Factors affecting spatial autocorrelation in housing prices : an empirical study of Hong Kong

Abstract

Real estate economists and practitioners have been cognizant of spatial autocorrelation in housing prices for more than three decades. In the early days, they relied to a huge extent on techniques developed in statistical science and focused exclusively on its identification and quantitative assessment in housing studies. It has been well-acknowledged that the presence of spatial autocorrelation in housing prices will compromise the applicability of conventional hedonic statistics, which may lead to biased and inconsistent estimates. In light of this, recent studies in the area have spawned an immense literature aimed at devising sophisticated econometric models such as hedonic spatial lag model and hedonic error model as correction methods. Interestingly, the underpinning factors attributing to its existence remain relatively theoretically unexplored due perhaps to a paucity of quality goereferenced housing data as well as the indifferent attitude espoused by the researchers. Although some general propositions regarding its causes have been proposed, they are accused of lacking inferential basis.

Acknowledging the above research gap, this thesis attempts to investigate the underlying factors affecting the formation of, and change in, spatial autocorrelation in housing prices. Specifically, we conjecture that spatial autocorrelation is crucially determined by one of the economic workings of the housing market—price determination process. It is posited that the occurrence of the spatial phenomenon is a direct consequence of how market participants in search of past information in the market ascertain current housing prices. Specifically, spatial autocorrelation is deemed to be established, or increase, when property traders infer current housing prices from past sales of properties (i.e. comparables) located in the same neighborhood as the subject houses.

Based on the above information search framework, we put forward three hypotheses to facilitate our examination. First, it is hypothesized that market volatility depresses spatial autocorrelation. As in any other commodity market, past sales transactions in real estate are important sources of price information, which become more obsolescent, and hence, fail to be a good price signal when the market is more volatile. Traders are compelled to rely less on past sales in establishing the current prices of the housing units. Accordingly, spatial autocorrelation will be diminished; following broadly the same line of logic, the second hypothesis is constructed, which states that market liquidity (defined as total market transaction volume) dampens spatial autocorrelation. Given that market liquidity reflects the amount of price information being circulated in the market, traders in accessing property values in a “thick” market do not have to necessarily infer from comparables that are located further away from the subject properties. Hence, a weaker spatial autocorrelation relationship between prices is resulted; third, building age is a critical factor in assessing a house’s redevelopment potential, whose value is largely independent of the transaction prices of the surrounding housing units. Given that a house’s total value can be perceived as an addition of its use value and redevelopment value (i.e. real option value of redevelopment), and that the former’s role decreases whereas the latter’s increases with building age in appraising its total value, it is therefore hypothesized that spatial autocorrelation decreases as building age increases.

Several spatial autoregressive hedonic models are developed, with which the three hypotheses are tested using geo-coded open market transaction data in Hong Kong for the period of 1997 to 2008. The results indicate no contrary evidence rejecting any of the hypotheses at the 1% significance level. They soundly confirm the roles market volatility, market liquidity and building age played in the price determination process in real estate, as well as in the formation of spatial autocorrelation.

The research findings of this thesis carry several straightforward but far-reaching implications beyond the theoretical literature, among which is the significance to, and impact on, property valuation and economic analysis of local housing systems. A better conceptual understanding on the causes of spatial autocorrelation in housing prices can greatly prompt the development of more parsimonious hedonic models, which are econometrically appealing. In this sense, statistical problems and complications (e.g. loss of degrees of freedom) associated with traditionally-used models, which generally routinely incorporate a long list of locational variables, can be circumvented. In addition, hedonic models derived based on our research findings, which give a simpler yet more realistic representation of the housing market, make real estate mass appraisals much less computationally-intensive. Real estate mass appraisals are used for a variety of purposes, including but not limited to taxation, mortgage assessment, government policy-making, and investment. They are of specific interest to accountants, bankers, real estate developers and investors, government authorities and economists.