From Linear to Learning: Machine Learning for the Price Elasticity of Gasoline Demand

Abstract

This paper introduces new insights into estimating the price elasticity of gasoline demand in the U.S. by integrating machine learning (ML) techniques within the Double/Debiased ML framework while accounting for endogeneity issues. The ML-based approach yields a price elasticity estimate of -0.0145, approximately 80% smaller in absolute value than the conventional two-stage least squares estimate of -0.0715. This disparity challenges the current understanding of gasoline demand responsiveness to price fluctuations, implying that policies designed to curb gasoline consumption through price mechanisms may be less impactful than anticipated. By showcasing the potential of ML techniques in econometric analysis, particularly in demand estimation, this paper paves the way for future research in this domain.

Keywords: Gasoline Demand, Price Elasticity, Double/Debiased Machine Learning, Causal Inference, Market-Based Instruments

JEL Classification: C14, C45, Q41, R41