Green logistics, transport decarbonization, and sustainable development: Evidence from emerging economies with a focus on Azerbaijan

This study investigates whether green logistics capacity is associated with transport-sector decarbonization in developing economies. Using annual panel data for ten countries over 2010–2021, the paper constructs a Green Logistics Index (GLI) from transport-service capacity, digital readiness, renewable-energy orientation, and reverse-normalized energy intensity. The preferred empirical specification uses country and year fixed effects and is complemented by robust covariance estimators, Driscoll–Kraay inference, common-correlated-effects checks, slope-heterogeneity diagnostics, stationarity and cointegration tests, and index-construction sensitivity analysis. The baseline equal-weighted GLI is negatively and significantly associated with per capita transport CO₂ emissions: a 0.10-point increase is associated with approximately 13.8% lower emissions. The result is substantively stable under PCA and entropy weighting, with corresponding exact semi-elasticities of −14.4% and −13.2%. The entropy coefficient remains significant under conventional TWFE and cross-section SUR/PCSE inference and marginally significant under clustered inference, indicating stable economic magnitude but covariance-sensitive precision. Trade openness and urbanization remain positively associated with transport emissions, while the EKC-type specification is unsupported. The combined evidence shows that the main result is not an artifact of equal weighting, although the observational design warrants interpretation as a conditional association rather than a causal effect. For Azerbaijan, the findings imply that Middle Corridor expansion should be paired with renewable-powered logistics infrastructure, energy-efficiency improvements, operational digitalization, and modal-shift policies. A leave-one-country-out analysis further shows that the GLI coefficient remains negative after every country is sequentially excluded; conventional TWFE supports the coefficient at the 5% level or better in nine cases and at the 10% level in the remaining case, while PCSE inference preserves conventional or marginal support in nine of ten exclusions. Endogeneity-oriented checks preserve a negative GLI coefficient in all five specifications: the association remains significant after excluding 2020–2021 and after absorbing country-specific linear trends, while lagged and first-difference estimates retain the expected sign but lose precision under PCSE. A significant one-year lead-placebo coefficient prevents causal interpretation and indicates that temporal persistence or residual time-varying confounding cannot be ruled out.