Can machine learning bring ESG ratings closer to small and medium-sized enterprises?
Can machine learning bring ESG ratings closer to small and medium-sized enterprises?

Summary/Abstract: Research background: Environmental, Social, and Governance (ESG) principles provide an important framework for companies to create value for stakeholders. Companies aim to enhance their performance in ESG as it has gained importance in investment analysis, with ESG ratings often used for this purpose. However, there is no objective way to calculate ESG scores, and small and medium-sized enterprises (SMEs) struggle to access ratings provided by score providers. Purpose of the article: The main goal of this paper is to investigate whether the application of an artificial neural network (ANN) and feature selection techniques can make it possible to identify and prioritize the key features affecting companies' ESG scores. Determining a set of key features from among the wide range of non-financial data reported by companies and then used by rating ESG score providers would benefit SMEs. It would allow them to report the most relevant non-financial data, that is considered in calculating ESG scores. Methods: A feedforward ANN was employed to predict corporate ESG ratings based on environmental, social and governance data reported by companies to Refinitiv. Specifically, ESG data from 1,194 companies was analysed in this study across seven diverse sectors: Banking Services, Diversified Retail, Telecommunications Services, Metals and Mining, Oil and Gas, Software and IT Services, and Specialty Retailers. The companies represented 61 countries from six continents. The data encompassed the period from 2017 to 2021. Sectoral representation varied, with company numbers ranging from 74 to 260. A sequential forward feature selection process was next implemented to identify the minimal feature subset of 186 parameters used by Refinitiv for accurate ESG score prediction. The ANN model was trained iteratively to predict the ESG score, starting with an empty feature set and progressively adding features that most enhanced the model's performance. The experiment was performed repeatedly for each sector. Findings & value added: This paper proposes using machine learning (ML) to bring ESG ratings closer to SMEs. The study shows that ANN can accurately predict ESG scores using Refinitiv data, while retaining ESG ranking consistency. Furthermore, feature selection can infer ESG scores with acceptable accuracy using a minimal subset of key company attributes. Specifically, a neural network can accurately predict a company's ESG score and determine its ranking using only around 13% of reported parameters. This approach may simplify access to ESG assessment for SMEs, allowing them to evaluate their performance with fewer parameters.

• Details
• Contents