ESG controversies and investor trading behavior in the Korean market

https://doi.org/10.1016/j.frl.2023.103750Get rights and content

Highlights

  • We use deep natural language processing to analyze textual news on ESG controversies.

  • ESG controversies significantly increase investors’ trading activities.

  • Domestic institutions sell stocks with controversies.

Abstract

This study examines how investor trading behavior changes following environmental, social, and governance (ESG) controversies by analyzing textual news data. We use deep-learning-based natural language processing to classify news articles into specific categories of controversy. ESG controversies generally increase investors’ trading activities regardless of their type, while their reactions differ by ESG pillar. Interestingly, domestic institutions tend to sell stocks with controversies.

Introduction

Environmental, social, and governance (ESG) activities receive considerable attention globally (Gillan et al., 2021; Kim et al., 2017). The importance of ESG is particularly stressed in Korea, which exhibits a leading emerging economy and financial market but faces various ESG-relevant concerns. For example, Korea suffers from severe air pollution, such as fine dust, and its carbon emissions remain consistently higher than the OECD average.1 In addition, various social issues, such as labor rights, worker safety, and monopolies, have emerged in the pursuit of rapid economic growth. Moreover, the Korean economy comprises unique governance structures that may potentially intensify the influence of governance issues, such as Chaebol. Hence, firms’ ESG performance is constantly recognized by market participants, investors, citizens, media, regulators, and entrepreneurs in Korea (Park et al., 2022; Ryu et al., 2016, 2017).

For firms engaging in ESG activities, avoiding ESG controversies is as important as improving ESG performance (Lin-Hi and Müller, 2013). ESG controversies are closely related to corporate social irresponsibility, which indicates corporate activities that conflict with stakeholders’ rights (Strike et al., 2006). Frequent ESG controversies about a firm may deteriorate its viability and trust and even affect its ESG activities (Aouadi and Marsat, 2018; DasGupta, 2022).

While most previous studies have analyzed the impact of corporate ESG activities on their performance or stock prices using low-frequency data, the effect of ESG-related news releases has only been studied recently. Krüger (2015) finds that stock prices react negatively to corporate social responsibility events, even when the news content is positive. However, the reaction is stronger to negative news. Capelle-Blancard and Petit (2019) claim that firm values deteriorate after negative news but do not significantly improve after positive news. In contrast, Serafeim and Yoon (2022) contend that stock prices increase with positive ESG news, but they do not significantly respond to negative ones, except those with high news coverage or social capital issues.

This letter examines investors’ reactions to ESG controversies by analyzing the abnormal trading behavior of different types of investors: domestic individuals, domestic institutions, and foreigners. We distinguish investor types based on our high-quality dataset and analyze whether various types of investors have different expectations for ESG controversies, providing further implications in addition to analyzing aggregate market reactions. An ESG controversy may affect investors’ trading activities if investors perceive it as a determinant of a company's fundamentals or an opportunity to earn profits. We identify specific categories of ESG controversies by classifying news articles using KoBERT, a deep-learning-based natural language processing (NLP) model. Using an event-study approach, we further analyze abnormal trading volumes for firms facing ESG controversies, depending on investor types and ESG pillars.

Section snippets

ESG controversy classification and data

We construct a classification for ESG controversies based on those provided by Refinitiv and Morgan Stanley Capital International. Our classification of ESG controversies consists of three pillars (i.e., environmental, social, and governance), nine main categories, and 19 subcategories.2

Bidirectional Encoder Representations from Transformers (BERT) is one of the most popular and recent NLP models (

Empirical analysis

To investigate investors’ trading behavior around ESG controversies, we use the following trading activity proxies: abnormal trading value (AV), turnover ratio (TO), and buy-sell imbalance (BSI). We winsorize all variables at the 1st and 99th percentiles.

First, AV for firm i on event day t (i.e., AVi,t) is measured as follows:AVi,t=Vali,tValiValm,tValm,where Vali,t and Valm,t denote the trading value of firm i and the aggregate trading value in the KOSPI market, respectively, on event day t.

Conclusion

This study investigates whether investors’ trading activities change following news articles revealing ESG controversies, providing contributions to the literature on investors’ perceptions of ESG controversies. We find abnormal trading volumes on days with ESG controversies, supporting the idea that ESG controversies affect investors’ decisions. We also find that the signs and significance of abnormal trading volumes vary according to the ESG pillar. Institutional investors, in particular,

CRediT authorship contribution statement

Jeongseok Bang: Methodology, Writing – review & editing, Methodology, Software, Data curation, Formal analysis, Visualization. Doojin Ryu: Methodology, Conceptualization, Investigation, Resources, Validation, Writing – review & editing, Supervision, Project administration, Funding acquisition. Jinyoung Yu: Writing – review & editing, Investigation.

Declaration of Competing Interest

There is no conflict of interest.

Acknowledgement

This work is supported by IREC, The Institute of Finance and Banking, Seoul National University (The project title is “ESG management, investments, and controversies: based on machine learning and big data”).

References (18)

There are more references available in the full text version of this article.

Cited by (12)

View all citing articles on Scopus
View full text