Disentangling the determinants of household energy expenditure: A quantile regression approach with machine learning

Abstract

Household energy consumption is a key driver of national energy demand and emissions. Understanding household behaviour is therefore essential for designing effective and socially sensitive energy policies. This study investigates the determinants of household energy expenditure in T & uuml;rkiye using microdata from the 2023 Household Budget Survey. Variable selection was conducted with machine learning algorithms, and quantile regression was applied to capture heterogeneity across different expenditure levels. The results show that socioeconomic and housing characteristics shape energy spending in diverse ways. In lower quantiles, household type, vehicle fuel choice, and heating systems are more influential, while in upper quantiles, income, residence type, and the number of automobiles dominate. Across all quantiles, appliance efficiency and fuel preferences remain important levers. These findings highlight that uniform policies are unlikely to succeed and that tailored, contextsensitive strategies are needed. By linking household behaviour with institutional realities, the study provides evidence to guide more inclusive and effective energy policy design.