Quantifying the drivers of residential housing demand – an interpretable machine learning approach

Abstract

Purpose

The paper employs a unique online user-generated housing search dataset and introduces a novel measure for housing demand, namely “contacts per listing” as explained by hedonic, geographic and socioeconomic variables.

Design/methodology/approach

The authors explore housing demand by employing an extensive Internet search dataset from a German housing market platform. The authors apply state-of-the-art artificial intelligence, the eXtreme Gradient Boosting, to quantify factors that lead an apartment to be in demand.

Findings

The authors compare the results to alternative parametric models and find evidence of the superiority of the nonparametric model. The authors use eXplainable artificial intelligence (XAI) techniques to show economic meanings and inferences of the results. The results suggest that hedonic, socioeconomic and spatial aspects influence search intensity. The authors further find differences in temporal dynamics and geographical variations.

Originality/value

To the best of the authors’ knowledge, it is the first study of its kind. The statistical model of housing search draws on insights from decision theory, AI and qualitative studies on housing search. The econometric approach employed is new as it considers standard regression models and an eXtreme Gradient Boosting (XGB or XGBoost) approach followed by a model-agnostic interpretation of the underlying effects.

Citation

Cajias, M. and Zeitler, J.-A. (2023), "Quantifying the drivers of residential housing demand – an interpretable machine learning approach", Journal of European Real Estate Research, Vol. 16 No. 2, pp. 172-199. https://doi.org/10.1108/JERER-02-2023-0008