The recent threat of a 'housing market bubble', and the ramifications for the economy of it bursting, has made the construction of an appropriate house price index a topical issue. In particular, the popular measures of changes in 'median' or 'mean' house prices are widely known to be unreliable, being highly sensitive to changes in the mix of properties sold (some of which is seasonal in nature) and the average quality of housing over time.
There are a number of methods proposed in the literature to improve the accuracy of house price index numbers. Typically, more simplistic methods have proved popular, particularly in Australia, as they are less data intensive. However, this thesis shows that housing data are accessible in Australia by compiling a dataset for the Brisbane metropolitan area, which makes it possible to employ more elaborate and refined methods of house price index construction.
The main objective of the thesis is to propose an improvement to existing methods of house price index construction by addressing three important oversights in the literature. Firstly, it is plausible that the shadow prices of property attributes evolve in a smooth manner, in line with the nature of consumer preferences over time. Existing methods either assume (explicitly or implicitly) that shadow prices are constant or change in a haphazard fashion from one period to the next. Secondly, the price of a house is spatially correlated with the price of houses that are in close proximity. The notion that 'location, location, location' is of particular importance is widely known in both public and academic forums. However, unlike the literature on house price prediction, existing models of house price indexes have failed to incorporate spatial autocorrelation. Thirdly, there has been little research into the accurate specification of index number formulae particular to the housing case, which is partly due to existing methods implicitly defining an index within their formulation. However, improvements in the comprehensiveness of housing data allows for methodologies that consider a wider spectrum of index formulae.
This thesis outlines a new methodology, which addresses these issues. Namely, a hedonic equation with smoothly time-varying parameters is estimated using a state-space formulation with spatially autocorrelated errors. Subsequently, the estimated model is used in constructing a housing price index using hedonic imputation (HI) methods. The formulated HI indexes are also applied to the traditional spatial errors model, which is new to the literature. The methods proposed in this thesis are illustrated using housing price data for the Brisbane metropolitan area, compiled from the property information service, RP Data.