Bayesian analysis of temporal and spatial trends of house prices in Norway

Abstract

The goal of this thesis is to analyse the temporal and spatial trends of house prices in Norway in a Bayesian setting. We will perform regression analysis of the data which will be modelled using structured additive regression models. This choice was made because structured additive regression models can be put into a computational framework of latent Gaussian models that can be analysed using integrated nested Laplace approximation (INLA). In addition, in a Bayesian setting each of the model parameters have their own posterior distributions from which we can get posterior means and credible intervals. The main findings were that after applying simple linear regression, new houses have both higher prices and higher price growths than used houses for all counties. Prices in Oslo grow much faster than in any other county. Including a spatially structured effect in the model, large geographical differences between counties were revealed. We conclude that the price differences between counties are reduced, taking the different population sizes into account.