Mass appraisal of rural properties using geographically weighted regression

Traditionally the classical linear regression models (CLRM) are used on mass appraisal of real estate, however, there has been the need to model the data spatially. The real estate values in rural areas are also affected by geographical location, nevertheless the modeling of the geographical effects has been used mainly in evaluations of urban areas. The purpose of this paper to use the geographically weighted regression models (GWR) in a sample of rural properties for the preparation of a map of standard ground value (MSGV) of an area of the North region of Rio de janeiro-Rj, Brazil. The proposed methodology was based on the application GWR, evaluate its fitness and performance with respect to the CLRM and produce the MSGVthrough the Kernel interpolator. The sample used was composed by 113 observations and 25 validation samples. The performance of the surface values obtained were analyzed through the median of ratios, Coefficient Of Dispersion (COD) and by the Price Relative Differential (PRD) and compared with the reference values recommended by International Association of Assessing Officers (IAAO). The GWR model was superior to the CLRM in all analyzed criteria. This led to the conclusion that the GWR yielded a better model adjustment and, therefore, was superior to the classical regression model. Regarding the use of Geostatistics to interpolate the values and generate the MSGV, the Kernel proved to be appropriate, as it generated a detailed surface, enabling the generation of values between the neighbors, in regards that it is difficult to obtain data for the entire studied area during field collection, especially collection, especially in rural areas.