Green Bond Issuance and Carbon Emissions: Can Causal Machine Learning Inform Forward-Looking Policy Decisions?

Green bonds finance projects intended to deliver environmental benefits, including reductions in greenhouse gas emissions. However, evidence that municipal green bond issuance lowers local carbon emissions remains limited and lacks the spatial and temporal granularity that municipalities need to make informed decisions. This study investigates the impact of U.S. municipal green bonds issued between 2009 and 2019 on carbon emissions while accounting for the confounding effects of local socioeconomic conditions. We employ a causal forest model within a double-biased machine learning framework to estimate the effects of issuance volume on emission trends. The analysis reveals substantial spatial and temporal heterogeneity: the largest reductions occur in specific counties and increase over time, with an implied abatement cost of $192 per ton of CO₂. Counties with higher concentrations of small and medium enterprises exhibit the strongest benefits. Comparisons between counties with and without prior issuance suggest that new green bonds generally reduce emissions, although magnitudes vary with local economic and infrastructure characteristics. These findings provide evidence on the effectiveness of municipal green finance, with implications for policy implementation and forecasting environmental outcomes, particularly for first-time issuers.