Spatial econometrics and the hedonic pricing model: what about the temporal dimension

Recent ready access to free software and toolbox applications is directly impacting spatial econometric modelling when working with geolocated data. Spatial econometric models are valuable tools for taking into account the possible latent structure of the price determination process and ensuring that the coefficients estimated are unbiased and efficient. However, mechanical applications can potentially bias estimated coefficients if spatial data is pooled over time because the applications consider the spatial dimension alone. Spatial models neglect the fact that data (e.g. real estate) may consist of a collection of spatial data pooled over time, and that time relations generate a unidirectional effect as opposed to the multidirectional effect associated with spatial relations. Through an empirical case study, this paper addresses the possible bias in spatial autoregressive estimated parameters when data consist of spatial layers pooled over time. An empirical study is made using apartment sales in Paris between 1990 and 2001. Estimation results and out-of-sample predictions confirm, at least for this case, the hypothesis that ignoring the time dimension and applying spatial econometric tools generate divergence among the estimated autoregressive coefficients, which can potentially engender other serious problems.(This abstract was borrowed from another version of this item.)