A carregar...
Publicação
Opening the Black Box of Sovereign ESG Ratings: Transparency, Explainability, and the Challenges of Data-Driven Indexes
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 7.28 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Despite the systemic importance of sovereign Environmental, Social, and Governance (ESG) ratings in global capital markets, their credibility is frequently undermined by the opaque, "black box" nature of proprietary methodologies, which limits transparency and interpretability. This study addresses this challenge by developing a transparent and replicable framework, supported by explainable machine learning (XAI), to construct and validate a sovereign ESG index for 37 European countries over the 2012–2020 period. Drawing on 52 indicators from publicly available World Bank data, the proposed approach integrates Principal Component Analysis (PCA), a multi-algorithm clustering strategy (KMeans, Agglomerative, Fuzzy C-Means, and DBSCAN), and a multi-criteria evaluation framework to construct and compare alternative composite ESG index specifications, identifying the most robust weighting scheme. The resulting composite index is then validated using XGBoost regressors and SHAP (SHapley Additive exPlanations), enabling a detailed interpretation of the drivers underlying ESG scores. The results show that the Rule of Law and Poverty Headcount ratios emerge as the dominant drivers of governance and social dimensions, respectively. However, the analysis also reveals important challenges associated with data-driven index construction. In particular, a fundamental tension emerges between statistical optimization and material policy relevance, particularly within the Environmental pillar. Specifically, the data-driven weighting scheme tends to reward structural underdevelopment while penalizing advanced economies undergoing active energy transitions, highlighting a "statistical vs. material" dilemma. By opening the black box of sovereign ESG ratings, this study demonstrates that methodological transparency, combined with XAI-based validation, not only enhances interpretability but also exposes inherent biases in data-driven approaches. The proposed framework enhances the credibility of sovereign ESG assessments and provides a foundation for more informed investment decisions and evidence-based public policy design.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Sovereign ESG Ratings ESG Transparency Composite Index Clustering Analysis Explainable Machine Learning Index Validation