Title: Spatial models in the space of covariates: Methodological and computational issues
Authors: Flavio Santi - University of Verona (Italy) [presenting]
Maria Michela Dickson - University of Trento (Italy)
Diego Giuliani - University of Trento (Italy)
Giuseppe Espa - University of Trento (Italy)
Abstract: Spatial dependence in the space of covariates of a regression model may arise for several reasons, including spatial trends of covariates, and model misspecifications. Just like spatial dependence over geographical or physical spaces, dependence of regression residuals in the space of covariates may lead to inconsistent estimates of regression parameters as well as of standard errors, thereby making inference unreliable. Although standard methodologies of spatial econometrics also hold when modelling dependence in the space of covariates, the very nature of that space poses new methodological and computational issues, mainly because of the dimensionality of the space of covariates, and because of the lack of isotropy in the spatial dependence. Both problems are analysed both from a theoretical and a computational point of view. In particular, it is studied how anisotropies should be modelled when the space of covariates is used as a proxy of the geographical space.