Spatiotemporal Model Analysis Of Urban Housing Price

Housing is one of the hotest social issues in China.The housing price is affected by the market supply and demand,macro socio-economy and national policies.It is also closely related to micro factors such as the geographic location and the housing characterstics.The analysis and forecasting of urban housing price is of great significance for housing security,reale-state development and urban planning.This paper aims to test and improve the geographical weighted regression(GWR)models for analysis of urban housing price in a class of resource-dependent cities.Huangshi City,a resource-dependent,land-scarce and multi-center city in Hubei Province,China,was selected as the research area.It is one of the typical resource-exhausted cities at the period of industrial transformation.The commercial housing data,such as flat location,unit price and various housing characteristics in the study area between year 2007 and 2012 were collected.In addition,the urban geographic data were prepared in GIS geodatabase.Based on statistical,spatial and temporal exploratory analysis,the spatiotemporal characteristics of urban housing price were summarized and several housing indicators were selected for modeling the housing price.The sample data were tested by regression models such as OLS,GWR and geographically and temporally weighted regression(GTWR).GWR or GTWR is a useful technique for exploring spatial nonstationarity by calibrating a regression model which allows different relationships to exist at different points in space.However,spatial autocorrelation could invalidate the model assumption and sometimes may result in residual dependency.This paper improved GWR and GTWR by introduced additional variables based on spatiotemporal windows.The size parameters for defining spatiotemporal windows were estimated by spatial and temporal statistics of all the sample data.The new window variables were calculated by averaging the explained variables which were located in its spatiotemporal window.The new variables were added in GWR and GTWR as improved regression models(IGWR and IGTWR).The main research findings are as follows:(1)Housing unit price in the study area shows significant spatial and temporal patterns,and is correlated with residential community level,plot ratio,green rate,the total number of building layers,the year of sale and the distance from regional center significantly on local scale rather than global scale.In addition,the influence is not obvious for land price on housing price.Due to scarcity of land,the plot ratio and housing prices have a significant positive correlation.(2)The case study indicates that GWR model could more accurately model the spatial relationships between housing price and housing indicators.Compared with the conventional linear regression model,the measure of goodness of fit(R2)of GWR is increased from 0.48 to 0.82,and the average prediction error is reduced from 18.3% to 8.4%.The model also provides a series of spatially distributed regression coefficients,which could be used to explain the locational variations of housing price.In addition,GTWR improve the model goodness of fit to 0.87.(3)By introducing the window based indicators in GTWR as new explanatory variables,the IGTWR model might estimate the impact of spatial and temporal autocorrelation between geographic data.For the case study,the modeling quality of IGTWR is improved in terms of the measure of goodness of fit(R2),the Akaike information criterion(AICc)and the residual sum of squares(RSS).The conclusions and major contributions of this research are summarized as follows:(1)The statistical,spatial and temporal exploratory data analysis is fundamental for modeling urban housing price.For deferent geographic and socioeconomic background of the urban regions,exploratory data analysis could explore the complex relationships between the housing price and potencial factors.It is nessasary to select pricing indicators by various exploratory methods for further modeling analysis.(2)For modeling analysis of housing price in Huangshi City and other silimar cities,GWR,GTWR and IGTWR models are much better than general linear regression.The spatiotemporal variation of housing price could be more practically explained by locational regression coefficients.In terms of modeling measures,GTWR model is better than GWR model,and IGTWR is better than GTWR.(3)Considering the spatiotemporal autocorrelation between data samples and the possible residual dependency in GWR and GTWR modeling,the additional explainary varioubles based on the spatiotemporal windows could improve the modeling quality.