Abstract
House prices tend to be spatially correlated due to similar physical features shared by neighboring houses and commonalities attributable to their neighborhood environment. A multilevel model is one of the methodologies that has been frequently adopted to address spatial effects in modeling house prices. Empirical studies show its capability in accounting for neighborhood specific spatial autocorrelation (SA) and analyzing potential factors related to house prices at both individual and neighborhood levels. However, a standard multilevel model specification only considers within-neighborhood SA, which refers to similar house prices within a given neighborhood, but neglects between-neighborhood SA, which refers to similar house prices for adjacent neighborhoods that can commonly exist in residential areas. This oversight may lead to unreliable inference results for covariates, and subsequently less accurate house price predictions. This study proposes to extend a multilevel model using Moran eigenvector spatial filtering (MESF) methodology. This proposed model can take into account simultaneously between-neighborhood SA with a set of Moran eigenvectors as well as potential within-neighborhood SA with a random effects term. An empirical analysis of 2016 and 2017 house prices in Fairfax County, Virginia, illustrates the capability of a multilevel MESF model specification in accounting for between-neighborhood SA present in data. A comparison of its model performance and house price prediction outcomes with conventional methodologies also indicates that the multilevel MESF model outperforms standard multilevel and hedonic models. With its simple and flexible feature, a multilevel MESF model can furnish an appealing and useful approach for understanding the underlying spatial distribution of house prices.
Highlights
The hedonic pricing model is one of the most widely used specifications for house price prediction
The introduction of a random effects (RE) term in the multilevel model increased R2 to 0.752 by accounting for spatial autocorrelation (SA) that was present within block groups
The Anderson-Darling test p-values indicate that residuals of the three models deviated from bell-shape curves, but the multilevel Moran eigenvector spatial filtering (MESF) model residuals were relatively closer to a normal distribution than were those for the multilevel model
Summary
The hedonic pricing model is one of the most widely used specifications for house price prediction. The prices of nearby houses tend to be comparable because of similar neighborhood amenities (e.g., access to public facilities, socioeconomic status) and their similar physical characteristics (e.g., lot size, house age) This potential correlation in space can violate the independent observations assumption in a hedonic model specification, and may lead to inefficient and biased parameter estimates [1]. House property values vary across geographic space, highly depending upon geographic location, house characteristics, and neighborhood environment Due to their heterogeneous nature, house values have traditionally been described with a hedonic model, which generally utilizes a linear regression specification that includes the attributes of houses and geographic locations as covariates to describe house transaction values [13]. This extended model achieves better model performance and an increased prediction accuracy by accounting for SA [1]
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
