Examining the nonlinear and spatial heterogeneity of housing prices in urban Beijing: an application of GeoShapley

Housing is essential for human well-being and economic stability. Major metropolitan areas, particularly in developing countries, face severe housing price challenges. Traditional Hedonic Pricing Models (HPM) have extensively examined the determinants of housing prices, often assuming linear relationships and overlooking submarket segmentation. While approaches such as Geographically Weighted Regression (GWR) address spatial heterogeneity, they may still struggle with capturing complex nonlinear interactions between housing attributes, neighborhood factors, and spatial dependencies. To overcome these limitations, this study combines Extreme Gradient Boosting (XGBoost) with the GeoShapley to better model nonlinear and spatially varying effects on housing prices. The GeoShapley summary plot reveals that spatial location (GEO) is the most influential feature, followed by distance to the CBD, housing age, and housing size, along with their interactions with GEO. Further analysis uncovers that larger suburban homes show weaker market performance compared to smaller units in central districts, revealing distinct submarket dynamics. Properties near the CBD, particularly in school districts and green landscapes, maintain higher value due to the spillover effects of educational and environmental amenities. Conversely, the negative correlation between proximity to Xizhimen Metro Station and housing prices highlights the complexity of metro accessibility, where factors such as station design might diminish the expected premium. These insights inform real estate policy and sustainable urban planning by spotlighting the importance of spatial heterogeneity and threshold effects, thus extending classical theories of urban housing markets to account for submarket-specific price formation processes.