Incorporating spatial heterogeneity in determinants of secondhand housing prices in Zhengzhou City

This paper demonstrates how to construct an effective and accurate real estate appraisal method using a geographically weighted hedonic regression equation. It considers the spatial distribution of second-hand residential houses and reveals the mechanism of various determinants impacting on houses by explicitly incorporating spatial heterogeneity. The paper shows that space matters and implicit valuation of housing attributes differs over space. Also, the results suggest that incorporating space explicitly in real estate valuation models will improve the accuracy of property appraisal. With the rapid development of China's real estate industry, traditional single asset valuation methods cannot meet the demand appraisal of properties. Combined with the impending introduction of property taxes, mass appraisal of real estate is needed. However, current property valuation techniques mainly rely on the ‘Land Datum Value Method’ and the ‘Market Comparison Approach’, which are both prone to inaccuracy and subjective opinions of the appraisers. Actually, in active real estate markets, an estimation of the relationship between housing determinants and the housing price requires rich samples of housing price. Therefore, a first step would be to quantitatively describe this relationship by means of geostatistical models, and provide an objective appraisal method for housing price. Besides, another problem imbedded in the current valuation methods is the lack of consideration for spatial heterogeneity. Residential properties are highly heterogeneous good, by virtue of their differing location. Even at the same time, prices vary greatly depending on the different structural, neighborhood and locational characteristics over space. However, existing real estate appraisal methods do not take spatial heterogeneity into account, resulting inaccurate assessment. This paper firstly analyses the research background including the national real estate market and the Zhengzhou housing market. Subsequently, previous studies on the Hedonic Price Model and spatial heterogeneity of housing prices are discussed. Next, the core empirical part of this research is presented. In this paper, housing attributes and prices for 8864 residential houses in Zhengzhou City analysed. Firstly, some spatial features are derived based on the Exploratory Spatial Data Analysis (ESDA) method in ArcGIS. Secondly, an Ordinary Leastsquares (OLS) estimation based on the Hedonic Price Model is presented to explain the basic correlation between the housing determinants and housing prices. Thirdly, the Global Moran’s Index is evaluated to examine if there is spatial autocorrelation problem in the OLS estimation. After that, a Geographically Weighted Regression (GWR) estimation is used to test the spatial heterogeneity in the price composition of the second-hand housing market in Zhengzhou City. The final step is to compare these two estimation methods and the conclusion can be drawn as follows: Compared with the OLS estimation, the GWR estimation has a better performance in the adjusted R2, AICc value, the Moran’s Index and sum squares of the residuals, indicating that the GWR estimation is more suitable for the evaluation of housing price. In other words, space matters in determining house prices and attributes of houses are valued differently over space. Also, the coefficients in the GWR estimation of these housing determinants vary in space with their different spatial locations and reflects the spatial heterogeneity of the influence of these indicators. This would indicate that cities follow particular patterns with respect to their morphology, which drives the difference in prices of housing attributes. These results suggest that, taking the spatial heterogeneity of housing price into account, will provide more effective and accurate models for the housing price evaluation. Particularly with respect to mass appraisal methods, incorporating spatial heterogeneity will result in more accurate estimates. Taxation departments can use the results of this paper to improve their computer assisted valuation models. Nevertheless, additional research may be needed, for instance to look into the questions of poly-centricity, as well as addressing current data (and computation) limitations.