Heterogeneities in the House Price Elasticity of Consumption

Abstract

I provide new evidence on the house price elasticity of consumption by exploiting micro-level consumption data from the Nielsen consumer panel for 2004 through 2016. I estimate elasticity as a non-parametric function of household characteristics, locations and time using Generalized Random Forest (GRF), a causal machine learning model. At the county-level, the average elasticity ranges from 0.04 to 0.16 with some neighboring counties being up to eight standard deviations apart, while household elasticities range from 0.01 to 0.2. Among all characteristics, having a child, household size, and the age of a household head create substantial disparities. I find that locations with volatile housing markets are less elastic. This means that failing to account for local heterogeneities overestimates the magnitude of total consumption responses in booms and busts. Moreover, local heterogeneities in elasticity camouflage the existing asymmetry in responses. Looking within a county reveals that households, especially more financially-constrained households, are more elastic in busts than in booms. Policymakers should account for this individual and geographic heterogeneity in consumption responses to house price changes when formulating policy.

Keywords: House Price Elasticity, Consumption, Generalized Random Forest, Machine Learning