Paper Type
Short
Paper Number
PACIS2025-1149
Description
The divergence in ESG ratings across different agencies poses certain challenges for investors and companies, undermining the credibility and utility of sustainability assessments. This paper proposes a multi-view learning framework integrating data from five major rating agencies to address this inconsistency. Utilizing Multi-View Autoencoders and Canonical Correlation Analysis, the model maps heterogeneous ESG data into a unified latent space, preserving agency-specific features while reducing noise and discrepancy. Validation through Spearman correlation analysis demonstrates strong alignment between the unified ESG score and individual ESG ratings, indicating an effective integration. Based on random forest regression models, the test of different ESG ratings’ capability for predicting stock returns reveal a 14% increase in R² with the unified ESG score, highlighting its superior explanatory power. The framework enhances ESG rating consistency and reliability, offering theoretical advancements in multi-view learning applications and practical benefits for stakeholders in sustainable investment and decision-making.
Recommended Citation
Li, Xinqi; Ma, Jiyuan; Guo, Qiaozhen; Cai, Xiaofang; Tang, Xin; and Huang, YJ, "Integrating Divergent ESG Ratings through Multi-view Learning: A New Framework for a Unified ESG Rating" (2025). PACIS 2025 Proceedings. 1.
https://aisel.aisnet.org/pacis2025/dig_plat/di/1
Integrating Divergent ESG Ratings through Multi-view Learning: A New Framework for a Unified ESG Rating
The divergence in ESG ratings across different agencies poses certain challenges for investors and companies, undermining the credibility and utility of sustainability assessments. This paper proposes a multi-view learning framework integrating data from five major rating agencies to address this inconsistency. Utilizing Multi-View Autoencoders and Canonical Correlation Analysis, the model maps heterogeneous ESG data into a unified latent space, preserving agency-specific features while reducing noise and discrepancy. Validation through Spearman correlation analysis demonstrates strong alignment between the unified ESG score and individual ESG ratings, indicating an effective integration. Based on random forest regression models, the test of different ESG ratings’ capability for predicting stock returns reveal a 14% increase in R² with the unified ESG score, highlighting its superior explanatory power. The framework enhances ESG rating consistency and reliability, offering theoretical advancements in multi-view learning applications and practical benefits for stakeholders in sustainable investment and decision-making.
Comments
Methods