Novel methods to demarcate urban house submarket-cluster analysis with spatially varying relationships between house value and attributes

In urban house market studies, urban housing market can be divided into a series of submarkets. Usually, submarkets are identified with either geographic locations or housing structural characteristics, or some combination of both. In this study, we propose an alternative to identify urban housing submarkets. Instead of using house characteristics or locations, we use the relationships obtained through a geographically weighted hedonic regression (GWHR) model. In particular, we apply a K-means classification on the coefficients obtained via GWHR to identify different submarkets. Data from the City of Milwaukee are used to test the model and procedure. Comparison of a regular cluster analysis using housing structural and neighborhood socioeconomic information and the proposed procedure is conducted in terms of prediction accuracy. The analytical results suggest that hedonic regression on demarcated submarkets is better than a uniform market, and our proposed method yields more reasonable result than the ones using raw data.