Bayesian Model Averaging for Identifying the Robust Determinants of Finland's Production‐Based Ecological Footprint

This study employs Bayesian regression and Bayesian Model Averaging (BMA) to analyze the determinants of the production‐based ecological footprint (EF) of Finland during 1990–2022. Using Hamiltonian Monte Carlo estimation and enumeration of the 256‐model space with the Bayesian Adaptive Sampling algorithm, the analysis accounts for parameter uncertainty and nonlinearity among the predictors (GDP and its square, renewable energy, foreign direct investment, urbanization, nuclear energy, and innovation), together with an FDI × urbanization interaction. To ensure the findings do not hinge on an arbitrary prior, inclusion probabilities are assessed across prior specifications. Three predictors are robust under every specification: renewable energy (negatively associated with EF), foreign direct investment (positively associated), and the FDI × urbanization interaction (negative), implying that the footprint‐raising effect of inward investment is attenuated in more urbanized settings. Support for an Environmental Kuznets Curve is weak once model uncertainty is accounted for: neither GDP nor its square is a robust driver. An out‐of‐sample comparison shows that BMA delivers better‐calibrated predictions than any single selected model, and information criteria (LOOIC preferred over WAIC, with reliable Pareto‐ k diagnostics) confirm model adequacy. The paper provides a robust framework for inference under model uncertainty and clarifies the production‐side drivers of Finland's environmental pressure.