•  
  •  
 

Journal of Property Tax Assessment & Administration

Abstract

Location is a key factor in the valuation of residential properties, but it can be a challenge to capture its impact well. This paper considers a Gaussian Markov Random Field (GMRF) approach to modelling property values that enables the location element of the model to be considered as a smooth process and allows a more flexible correlation structure than alternatives. The model was built to value approximately 1.5 million properties in Wales, UK, for a revaluation planned by the Welsh Government. Due to the large scale of the model a Bayesian technique, known as the Integrated Nested Laplace Approximation (INLA), was employed to make the model computationally feasible. This method allowed spatial dependence to be accounted for and improved model performance over a mixed model with nested geographical variables.

First Page

37

Last Page

53

Keywords

Statistics, Real property - Valuation

Download

Share

COinS