This study investigates the performance of sustainable global equity ETFs by applying returns-based style analysis (RBSA) enhanced with machine learning techniques, focusing on the investment styles driven by Environmental, Social, and Governance (ESG) factors.

The conventional Sharpe's style analysis method, which regresses ETF returns on a set of benchmark indices to determine exposure to various factors and investment styles, is expanded in this study by incorporating a Gradient Boosting Regressor (GBR) model. This approach addresses key limitations in traditional RBSA, including the inability to handle high multicollinearity among factors and the lack of statistical significance assessment of style weights. The GBR method, recognized for its robustness in managing non-linear relationships and multicollinearity, provides a more robust evaluation of the ETFs' exposure to the distinct ESG factors. The paper also introduces bootstrapped confidence intervals to assess the stability and significance of the ESG factor contributions on the ETFs returns. By generating multiple bootstrap samples, the confidence intervals offer insights into the variability and reliability of the estimated factor exposures.

Key findings reveal that the Environmental factor consistently exhibits the highest contribution to ETF returns, reflecting the significant market prioritization of environmental considerations in ESG investing. Conversely, the Governance factor, although impactful in specific cases, generally shows lower and more variable contributions. The Social factor's influence varies widely, demonstrating inconsistent importance across the ETFs.

The study's innovative integration of machine learning techniques with traditional RBSA provides a deeper understanding of ESG-driven investment strategies, enhancing the explanatory power of the model and offering investors more robust insights into the performance dynamics of sustainable ETFs. This approach not only broadens the methodological toolkit for style analysis but also underscores the critical role of advanced statistical methods in refining investment performance evaluation in the context of ESG-focused portfolios.