Spatial Clustering of Housing Construction in the Tokyo Metropolitan Area: An Application of Spatially Clustered Fixed-Effects and Spatially Correlated Random-Effects Models

We proposed two types of econometric models, a spatially clustered fixed-effects model (SCFEM) and a spatially correlated random-effects model (SCREM), to examine area-based panel data. We investigate what factors influence housing construction in the Tokyo Metropolitan Area, incorporating unobservable factors, local regulatory differences in housing development and spillovers of local public or private goods, which may cause spatial clustering or correlation of housing construction. The SCFEM is a type of fixed-effects model where a cluster has the same effects, so that we have to find which areas constitute a cluster. The issue of finding clusters can be regarded as a problem of model selection from too many possible models. We adopt an aggregate prediction error as a model selection criterion, which is estimated by a resampling method, namely leave-one-out cross-validation. We showed by simulations that the estimated parameters of concern are more efficient than the within estimates. The SCREM is a model where the random-effects are spatially correlated. We use the concentrated maximum likelihood method for the estimation. The unobservable area-effects are large in the east, west and north areas of the TMA but small in the south, where regulations against development are more severe than in the other areas. Clusters are found in huge cities