Machine Learning in Green Bond Pricing

Abstract

The pricing of green bonds has emerged as a critical issue of international scholarly and policy interest amid the rapid expansion of the global green bond market. Existing empirical research, however, remains constrained by limited sample coverage and superficial model comparisons, leaving substantial gaps in the understanding of the key determinants of green bond yields. This study examines a comprehensive sample of 7,062 Chinese green bonds issued between 2016 and 2024, employing three machine learning algorithms—Random Forest (RF), XGBoost, and Support Vector Machine (SVM)—to predict green bond coupon rates. The empirical results indicate that XGBoost achieves the strongest predictive performance, followed closely by Random Forest, whereas SVM exhibits comparatively weaker performance. Feature importance analysis reveals that macroeconomic factors—particularly year-on-year GDP growth rate, year-on-year M2 growth rate, and year-on-year CPI growth rate—constitute the most critical determinants of green bond pricing. Extensive robustness tests, encompassing alternative train-test split ratios and cross-validation settings, confirm that these findings remain stable across different experimental configurations. This study offers timely theoretical support and practical guidance for investors and regulatory authorities in green bond market decision-making.