hence less reliable for public scrutiny.
However, the consistency of CCP data between third-party providers has not been empirically investigated. Third-party data providers generate large datasets based on data from corporate reports and their own estimation methods. The latter approach is used primarily when companyreported data are not available. Thus, our first research question deals with the general level of consistency for CCP data between different thirdparty data providers.

Mandatory and voluntary reporting schemes
Most companies report CCP data according to either mandatory or voluntary reporting schemes (Andrew & Cortese, 2011). Several regions and nations require energy-intensive companies to report annual carbon emissions of specific facilities according to a mandatory reporting scheme.
For example, the European Union Emission Trading System (EU ETS) covers approximately 11,000 facilities, including large power stations and manufacturing plants, which cover around 45% of the GHG emissions in the European Union (EU Commission, 2018b). The U.S. Environmental Protection Agency (USEPA, 2018) requires facilities that emit more than 25,000 metric tons of CO 2 -equivalent (CO 2 e) emissions per year to disclose their annual emissions through the Greenhouse Gas Reporting Program (GHGRP). This program covers over 8,000 facilities and roughly 50% of the total U.S. emissions (USEPA, 2017).
Mandatory reporting schemes were established to monitor facility-level CCP data according to strict, formal rules for accounting and reporting (Perrault & Clark, 2010;Sullivan & Gouldson, 2012). Since these schemes consider carbon emissions at facility level, all covered emissions related to Scope 1. It can be assumed that emission data reported under a mandatory regulation is regarded as highly standardized and consistent. Nevertheless, it has been found that other aspects, such as reporting boundaries and emission factors, can be manipulated and, thus, may affect data comparability (Dragomir, 2012;Sullivan & Gouldson, 2012;Talbot & Boiral, 2018).
Beyond mandatory schemes, several voluntary carbon reporting initiatives have emerged. Voluntary reporting schemes have an advantage over mandatory reporting, as "voluntary disclosure programs like the CDP may allow firms to engage with stakeholders such as investors and employees more directly than mandatory reporting requirements and serve as a way to improve internal management of GHGs" (Matisoff et al., 2013, 297).
Undoubtedly, the most successful voluntary initiative is the CDP (formerly the Carbon Disclosure Project) with more than 2,400 companies disclosing CCP data (CDP, 2017). Since 2003, the CDP collects data via an annual questionnaire from major corporations in various capital markets.
Despite its success, several studies have raised concerns regarding the quality of data gathered and published by the CDP (CDP, 2010; Andrew & Cortese, 2011;Matisoff et al., 2013;Stanny, 2018).
The distinction between mandatory and voluntary reporting schemes is relevant in this article because they differ in scope and scheme. On the one hand, mandatory reporting schemes are stricter in terms of accounting methodology, but they are limited to direct emissions (Scope 1) and pertain to individual facilities (i.e., not entire firms). On the other hand, voluntary disclosures encourage companies to report both direct (Scope 1) as well as indirect emissions (Scopes 2 and 3). While such schemes encourage firms to report emissions for the entire firm and its value chain, they are much more lenient in accounting methodology, and this makes the comparability between firms difficult (Perrault & Clark, 2010). We explore the extent that mandatory and voluntary reporting schemes affect the consistency of CCP data between third-party providers.

Third-party estimation methods
Although the number of firms disclosing their carbon emissions has increased over the years, many companies still do not report their carbon emissions. In those cases, third-party providers can use their own estimation methods to fill these data gaps. Such estimation methods are also essential in cases where investors and other stakeholders would like to assess the carbon footprint of supply chains since CCP data are typically scarce further along the supply chain.
The literature suggested three general approaches for estimation methods (Goldhammer, Busse, & Busch, 2017). First, a process analysis (PA) approach using primary and secondary production process data and the associated carbon emissions (Minx et al., 2009;Suh & Huppes, 2009;Wiedmann & Minx, 2008). This approach has been used extensively to estimate carbon footprints on a product level (Burkhardt, Heath, & Cohen, 2012;Dolan & Heath, 2012;Mazor, Mutton, Am Russell, & Keoleian, 2011) as well as on a firm level (Block et al., 2011;Gooding, 2012;Lee & Cheong, 2011). This approach is very detailed and requires a lot of specific information. At the same time, boundaries for proper calculation need to be established, which may result in a lack of completeness of the estimation (Minx et al., 2009;Suh & Huppes, 2009).
Second, input-output analysis (IOA) uses the input-output tables of national accounts to allocate the carbon emissions of the economy to smaller units on a sectoral basis. This approach has been applied to carbon footprint calculation at the firm level (Minx et al., 2009), for example, by ascribing emissions to individual firms according to their relative proportion of the firm's sales compared the sectors overall sales numbers.
Assessing emissions per unit of turnover appears to be an appropriate frame of reference since it allows the most comparable analysis of individual CCP in the absence of any further company specific information. This method can be extended to a multi-regional input-output (MRIO) analysis (e.g., Lundie, Wiedmann, Welzel, & Busch, 2019). Therefore, MRIO analysis can also capture the Scope 3 emissions along the entire supply chain (Huang, Lenzen, Weber, Murray, & Matthews, 2009). The estimations resulting from MRIO analyses represent averages and do not consider firm specific aspects (Minx et al., 2009).
Third, hybrid approaches merge the strengths of both PA and IOA. They have been coined the state of the art for carbon footprinting (Wiedmann, 2009). The Norwegian "Klimakost" model is one example of a hybrid approach being applied to companies, municipalities and other organizations (Larsen, Solli, & Pettersena, 2012).
Third-party data providers combine and extend these approaches to estimate CCP data. ISS-Ethix estimates CCP data with the help of more than 800 sub-sector specific models, where every sub-sector is analyzed to identify emission predictors (ISS-Ethix Climate Solutions, 2019). MSCI ESG employs different estimation methods to determine CCP, for example, based on a company's previous emission intensities (MSCI ESG Research, 2019). Sustainalytics uses more than 80 different estimation models to determine Scope 1 and 2 emissions, which are based on historical company data and non-linear regressions (Sustainalytics, 2019). The 80 different models are then calculated as average emissions per million USD revenue.
Thomson Reuters ESG applies one of three different estimation methods to determine Scope 1 and 2 emissions in the following order, according to data availability: (a) previously reported emissions, (b) energy consumption, and (c) median emissions of industry or business sector (Thomson Reuters ESG, 2019). Finally, Trucost uses its own Extended Environmental Input-Output (EEIO) model based on industry-specific environmental impact data, and the flow of goods and services between economic sectors (Trucost ESG Analysis, 2019).
While the individual approaches differ, the third-party providers do share some commonalities in the estimation methods. Thomson Reuters ESG and Sustainalytics, for example, use similar business metrics to calculate industry average CCP data in the form of intensities and then convert those back to absolute emissions. ISS-Ethix and Trucost base their estimations on sector and country-specific models utilizing industry input-output tables. It appears that most third-party providers use industry-based methods over location-based methods. While various estimation methods can still yield different results, a high degree of consistency between third-party data estimations would be desirable. Thus, our third research question examines how consistent estimation methods are between third-party providers.

Sample selection
In order to establish a representative sample, we selected third-party data providers based on two criteria. The first criterion is data type-data providers must offer company-wide CO 2 e emission data in Scopes 1 and 2 as a minimum requirement, and if available, Scope 3 emissions. We excluded any data providers that only offered carbon scores or climate ratings to represent CCP information. The second criterion is data rangethird-party providers must offer historical data for more than 5 years of same-firm data. Additionally, they should provide a global coverage of firms. We excluded any third-party data providers with only a few years of coverage, including those that have discontinued this service, as well as third-party providers focusing on a single country.
We derived a list of major third-party CCP data providers, including Bloomberg, CDP, ISS Ethix, MSCI, Sustainalytics, Thomson Reuters ESG, and Trucost. Table 1 shows the data type, timespan, coverage of companies having reported data, total company coverage, and additional details of each provider. We obtained this information through the providers' websites and downloadable factsheets. When information was not available, we contacted the data providers directly, and asked them to provide supplemental information.
We matched CCP data from different providers for the same firm using the International Securities Identification Number (ISIN), resulting in a sample of 15,485 firms between 2005 and 2016. From this sample, we were able to draw firm-year observations where at least two providers can be compared: 50,793 observations (Scope 1), 50,609 observations (Scope 2), and 12,355 observations (Scope 3). The data correspond to the years of actual emissions. We calculated the pairwise Pearson correlation coefficient to analyze the strength of the relationship and thus the degree of consistency between two individual data providers. In addition, we also calculated the Spearman rank correlation coefficients, as they are less affected by potential outliers in the data.
In order to scrutinize the effects of mandatory emission reporting schemes, we utilized facility level carbon emissions as provided by the USEPA GHGRP and the EU ETS. We matched these facility emissions to their respective primary owners on a firm level. This allowed us to calculate aggregated facility emissions that are matched to one company. The USEPA GHGRP database provides information on the ownership structure of each facility through company names and the ownership percentage. Using these company names, we matched facilities to a single company ISIN. In cases where a facility has multiple owners, we considered the primary owner as the principal owner for our analysis. For the EU ETS, the World Carbon Market Database (WCMD, 2019) offers facility to company matching tables, including the ISINs. With the ISIN as the identifier, we were able to compare aggregated facility level emissions with CCP data from third party providers.

Descriptive statistics and data handling
The descriptive statistics illustrate a heterogeneous picture for all emission scopes (see Table 2). Maximum values, means, and standard deviations vary substantially across providers for all scopes, pointing toward high levels of inconsistency. The large differences for maximum values made it Note. Table 1 contains an overview of major CO 2 e emission data providers. The information in this table is derived from the respective database documentation of each provider. The timespan corresponds to the years of emissions and not years of reports.
necessary to further investigate potential outliers. Since Pearson correlation coefficients are sensitive to extreme outliers, we manually investigated some of the extreme outliers, for example, 88,000,000,000 t CO 2 e emissions indicated as the annual Scope 3 emissions of one company, which is more than the annual anthropogenic emissions. The additional analysis of the descriptive statistics in Table 2 led to the conclusion that the dataset had to be adjusted by removing severe outliers. To minimize the effect of potential errors in the data, we removed outliers by deleting 0.05% of the observations at the low and high end of the distribution for each provider. This adjusted dataset shows significantly higher correlations in all scopes and was used as a basis for further analyses.
Several data providers were not included in specific analyses due data availability reasons. ISS Ethix could not be included in many analyses, as they only offered combined Scope 1 and 2 data and estimated Scope 3 data. Sustainalytics was left out of Scope 3 analyses, as they do not provide Scope 3 data at all. Bloomberg does not reveal if the data stem from company reports or third-party estimates; therefore Bloomberg only included in general analyses (Bloomberg Finance L.P., 2019). The analyses for third-party estimated Scope 3 data was limited to two providers offering related data, ISS Ethix and Trucost. Note. Table 2 shows the descriptive statistics of the data as it is available directly from each provider. No alterations to the data were made. All emissions are measured in metric tons.

RESULTS
With respect to RQ1, the results show that Scope 1 data are highly consistent between most data providers with an average Pearson correlation coefficient of 0.97 and average Spearman correlation coefficient of 0.95. As shown in Table 3, a general pattern emerges: Scope 1 data are more consistent than Scope 2 data, and Scope 2 data are more consistent than Scope 3 data, where the biggest inconsistencies can be found.
The Spearman rank correlations exhibit similar patterns for all our analyses and generally confirm that low Pearson correlation levels are not exclusively the result of individual outliers in the data.
Additionally, we investigated the development of consistency between third-party providers over time (compare Figure S1 in the Supporting Information). The general expectation would be that third-party data providers gain experience in gathering and processing CCP data over time.
However, the overall data consistency fails to improve substantially. On the contrary, data on Scope 3 emissions became even less consistent over time.
In answering RQ2, we use facility-level emission data from the EPA GHGRP and EU ETS. the USEPA GHGRP for two reasons. First, the sustainability report includes global Scope 1 emissions, while the USEPA GHGRP report only covers U.S. facilities. Second, the calculation methods they used for the facilities covered by the EPA GHGRP is different than for the facilities outside of this regulation and facilities with less than 25,000 t CO 2 e emissions annually (Enbridge Inc., 2017).
Given the results in Table 4, we expected a low consistency measured through correlation coefficients between aggregated facility level emissions and company Scope 1 emissions. The results in Table 5  Next to mandatory schemes we investigate the extent of how voluntary reporting schemes affect the consistency of CCP data between thirdparty providers. To measure this properly, we first established a baseline by investigating the correlations of CCP data indicated as companyreported (Table 6). We found that company-reported data increased the overall correlation levels in Scopes 1 and 2. Moreover, Table 6  Note. Table 4 shows the facility level CO 2 e emission data provided under the EPA GHGRP and the EU ETS, compared to the Scope 1 emission data from third party providers. For each provider, we show the share of emissions under the EPA GHGRP and the EU ETS of the total reported Scope 1 emissions. Note. Table 5 shows the Pearson and Spearman correlation coefficients between the CO 2 e emission data provided under the EPA GHGRP and the EU ETS and the third party providers in our study. * Indicates a significance level of p < 0.01, number of observations in parentheses.

TA B L E 5 Correlation between aggregated facility emissions and Scope 1 emissions from data providers
that the combination of Scopes 1 and 2 into a single data item also yields higher correlation levels compared to Scope 2 alone. Consistency in Scope 3 emissions does not improve with company-reported data when compared to the overall findings (Table 3), which again reveals how difficult it is to properly gather these data.
Next, we created a subsample of company-reported data only for firms included in the CDP database. The findings show opposite trends when compared to the results in Table 6. We detected lower correlations for Scopes 1 and 2. This suggests that reporting to the CDP can potentially cause further data inconsistency. One explanation for greater inconsistency may be related to different values presented as "company-reported data" (i.e., The results in Table 7 illustrate that-not surprisingly-correlation levels between data providers are lower when using only third-party estimations for both Scope 1 and 2 emissions. For example, the Pearson correlations of Scope 1 emissions between MSCI and Sustainalytics decrease from 0.99 (company-reported) to 0.77 (third-party estimates) and from 0.98 (company-reported) to 0.69 (third-party estimates) for Scope 2 emissions. When combining Scopes 1 and 2, the correlation levels improve for third-party estimated data even though the underlying estimation models used by the data providers differ considerably. The two estimations for Scope 3 emissions yield inconsistent results. In this case, ∼97% of observations differ by more than 10% between the two providers. These low correlations highlight the differences in third-party estimation results. An extreme example of this is the WH Group, the world's largest pork producer, where TruCost estimates 39,839,717 and ISS 6,707,330 t CO 2 e Scope 3 emissions in 2015. Note. Table 6 shows the Pearson (lower left side) and Spearman (upper right side) correlation coefficients between third party data providers. The data included in this table is limited to those observations that were clearly identified as originating from company reports by the provider. * Indicates a significance level of p < 0.01, number of observations in parentheses. Note. Table 7 shows the Pearson (lower left side) and Spearman (upper right side) correlation coefficients between third party data providers. The data included in this table is limited to those observations that were clearly identified as originating from estimations by the provider. * Indicates a significance level of p < 0.01, number of observations in parentheses.

DISCUSSION
The findings are important along three dimensions-scope, scheme, and source. According to scope, we observe as the complexity of carbon accounting increases from direct emissions (Scope 1) to indirect emissions (Scope 2 and 3), the consistency of CCP data between third-party providers decreases. The higher levels of data consistency in Scopes 1 and 2 is most likely a consequence of investors' requests for more transparent disclosure in these areas (Sullivan & Gouldson, 2012). Thus, it seems essential to engage further in promoting and requesting firms to follow a standardized approach, even in the more complex Scope 3 realm (Matisoff et al., 2013).
According to reporting scheme, mandatory reporting schemes provide little benefit in generating higher consistency of CCP. Since these reporting schemes are based on facility-level data, they deliver an incomplete picture of companies' overall Scope 1 emissions. CCP data stemming from mandatory reporting schemes, including the USEPA GHGRP and EU ETS, cover between 33% and 49% of firms' Scope 1 carbon emissions. Additionally, CCP data stemming from voluntary reporting schemes do not improve data consistency. If firms report different values through different reporting channels, this creates further inconsistency. This is in line with previous research findings that the use of different data sources, that is, CDP and sustainability reports, leads to higher levels of inconsistency (Depoers, Jeanjean, & Jérôme, 2016;Perrault & Clark, 2010).
With respect to source, the consistency of third-party estimated data is lower when compared to data stemming from company reports. However, the combination of estimated data for Scope 1 and 2 emissions provides a surprising result of higher consistency. This outcome illustrates that the applied estimation methods can improve consistency to a certain extent. The biggest inconsistencies emerge for Scope 3 emissions-for both company reported as well as estimated data. More importantly, third-party providers do not provide transparent coverage of the categories of Scope 3 emissions. Thus, transparency of the Scope 3 sources will become important as more companies set reduction goals along supply chains according to science-based targets (Rekker, Humphrey, & O'Brien, 2019).
These findings reveal two points for further discussion: (a) the value of fostering consistency and (b) the significance of third-party estimation methods. Fostering CCP data consistency is important for several reasons. First, consistent CCP data are a necessary requirement for a range of stakeholders in order to measure, analyze, and compare CCP data between firms and over time. Achieving high levels of CCP data consistency should ultimately create a "level playing field between firms and across industries" (Bowen & Wittneben, 2011, p. 1029. Second, consistent CCP data are the foundation for thorough stress tests for banks and investors considering CCP data in investment decisions. For example, the TCFD (2017) has developed recommendations for climate-related financial disclosures for a better understanding of related financial risks. However, the TCFD (2017, p. 1) stresses the lack of consistent CCP data "hinders investors and others from considering climate-related issues in their asset valuation and allocation processes". Third, as more countries and regions implement carbon taxes and emission trading schemes, CCP data-notably beyond the already covered facility level data-will become more important from a financial perspective. Thus, fostering consistent CCP data covering all three emission scopes is essential for accurately assessing the business risks and opportunities in pathways toward decarbonization (Griffin et al., 2017;Stanny & Ely, 2008).
Third-party estimations play an important role in the CCP data context, as data providers fill in data gaps when companies do not or only partially report emission data themselves. This is especially the case for companies that are not listed on stock exchanges, covering a majority of small and medium-sized enterprises (Bradford & Fraser, 2008). As such, third-party estimations will continue to play a significant role. However, at the same time, the data based on estimations is another source for inconsistency. First, third-party estimation methods vary between and within each provider (e.g., Thomson Reuters applies one of three different estimation methods depending on data availability). This explains why the results of third-party estimations are rather inconsistent. Second, the underlying assumptions and estimation rules are not fully disclosed in a transparent manner. This explains why it remains challenging from an outsider's perspective to determine the reasons for the prevailing inconsistency in full detail. Notably, it remains not clear what data providers do and do not include in their Scope 3 estimations.
Most data providers use estimations based on input-output analyses. The advantage for third-party providers is that the data are publically available and user-friendly (Huang et al., 2009). However, the clear disadvantage is that results remains an estimate based on industry or sector averages, that is, you cannot detect the good performers from the laggards within any given sector (Minx et al., 2009). Process analysis could mitigate this issue; however, it requires a lot of company and process-specific information, and thus, it is likely to play rather a minor role for data providers who need to handle huge amounts of data points. In sum, third-party estimates make CCP data more comprehensive and wide-ranging, closing data gaps specifically regarding supply chains and small and medium-sized enterprises (Comas Martí & Seifert, 2013). It appears very likely that stakeholders will have to rely on third-party estimated data in the coming years. In terms of data consistency, we derive two implications: First, data providers should use a combination of the most reliable estimation methods-according to data availability. Second, providers should disclose in more detail how the estimations are conducted.

CONCLUSION
These results on the consistency of CCP data between third-party providers are relevant for future research, corporate management, financial markets, data providers as well as policymaking. First, academic researchers must be mindful of the consistency of CCP data for future research settings, as it can significantly affect the results. We would advise researchers to apply three practicable steps to increase the consistency of CCP data in future studies. First, extreme outliers should be examined to see if they are errors or actual outliers. For example, a company using renewable energy may have zero Scope 1 emissions-so the provided data are indeed an outlier but nevertheless correct and, thus, should be included. Second, scholars should repeat their analyses for data sets with and without outliers. The results show whether the outliers make a difference and to which scale they affect the results. The third step is to carefully select the source of CCP data, not just focusing on emission scopes, but also considering the differences of corporate-reported data and third-party estimation methods. If researchers are able to distinguish the two sources of data, we would suggest reporting the results in different models. For example, one model could cover for the entire sample and one model could cover only the company-reported data as a robustness check. Furthermore, future research could test the correlations of third-party estimated CCP data between the provider amounts and the maximum industry adjusted amount, also known as the Engaged Tracking amounts. The Engaged Tracking data (ET Carbon Dataset) has been used in several recent studies (Belkhir & Elmeligi, 2019;Jackson & Belkhir, 2018).
Second, corporate management should avoid situations where different CCP data are communicated or presented externally. Management should be very careful to check the consistencies in the CCP data, including the CDP and their own sustainability reports. The tolerated latitude of carbon accounting discourages consistent data are reflected in the inconsistencies between third-party data providers. Furthermore, companies should be aware of the release of CCP data based on third-party estimation methods, notably in the Scope 3 context. Managers could start gathering and disclosing Scope 3 emissions on their own as well as become more involved in improving standards for related estimation methods.
They should join in a wider discourse on transparency and clarity of CCP data, especially what are the most consistent approaches for all scopes.
Furthermore, with respect to Scope 3 emissions, it would be desirable to develop mutually agreed minimum standards. In sum, data consistency in Scope 3 would particularly benefit from a universally accepted standardization. Eventually, extending the scope of coverage to include small companies would also benefit from user-friendly accounting tools and reporting standards for these firms (Bradford & Fraser, 2008).
Third, our results point toward a challenge for a key player in the future decarbonization process, financial markets. On the one hand, CCP data are becoming increasingly relevant for investment appraisals (Depoers et al., 2016;Matsumura et al., 2014;Reid & Toffel, 2009). On the other hand, the consistency of data is not guaranteed, and could be hindering adequate analyses and assessments. Investors should engage firms to follow a standardized approach accordingly. Moreover, it is important to further raise the awareness in financial markets that capturing climate risks in investment portfolios must go beyond pure carbon footprints. Ex-post data about carbon performance on its own typically does not reveal any information for future risks. Further information about individual assets-such as their carbon dependency, decarbonizing options, and adaptation-related exposures-is required (Diaz-Rainey et al., 2017). Thus, carbon footprints are an essential step of a holistic climate risk management approach within investment appraisals.
Fourth, data providers should increase the transparency about their own estimation methods. We are aware that further standardization is not the most likely way forward regarding third-party estimation methods. It is well known that different estimation methods inherently yield different and, thus, inconsistent results. As our analysis shows, the correlation results for combined Scope 1 and 2 emissions for third-party estimated data do not diverge too greatly from reported data. As such, estimating CCP appears to be a promising way to close related data gaps. However, the data provider could help by increasing transparency of the details of their estimation methods. For example, data providers could indicate to which extend and in which cases they use sector-averages to infer firm-level emissions. Thus, investors can decide which third-party estimation methods they find most adequate and reliable. Eventually, the argument to move beyond publicly listed firms is also valid in the estimation context: phasing in estimations for firms other than large listed corporations will be an important step in order to pave the way for an economy-wide decarbonization.
Fifth, the findings of this article are also relevant for policymakers. For example, the recently released "Action Plan: Financing Sustainable Growth" of the EU Commission (2018a) seeks to reorient capital flows toward sustainable investment and manage financial risks stemming from climate change. It intends to develop sustainability benchmarks, which is only possible with a sound and standardized methodology to calculate carbon footprints. Furthermore, risks associated with climate change shall play an important role. All these ambitious efforts require high quality CCP data. This article illustrates some current shortcomings and severe challenges ahead.