Avaliação em massa de imóveis urbanos através de modelagem de regressão espacial

Abstract

With the increasing demand for efficiency in municipal revenue collection, real estate appraisal has gained even more relevance due to the fundamental role of the Property and Territorial Urban Tax (IPTU). Since the implementation of Complementary Law No. 101 of May 4, 2000, known as the Fiscal Responsibility Law (LRF), municipalities have faced an urgent need for more robust methods to calculate the market value of properties, driven by the pursuit of tax equity. Detailed and precise planning is essential to ensure that the values assigned to properties fairly reflect their characteristics and location. This study modeled parameters for the appraisal of urban properties, promoting fair proportionality in the generation of Generic Value Maps (PVG). The research was conducted in Camaçari, a municipality in the Metropolitan Region of Salvador (BA), which has over 300,000 inhabitants and stands out as one of Brazil's main industrial hubs, hosting the Camaçari Industrial Pole. The study covered areas in the neighborhoods of Arembepe, Barra de Jacuípe, Boa Esperança, Genipabu, and Vale do Landirana, using a sample of 124 land plots for modeling and 12 for validation. The methodology compared the Ordinary Least Squares (OLS) regression with the global spatial model Conditional Auto Regressive (CAR) and the local Geographically Weighted Regression (GWR) model. The models were evaluated based on quality indicators such as the Akaike Information Criterion (AIC), Log Likelihood (LIK), coefficient of determination ($R^2$), and Root Mean Square Error (RMSE), in addition to specific metrics for mass appraisal, such as Median Ratio (Med R), Coefficient of Dispersion (COD), and Price Related Differential (PRD), according to the International Association of Assessing Officers (IAAO) criteria. The results indicated that GWR was able to significantly reduce spatial effects, outperforming the other models in most quality and performance criteria. The methodology was combined with Ordinary Kriging to generate unit value surfaces. The semivariogram adjusted to the Gaussian model allowed the kriging of predicted values, and the surface generated by the GWR predicted values achieved the lowest RMSE and better performance indicators, close to the limits recommended by the IAAO. It is concluded that GWR modeling was the most effective in representing spatial variations in real estate values in the studied area, and the proposed methodology for PVG generation demonstrated great potential for application by municipalities in the mass appraisal of urban properties, contributing to fairer and more precise tax collection, especially for taxes such as IPTU and the Real Estate Transfer Tax (ITBI).