Evaluating the scale-dependent impact of environmental ratings on green credit using machine learning and threshold regression

Abstract This study examines the association of environmental (E) ratings and green credit allocation among Chinese commercial banks using a panel dataset from 2010 to 2021. Initial linear models unexpectedly reveal that E ratings do not show a statistically significant unconditional association with green credit balances (GCB) once bank-level heterogeneity is rigorously controlled. To unravel this, we systematically integrate machine learning methods with econometric threshold analysis. A Hansen panel threshold regression identifies a critical institutional scale threshold (total loan volume). Crucially, E ratings exhibit a positive and marginally significant association with GCB only for banks exceeding this scale, while the association remains indistinguishable from zero for smaller banks. This provides the key to interpreting the linear model results: the association between environmental performance and green credit allocation appears strictly conditional on a minimum scale. Further analysis confirms a “scale-ratio paradox”, where higher E ratings increase the absolute volume of green credit for large banks but fail to improve, and may even reduce, its portfolio share (GCR). These findings support a risk-signaling rather than a pure efficiency-enhancement narrative, underscoring the need for differentiated, scale-contingent policies.