Spatial Dependence, Housing Submarkets, and House Prices

This paper compares the impacts of alternative models of spatial dependence on the accuracy of house price predictions in a mass appraisal context. Explicit modeling of spatial dependence is characterized as a more fluid approach to defining housing submarkets. This approach allows the relevant “submarket” to vary from house to house and for transactions involving other dwellings in each submarket to have varying impacts depending on distance. We compare the predictive ability of different specifications of both geostatistical and lattice models as well as a simpler model based on submarkets with fixed boundaries. We conclude that – for our data – no spatial statistics method does as well in terms of predictive ability as a simple OLS model that includes a series of dummy variables defining submarkets. However, of the spatial statistics methods, geostatistical models provide more accurate predictions than lattice models. We argue that this is due to the fact that the kriging procedure used to make predictions in a geostatistical framework directly incorporates spatial information about nearby properties. That is not possible in a lattice framework due to the reliance on a matrix of weights that incorporates relationships only for the sample of properties that transact.