Decoding Energy Consumption in the ESG Era: Panel Data Evidence and Machine Learning Insights

This research evaluates how energy consumption per capita (ENUS) is affected by Environmental, Social, and Governance (ESG) factors, with particular focus on the Environmental pillar and its relationships with energy systems. The key research question is whether and how indicators related to ESG provide additional explanatory power for capturing variations in ENUS across countries and time, and whether machine learning techniques offer novel ways to address this problem beyond traditional panel econometric models. To this end, this research aims to combine panel econometric models and machine learning procedures, using a large dataset from the World Bank that provides internationally comparable data for approximately 161 countries during the period 2004-2023. From a methodological standpoint, this research will consist of three key steps. In a First Step, a sequence of fixed-effect, random-effect, and weighted least-squares models will be applied, with a specific focus on identifying how a large set of ESG-related variables relate to ENUS in a structural equation, controlling for country-specific unobserved heterogeneity. As a second step, this research will explore a sequence of clustering procedures, with a specific interest in identifying a number of regimes across which countries systematically co-pattern energy use with emissions, climate, and natural resources in a shared, multidimensional setting. In a third and final step, this research will evaluate a set of machine learning techniques through a sequence of assessments, with a specific focus on the K-Nearest Neighbor algorithm. It will identify that this technique is one of the most accurate models across a set of normalized criteria, such as accuracy and fit parameters. Model explanation will be improved through dropout loss and additive explanations, consistently assessing individual ESG variable weights in describing energy use. The analysis provides novel evidence that a strong and complex multideterminant pattern defines the relationships between ESG factors and energy use, with a strong influence from environmentally driven indicators, emissions intensity, energy efficiency, and natural resource usage, and with a complex interplay between social and governance pillar variables in driving energy consumption through development, institution, and structure variables.