CO 2 emissions, financial development, and economic growth in Nigeria: small-sample time-series evidence and an exploratory machine-learning stress test

This study examines Nigeria’s carbon dioxide emissions per capita over 1986–2024 using 39 annual observations, linear and nonlinear autoregressive distributed lag models, predictive-precedence tests, and machine learning. Cointegration tests support long-run co-movement, but individual coefficients are sensitive to lag restrictions and regime controls. Electricity consumption provides the most persistent association; income curvature and financial and exchange-rate effects remain fragile. Unadjusted machine-learning models fail to outperform a naïve forecast. A post-2015 indicator improves some specifications without establishing a permanent break or decoupling. The findings emphasize small-sample uncertainty and conditional, rather than causal, policy implications.