Estimating Housing Prices through a Spatial GAMLSS Modeling Approach

초록

Estimation of housing prices is of great importance in various sectors, including property tax assessment and collateral valuation. The multi-household house (MHH) is an emerging property type for the application of the automated valuation method in South Korea. Thus we choose Seoul, the capital of South Korea, as our study area, and apply the hedonic price model to MHH, a common tool in property valuation. We suggest an alternative approach to estimate housing prices – generalized additive models for location, scale, and shape (GAMLSS) – in order to overcome the limitations inherent in the traditional hedonic price model, those limitations being: lack of theoretical background for a functional form, assumption of a linear relationship between the explanatory and response variables, and little account of spatial autocorrelation in model building. We employ the GAMLSS with the spatial autocorrelation of the data being taken into account. We show that a non-normal distribution could give a better fit for house prices and illustrate nonlinear effects of the explanatory variables, as well as the spatial effect of house locations on the price of MHH. We hope that the spatial GAMLSS modeling approach in this study will lead to better property tax assessment and collateral valuation.

키워드