Hierarchy and spatial autocorrelation effects in hedonic models.

Hedonic housing price models should deal with spatial autocorrelation in order to avoid bias and inconsistency in the coefficient estimates. Multilevel models have been presented as a way of properly considering the effects of neighbourhood amenities operating at different spatial hierarchies. In this paper, we test this feature by specifying a three-level model for a dataset on downtown Madrid, which includes spatially lagged explanatory variables. The results show that hierarchical models are not capable of fully capturing spatial autocorrelation so that more effort should be done in developing the appropriate spatial multilevel models.