Natural Resource Depletion, Trade Flows, and CO ₂ Emissions: A Panel Data Econometrics and Machine Learning Analysis of the Top Three Economies

ABSTRACT

This study examines the connections between trade flows, the depletion of natural resources, and total emissions(kt) in China, India, and the United States between 1992 and 2022 using a combination of panel econometric and machine‐learning methodologies. While panel econometric models like Pooled OLS, Fixed Effects, and Random Effects were used to evaluate structural and country‐specific relationships, machine‐learning models like Artificial Neural Network (ANN), Random Forest (RF), and Long Short‐Term Memory (LSTM) were used to capture nonlinear and temporal patterns in annual emission dynamics. The Fixed Effects specification demonstrated the greatest explanatory power among the econometric models. Under the chosen validation scheme, LSTM outperformed the other machine‐learning models in terms of predicted errors. The forecasting results are taken cautiously due to the restricted annual time‐series sample, and out‐of‐sample testing and chronological validation were used to lower the danger of overfitting. Although the policy implications should be interpreted cautiously due to the aggregate character of the data and the lack of direct policy‐effect assessment, when combined, the two methodologies offer complementary evidence for understanding emissions patterns and for guiding policy discussion. The policy interpretation of machine‐learning models is cautious and is meant to supplement rather than replace the structural evidence from the econometric analysis because they are less transparent than econometric specifications. To guarantee comparability between the econometric and forecasting components, all variables were harmonized to uniform annual definitions and units.