An analysis of land prices: A structural time-series approach

Abstract

This paper analyses spatio-temporal variation of land prices in two single localities by means of structural time series modelling formalism that combines the flexibility of a time series model with that of the interpretation of a regression analysis. The extension of conventional hedonic models by introducing unobserved components for trend and cycle resulted to significant improvements in their post-sample predictive accuracy. In predictive testing, for most model formulations the unobserved component approach generated only a marginal average prediction error when compared to the orthodox hedonic models, which, in contrast, yielded to a considerable amount of systematic prediction error. It therefore seems that the structural time-series modelling paradigm offers a more viable alternative to the hedonic analysis of land prices than the conventional approach based on least squares estimates. The effect of slope component in the trend specification was found to be statistically insignificant, which implies that the elementary local level model would be the most adequate description of the long-term land price movements.