Extracting ESG Scores for Quantitative Text Analysis

KCGS’ ESG standards

The Korea Institute of Corporate Governance and Sustainability (KCGS), an ESG evaluation agency, has been assessing ESG management across all industries, including financial and manufacturing sectors, since 2011. KCGS has consistently evaluated the ESG management practices of domestic companies for many years based on Korean ESG standards, offering the advantage of confirming the evaluation of ESG management elements for each specific sector(Lee and Rhee, 2020). KCGS' ESG best standards establish Korean standards for each ESG element by continuously reviewing overseas ESG evaluation methods. In particular, it is recognized as Korea's premier ESG evaluation institution, with the majority of domestic researchers conducting empirical analyses using KCGS's ESG rating scores(Huh, Kim, Lee and Lee, 2022; Lim, Cheung and Son, 2019).

In the case of another Korean-style K-ESG evaluation standard, 61 core criteria are presented by analyzing the evaluation items and disclosure standards most frequently covered by domestic and foreign rating agencies. However, the K-ESG guidelines primarily focus on regulatory responses rather than a comprehensive concept encompassing all ESG activities. As a result, they have not yet substantially provided evaluation scores and grades based on the K-ESG standards, and there have been no instances of their use in research. Therefore, in the Korean context, KCGS' ESG best practices are considered suitable as unstructured data for establishing an ESG dictionary.

The ESG best practice standards activities managed by KCGS can be summarized as follows: The best standards in the environmental field (E) primarily emphasize environmental management plans, environmental management principles, and areas of implementation that companies should focus on. Additionally, they provide detailed explanations of all activities related to environmental matters, encompassing environmental responses, along with standards (guidelines) aligned with environmental performance management.

The best standards in the social sector (S) offer guidance on corporate social responsibility activities by establishing benchmarks for social responsibility. These standards place a strong emphasis on stakeholders and categorize them into workers, partners, competitors, consumers, and the community. They address significant issues related to social responsibility management, aiming for shared growth through areas such as labor, human rights, fair operations, fair trade, sustainable consumption, information protection, and community involvement. The content reflects a multidimensional approach to social responsibility management as its foundation.

The governance (G) standards reflect global best practices. These standards, in particular, outline a corporate governance structure that promotes sustainability. They place a strong emphasis on the active role and responsibility for ESG management, with a central focus on the board of directors, and underscore the leadership's role and responsibility. From a governance perspective, these standards not only highlight aspects like risk assessment and management, compensation, and effective communication with shareholders and stakeholders but also encompass comprehensive governance activities for sustainable management, including information management and audit organizations.

Research method

The technique of quantifying ESG values derived from unstructured text is presented in <Figure 1>. Different organizations, such as KCGS and K-ESG, have developed standards for evaluating domestic ESG. Among them, KCGS stands out for disclosing and continually improving Korean-style ESG standards, offering exceptional interpretive value due to its extensive text data compared to other organizations' ESG activities. Therefore, in this study, we establish an ESG dictionary based on KCGS standards to extract time-series score data for corporate ESG management. Consequently, in this study, we establish an ESG dictionary based on KCGS standards to extract time-series score data for corporate ESG management activities. We collect ESG reports for each company and employ them as analysis data to derive numerical quantifications.

Specific methods for quantitative analysis using unstructured text are as follows. Firstly, KCGS' ESG best practices utilize natural language processing to remove special characters, stop words, formulas, and irrelevant language expressions like punctuation marks and URLs. The Korean language features 5 languages and 9 parts of speech, and due to its linguistic characteristic of combining particles in word forms, such as nouns, pronouns, rhetorical elements, verbs, and adjectives, word errors and bias phenomena can emerge when analyzed by morphology. Therefore, sophisticated preprocessing is required, and it needs to be composed of formal words(Kim, Jo, Choi and Chung, 2016). In this study, we employed a self-contained morpheme unit analyzer through MATLAB's Korean natural language analysis tool package to perform natural language processing, including particles and conjunctions.

Next, the process of building an ESG dictionary from KCGS' ESG standards is as follows. The algorithm typically used in word embedding modeling is the TF-IDF method. However, the TF-IDF method calculated words appearing in lower-order words higher in the inverse frequency weight calculation between documents, resulting in a few phrases being presented as keywords. On the other hand, the Word2Vec algorithm constructs words with related topics when extracting topic words, so it removes words that deviate from the topic and consists of words from major groups. In this respect, Word2Vec was ultimately selected as it was determined to be a relatively suitable modeling method for deriving key words for measuring ESG scores.

Word2Vec algorithm, which generates topic-related words and eliminates irrelevant ones. The selection of subject words based on hyperparameters greatly influences the algorithm's outcome, so experimentation to optimize these parameters is crucial. Therefore, experiments to determine optimization hyperparameters are necessary. To this end, experiments are conducted to find optimization parameters based on the minimum frequency of appearance, learning dimension vector size, window size, and number of learning iterations, which have a major influence on parameters. In this process, it is determined whether the keyword constitutes an appropriate topic or is optimized. By comparing the similarity between the original document, which is the correct answer set, and the derived keyword, the value at which the numerical value reaches the maximum is found.

By comparing the similarity between the original document and derived keywords, we determine the appropriate topic and optimization level. The final ESG keyword dictionary is created by applying optimal hyperparameters and selecting 300 identical dimensions for each dictionary selected. Here, because the dimension sizes of different unstructured texts may cause deviations due to size effects in similarity measurement, each dictionary selected subject words with 300 identical dimensions. This completes the derivation of the final ESG keyword dictionary using KCGS data.

Another step involves constructing ESG reports for individual companies. These reports are obtained from the public data room of most domestic KOSPI-listed companies and undergo the same natural language preprocessing as the ESG dictionary. Through optimization experiments for each report, we create an individual ESG report, and the time-series ESG score of each company is calculated by analyzing the similarity between documents. As examples, Samsung Electronics in the manufacturing sector and Kookmin Bank in the financial sector were used to extract their respective ESG scores. This process is repeated ten times to gather comprehensive data.

Reference

Lee, J. G. and Rhee, J. H. (2020),“Current Status and Future Directions of Research on “Sustainable Management”: Focusing on the ESG Measurement Index,” Journal of Strategic Management, 23(2), 65-92 (In Korean).

Lim, H. J. (2021),“Analysis on ESG Issues of SMEs Using Text Mining,” The Journal of Humanities and Social science, 12(4), 469-482 (In Korean).

Kim, J. Y., Jo, W. Y., Choi, J. H. and Chung, Y. I. (2016),“Linking Findings from Text Analyses to Online Sales Strategies,” Journal of The Korean Operations Research and Management Science Society, 41(2), 81-100 (In Korean).

Huh, J. H., Kim, T. M., Lee, S. R. and Lee, C. L. (2022),“The Relationship between ESG and Brand Equity : A Mixed Method Approach,” Journal of Business Research, 37, 1-20 (In Korean).