Carbon Emissions and TCFD Aligned Climate-Related Information Disclosures

Alignments with other frameworks can be assessed at: https://​www.​tcfdhub.​org/​ignment/​

Computerized textual analysis refers to the textual analysis technique which uses computing power to extract relevant texts from big data and convert the qualitative information into climate disclosure scores.

Specifically, the U.S. SEC issued the Commission Guidance Regarding Disclosure Related to Climate Change in 2010, which provides interpretive guidelines related to climate change for firms under the existing SEC disclosure requirements. UK is considered as highly polluting nation with severe environmental issues (Giannarakis et al., 2017; Liao et al., 2015). The UK government made updates to The Company Act Regulation 2013 and 2016 that listing firms are mandated to reveal nonfinancial information on future development and performance. The Financial Conduct Authority in UK also released a proposal for certain listed firms to state whether they made disclosures consistent with the TCFD recommendations in their annual financial reports. The Australian and Canadian governments have weighed in with new guidance on climate change disclosure in alignment with TCFD reporting framework. Therefore, firms from these four countries are pioneers in taking initiatives to respond to climate change issues.

For example, in line with the Paris Agreement, the UK and other EU members jointly acted to reduce GHG emissions by at least 40% by 2030 and at least 80% by 2050 relative to the 1990 level. Canada and Australia set a target of 30% and 28% GHG emission reduction by 2030 while the US agreed to a 28% target of reduction by 2025 compared to the 2005 level. Hence, among the developed countries, these four countries have made significant commitments to reduce GHG emissions.

Firms that do not disclose climate information (i.e., zero CCSS, see Climate-related Information Disclosure Measure: Step 2) are filtered out from the sample in a year.

The GHG Protocol defines Scope 1 as direct GHG emission from source owned or controlled by the company; Scope 2 as indirect GHG emission from the generation of purchased electricity consumed by the firm.

The number of emission types included in Scope 3 have maintained at constant 17 types since 2013 according to CDP database, as described: 1. Purchased goods and services; 2. Capital goods; 3. Fuel-and-energy-related activities (not included in Scope 1 or 2); 4. Upstream transportation and distribution; 5. Waste generated in operations; 6. Business travel; 7. Employee commuting; 8. Upstream leased assets; 9. Downstream transportation and distribution; 10. Processing of sold products; 11. Use of sold products; 12. End of life treatment of sold products; 13. Downstream leased assets; 14. Franchises; 15. Investments; 16. Other (upstream); 17. Other (downstream).

Alternatively, we assume firms without Scope 3 emission data (i.e., not reporting) have zero emission in a year. We calculate carbon emission as summation of Scope 1, 2 and 3 and redo all regressions in our paper. We find that our results do not alter. These results are not reported but will be available upon request.

We use annual reports, sustainability reports and stand-alone climate change reports to construct an enriched vocabulary. However, in each country sample, firm-level CCSS is estimated using only annual reports. The International Integrated Reporting Council (hereafter IIRC) encourages firms to adopt integrated reporting, which aims to present financial reporting and sustainability reporting in annual reports. More specifically, IIRC (2019a, 2019b) reports approximately 50% of the 200 largest Australian listed firms leverage the principles of integrated reporting in annual reports by including more sustainability information. Additionally, ASIC (2018) constructed a review of reports of Australian listed firms, including annual reports, sustainability reports and climate change reports and found over 40% of reports of ASX300 firms contain climate change information between 2011 and 2017, which is consistent with our findings. Moreover, the Companies Act 2006 Regulations 2013 mandates UK listed firms to issue a strategic report disclosing information on environmental matters and sustainability when material in their annual reports. We find supporting evidence of over 90% of firms in UK sample report climate change information in their annual (or integrated) reports. Our results also show that on average, 50% of firms in Canada disclose climate-related information in the annual reports, with the figure being 40% in US firms between 2010 and 2018. Therefore, it suggests to a certain extent an increasing proportion of annual reports are being integrated to incorporate more sustainability information. Overall, excluding sustainability reports might have little effect on the issue that we possibly omit the firms reporting climate change information only in sustainability reports. However, we also acknowledge this as a limitation that we do not include sustainability reports or stand-alone climate change reports in our sample.

It is not difficult for information extraction since TCFD supporters tend to disclose climate change information in line with four categories under a separate and structured section. In this study, we copy all textual climate-related information under each category and save them in the word files for text processing in the next step.

We set n to two and three as suggested by Gentzkow et al. (2019), which produces pairs of two and three consecutive words.

We consider the tokenized words by bigrams used to differentiate firms whether they report climate-related information or not. Compared with single-word terms that are insufficient to capture the textual patterns, the n-gram modeling can generate richer representations for climate change information and avoid capturing noised terms, such as ‘investment climate’ that are irrelevant to climate change information.

tf is calculated as the number of a word’s occurrence in a document; idf is a measurement of whether a word is common or not across all documents, calculated as the logarithm of the ratio of total documents to the number of documents containing a given word. f_(t,d) is the occurrence of term t in a document; N is the number of total documents; n_t is the number of documents containing term t.

\(Similarity score=cosine\left({tf\_idf}_{CCV},{tf\_idf}_{firm,i}\right)=\frac{\left({tf\_idf}_{CCV,j}\right)\left({tf\_idf}_{firm,i,j}\right)}{\left|{tf\_idf}_{CCV,j}\right|\left|{tf\_idf}_{firm,i,j}\right|},\) where, \({tf\_idf}_{CCV,j}\) is the \(tf\_idf\) of term j in CCV; \({tf\_idf}_{firm,i,j}\) is the \(tf\_idf\) of term j in firm \(i\)’s annual report. Hence, it ranges from zero to one.

FOG = 0.4 * [average sentence length + 100 * (the number of words with 3 syllables or more / the number of words)]; SMOG = 1.043 * sqrt(30*the number of words with 3 syllables or more / the number of sentence) + 3.1291.

We also use the carbon emission intensity variable of Luo (2019a, 2019b) for robustness, calculated as natural logarithm of the ratio of total emission to firm sales. Compared to other two carbon emission measures, using scaled carbon emission by sales does not alter our key finding regarding the relation between climate-related information disclosure and carbon emissions. The regression results are not reported in the paper but are available upon request.

We control for the effect of Emission Trading Scheme (ETS) by replacing country fixed effect with a ETS dummy variable in the regression model. We find that ETS has positive effect on climate change disclosure but presence of ETS dummy variable does not alter our regression results. The regression results are not reported in the manuscript but will be available upon request.

This finding is not conflicted with our main results since environmental performance, proxied by ENVSCORE from Datastream indicates relative rating of a company based on reported environmental information. It evaluates the effects of company activities regarding resource utilization on overall environment including air, land, and water usage.

We run additional regressions using one-year lagged carbon emission variables and the models from Tables 8, 9. The results obtained using one-year lagged carbon emission variables are highly consistent with the results obtained using contemporaneous carbon emission variables. The additional results are not presented in this manuscript and will be available upon request.

See Fig. 2 for carbon emissions in the four countries.

We also replace TCO2 with ADJTCO2 and rerun all regressions, presenting similar results. The additional results are not presented in this manuscript and will be available upon request.

Heckman stage 1 test results are not reported in this manuscript, but they are available upon request. Inverse Mills ratios (IMR) are estimated in Heckman stage 1 test and used as an independent variable in Heckman stage 2 test. A significant IMR indicates significant self-selection bias in the sample.

There is a distinction between Supporter dummy in Table 13 and SUPPORT dummy in Appendix 3. SUPPORT equals to 1 if the firm is a TCFD supporter in year t since SUPPORT dummy depends on the date when the firm pledges support for TCFD. For example, if the firm expresses their support in 2017-year, SUPPORT dummy will equal to 1 in 2017 and 2018 years only whereas Supporter dummy equals to 1 in all years between 2010 and 2018. Creating Supporter dummy for the difference-in-difference test is to avoid the technical issue that Supporter would be the same as the interaction variable, Supporter* Event, in the difference-in-difference test.