Distributional heterogeneity in the effect of green innovation on carbon emissions across G20 economies

This study investigates whether green innovation reduces carbon dioxide emissions in G20 economies over 1985–2024. It develops a diagnosis-driven empirical strategy in which estimator selection follows formal tests for cross-sectional dependence, slope heterogeneity, integration, and cointegration. The results reveal strong cross-sectional dependence, reject slope homogeneity, and show that heterogeneity is driven largely by a small number of influential economies, particularly India and Türkiye. Long-run relationships are estimated using the mean-group estimator and the outlier-robust trimmed mean-group estimator. Conventional mean-group estimates are largely statistically insignificant, whereas the trimmed estimator delivers precise estimates of all five long-run coefficients, highlighting the importance of outlier-robust aggregation in heterogeneous panels. The preferred results trace an inverted-U income–emissions profile. A one-percentage-point increase in the renewable-energy share is associated with a 0.032-ton long-run decline in per-capita emissions, and a doubling of renewable-energy patenting with a 0.051-ton decline. Panel quantile regression with common shocks shows that the innovation effect is distribution-dependent. The effect is statistically indistinguishable from zero in low-emission conditional states, and it strengthens monotonically toward the upper tail. At the 90th percentile, a doubling of patenting is associated with a reduction of roughly 0.065 tons. These findings identify renewable-energy deployment as a distribution-wide decarbonisation lever and green innovation as a targeted one, operating where emissions are highest.