A spatial autoregressive geographically weighted quantile regression to explore housing rent determinants in Amsterdam and Warsaw

Abstract

A hedonic approach is typically performed to identify housing rental or sales price determinants. However, standard hedonic regression models disregard spatial autocorrelation of prices and heterogeneity of housing preferences across space and over price segments. We developed a spatial autoregressive geographically weighted quantile regression (GWQR-SAR) to address these shortcomings. Using data on the determinants of residential rental prices in Warsaw (Poland) and Amsterdam (The Netherlands) as case studies, we applied GWQR-SAR and rigorously compared its performance with alternative mean and quantile hedonic regressions. The results revealed that GWQR-SAR outperforms other models in terms of fitting accuracy. Compared with mean regressions, GWQR-SAR performs better, especially at the tails of the dependent variable distribution, where non-quantile models overestimate low rent values and underestimate high ones. Policy recommendations for the development of private residential rental markets are provided based on our results, which incorporate spatial effects and price segment requirements.