Modelos estatísticos para geração de plantas de valores genéricos em áreas urbanas

Abstract

The calculation of the Tax on Land and Urban property (IPTU) is based on property market value, which must match the value of cash sale, usually established in the city table of general values (PVG). These taxes provide municipalities with resources that are used in social services. However, there are municipalities, especially small ones, that do not collect IPTU taxes. This is due to outdated real estate cadastre, besides the lack of qualified personnel, financial resources and robust and easy methodology to determine real estate market value. Therefore, the present work aims to apply the combination of the spatial regression model and location factor modeling to determine the market value of each property in a small city for the generation of the table of general values (PVG). The study was conducted in the city of São Gotardo/MG, in the middle region of the Triângulo Mineiro/Alto Paranaíba, which collected R$ 17.68 per person with property tax (IPTU), in 2012. One hundred and eighty-four samples of residential real estate evaluations made by the Caixa Econômica Federal in 2012 and 2013 were used. All models were generated using 166 samples, since 10%, ie, 18 properties, were excluded to evaluate the quality of the prediction of the final regression model. Aiming to analyze the application of spatial models, four multiple regression models were generated based on the logarithm dependent variables of the total and unit value and the independent variables related to the construction characteristics of the buildings, according to previous studies. Additional variables related to the characteristics of the land were also tested. For the models with spatial error dependence, a spatial error model was generated to determine a new homogenized variable encompassing the location factor (VH), which was used as independent variable of a new linear regression model. The best regression model was selected based on the observance of the assumptions of the linear regression model and the analysis of the lowest Dispersion Coefficient resulting from the 18 samples that were not used in the generation of regression models. The model with the logarithm dependent variable of the unit value and the homogenized variable as independent showed the best result and observed all the assumptions. Thus, it was demonstrated that the homogenized variable generated from the spatial error model improves the performance of the linear regression model, since it includes the location factor of the property in the independent variables. Therefore, in order to determine property value with the highest possible accuracy, it is necessary to consider, besides real estates, other aspects often neglected by the government in the valuation of the basic value for tax (IPTU) collection.