Predicting Property Rating Values Using Geographically Weighted Regression

Currently, the market value for rating valuation applied in Malaysia is the single property valuation technique. This technique is not efficient enough, involving high costs and large labor force because rating involves valuation of large number of properties. Multiple Regression Analysis (MRA) was applied due to these weaknesses. However, the MRA fails to account for the spatial effects (spatial heterogeneity and spatial dependence) inherent in property data. In this study, the Geographically Weighted Regression (GWR) model is introduced as a new method to value rating properties. The GWR model is able to capture spatial heterogeneity by allowing different relationships to occur between variables at different points in space. This study has two objectives. The first objective is to determine the attributes to be used for MRA and GWR model in this study. Data for this study were collected from two local authorities to represent rent and transaction data-based rating. Data for rent was obtained from Majlis Perbandaran Kajang (MPKj) and data for transaction was obtained from Majlis Perbandaran Kulai (MPKu). Final attributes for rent–based rating area are land area, main floor area, ancillary floor area, type of ceiling, property position, property type, age of building, distance to centre business of district and neighborhood quality and the attributes for transacted-based rating area are land area, main floor area, additional floor area and floor finishing. The second objective is to compare the performances of the GWR model with the MRA model in predicting rating values in the study areas. The result of R2, Adjusted R2, F-test and standard error of estimates proved that the GWR model provides better fitness compared to the MRA model. Residual analyses also reveal the same conclusion where residual for the GWR model is smaller in absolute values and probability distribution close to normal. The GWR model has also successfully captured spatial heterogeneity in almost all attributes. The prediction assessment of out-sample observations also revealed that the GWR model is able to produce better prediction. The ability of the GWR model to capture spatial effects is the main reason for this model to perform better; the GWR model is able to solve spatial heterogeneity problem explicitly and spatial dependence problem implicitly. Thus, the GWR which has been proven to be able to produce accurate prediction with small number of attributes should be used for rating valuation in Malaysia.

Preview PDF ITMA_2010_1A.pdf Download (542kB)

Actions (login required)

View Item