Authors: Mehak Sachdeva*, Arizona State University
Topics: Spatial Analysis & Modeling, Geographic Information Science and Systems, Quantitative Methods
Keywords: Local modeling, GWR, MGWR, spatial regression, multiscale house price modeling
Session Type: Paper
Start / End Time: 11:50 AM / 1:05 PM
Room: Virtual Track 2
Presentation File: No File Uploaded
Real estate market analysis has long been an active area of inquiry and one that reveals much about people’s preferences regarding housing attributes. Housing prices tend to exhibit spatial dependencies with high value houses tending to cluster with other high value properties and low value properties tending to be located near other low value properties. Additionally, house prices tend to vary across space due to differences in corresponding structural and neighborhood characteristics and these processes are likely to be unstable over varying scales of measurement. There is abundant empirical research to date using spatial extensions to traditional hedonic models to address spatial non-stationarity observed in house prices such as geographically weighted regression (GWR), spatial lag models etc. However, very few, if any, applications in real estate research have attempted to explore the unique scales at which different processes affect house prices. Using house price data in King County, WA, we apply a multiscale extension to GWR, multiscale geographically weighted regression (MGWR), to measure and investigate spatial variations in the processes affecting house prices at varying scales. In a novel attempt, we also quantify the intrinsic locational value housing units have beyond the determinants used to define traditional hedonic price models. In addition, we demonstrate that the MGWR modeling technique provides much better results than traditional approaches to house price modeling and reveals intricate housing submarkets often overlooked by other techniques.