Ommitted locational variates in hedonic analysis: A semiparametric approach using spatial statistics

A frequent assumption of hedonic price estimation using property market data is that spatial autocorrelation of regression residuals is a feature of the error generating process. Under this assumption, spatial error dependence models that impose a specific spatial structure on the error generating process provide efficient parameter estimates. In this paper we argue that spatial autocorrelation is induced by spatial features influencing property prices that are not observed by the researcher. Whilst many of these features comprise the subtle nuances of location that might adequately be handled by modelling the error process, others may be substantive spatial features whose absence from the model is likely to induce omitted variable bias in the parameter estimates. Accordingly we propose an alternative estimation strategy. We use spatial statistics to determine the nature of spatial dependence in regression residuals. Subsequently we adopt a semiparametric smooth spatial effects estimator to account for omitted locational covariates over the spatial scale indicated by the spatial statistics. The parameter estimates from this model are found to differ significantly from those of a spatial error dependence model.