Testing the Mitigating Effects and Mechanisms of Green Finance on Carbon Emissions: A Dynamic Panel Model and LSTM Causal Analysis Approach

Abstract

Based on panel data from China’s provinces covering the period 2006 to 2022, this study integrates econometric and machine learning methods to examine the impact of green finance on carbon emissions, its mechanisms of action and heterogeneous characteristics. We first construct a comprehensive green finance index and then establish a dynamic panel GMM model, finding that the effects of green finance on carbon emissions exhibit time lags and sectoral differences. In high-carbon industries and in central and western regions, green finance may temporarily drive up emissions in the short term by supporting capacity renewal or ‘greenwashing’ behaviour; however, in the medium to long term, it achieves net emission reductions through two intermediary channels: industrial structure optimisation and green technological innovation. This long-term mitigating effect is more pronounced in eastern regions. To further examine possible non-linear responses, the study also employs a multi-layer perceptron model and conducts a counterfactual simulation by increasing the green finance index while holding other covariates unchanged. The simulation shows that the response of carbon emissions differs across provinces. In several high-carbon provinces, a higher level of green finance still coexists with relatively high emissions in the short run. This result indicates that green finance does not work as an immediate emission-reduction instrument. Its effectiveness depends on the quality of financed projects, local regulatory capacity, and the coordination between financial policy and industrial policy. The econometric and machine-learning results therefore point to a common implication: green finance can support China’s dual-carbon goals, but its effect follows a staged and region-specific process. Early-stage emission rebounds should not be ignored, and policy design needs to distinguish between regions, sectors, and different phases of transition.