Research On The Spatial Variation Of The Urban Housing Land Prices Based On Geographically Weighted Regression Model

The formation of urban housing land prices results from multiple factors taking effects within urban area in geographical space. Exploring the law of housing land price spatial variation and discovering the spatial relationship between housing land prices and influencing factors facilitate to gain deep insight into the distribution characteristics of housing land prices, as well as the internal law and the mechanism of formation, thereby providing scientific decision-making basis for land market surveillance. The research was implemented both theoretically and empirically to study the spatial variation of urban housing land prices.Theoretically, based on reviewing existing researches, this paper summarized the theoretical basis of the spatial variation of urban land prices and made a clear definition. It proved that the formation of urban land prices was related to the geographical locations and presented the spatial reliance and the spatial heterogeneity, thus resulting in spatial variation pattern of land prices in geographical space.Empirically, this paper made a case study for housing land prices in Hangzhou, and the methods from the Geographical Information System (GIS) and its spatial analysis function were adopted for a quantitative research. The Exploratory Spatial Data Analysis (ESDA) was applied to discover the structure of the housing land price data distribution and how it changed in geographical space. The results showed that the price data in Hangzhou did not accord with the Gaussian distribution; in geographical space, its changing trends in the horizontal and vertical direction were separately in the form of conic curve. Further, spatial autocorrelation analysis was employed, and it revealed the spatial clusters of housing land prices in Hangzhou:the main urban area was the hot-spot while the northern and northeastern city was on the opposite; meanwhile, in substantial areas, housing land prices were in random distribution. Furthermore, the Universal Kriging Interpolation was applied to produce a continuous surface of housing land prices in Hangzhou, so the spatial variation pattern was concluded. The result found its monocentric spatial structure:the prices peaked at the CBD and declined gradually from the city center to suburban areas, meanwhile with several exceptions in the urban area.Further, this paper made an intensive study of Geographically Weighted Regression (GWR) Model and Mixed Geographically Weighted Regression (MGWR) Model, including their foundations, weight acquisition and fitting method. Meanwhile, this paper analyzed the suitability of GWR Model in land price researches. Thus, GWR Model was applied into the study of spatial variation of urban housing land prices. In the case study of Hangzhou,9 locational factors and 2 parcel-specific factors were selected and quantified. Thus, the GWR Model was built to analyze the quantitative relations between housing land prices and the influencing factors, and their relations were presented by means of map visualization. The result revealed that, on one hand, housing land prices in Hangzhou were affected by the specific attributes of parcels; on the other hand, locational conditions played essential roles. The distance to the West Lake and to hospitals, and the plot ratio of land, respectively presented their negative correlations with housing land prices, while the distance to expressway showed its positive correlations with housing land prices; meanwhile, the influences from the CBD/main commercial districts, the metro stations, department stores and the Jinghang Canal respectively varied a lot among different geographical locations. By model comparison, it showed the GWR Model fitted housing land prices much better than the ordinary linear regression model did. The significance test of the GWR Model regression coefficients revealed that the degree of spatial nonstationarity made differences among those influencing factors. Accordingly, the MGWR Model was built; depending on whether the influencing factors with significant spatial nonstationarity or not, the MGWR Model applied the Weighted Least Squares (WLS) fitting estimation to the former factors (the local variables) while applied the Ordinary Least Squares (OLS) fitting estimation to the latter ones (the global variables). The fitting process of the MGWR Model was implemented by means of both the PLES method and the Back-Fitting iterative method in programe codes. Model comparison showed that the result of the MGWR Model had better goodness of fit than that of the GWR Model. Compared with the GWR Model, the MGWR Model made a more satisfactory interpretation for the spatial relationship between urban housing land prices and the influencing factors.