A Review of Empirical Studies on Property Price Gradients
地产价格梯度研究进展

The paper reviews the methodologies, assumptions and improvements in empirical studies of property price gradients. The hedonic pricing model and the repeat-sales model are two basic methodologies of estimation. The hedonic pricing model is widely used as it requires fewer price data and the functional form is flexible to include different explanatory variables. The repeat-sales model uses a pair of observations for a single property, helpful for researchers when choosing variables, but subject to the availability of pair-wise transaction records. New methods which are able to reveal the complexity of price gradient curves are also employed, such as the spline function and the nonparametric approach. There are three different assumptions on the spatial structure of an urban area: the monocentric assumption, the non-monocentric/polycentric assumption and no priori assumption. When the monocentric assumption is used, the distance-to-CBD variable is included. Euclidean distance is used most frequently, but it is as unable to represent the accessibility to CBD. With the distance and location dummy variables, the fitting accuracy of the model is also improved. As American modern urban areas have become increasingly decentralized and polycentric, a non-monocentric or polycentric assumption was suggested, but it is not easy to identify each sub-center accurately. Recently, some scholars suggest a location dummy variable without any assumptions about the location of the center to cope with the problem. Empirical studies of Chinese domestic situations mainly focus on Beijing. It is believed that the monocentric assumption is suitable for Beijing and most of other cities, which suggests that influential sub-centers have not yet formed. Attention has been given to the suburbanization of mega-cities since 1982, and the effect of sub-centers on price gradients in some cities has been of concern. The hedonic pricing model is used when estimating price gradients, but the repeat-sales model remains untested. Ignoring the heteroscedasticity of estimations due to spatial autocorrelations may result in failure of the negative gradient in empirical studies. Numerous methods have been proposed to test and rectify spatial autocorrelation. To conclude this review we propose that further research focus on sample data selection, setting variables, assumptions on the urban spatial structure, and improvements of the estimation models.