Modelo hedônico espacial para avaliação em massa de imóveis de Fortaleza

• Autor(a) principal: Moreira, Felipe Fernandes
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Ceará (UFC)
• Texto Completo: http://www.repositorio.ufc.br/handle/riufc/56316
• topic: Avaliação de imóveis; Mercado imobiliário; Modelos hedônicos; Big data; Aprendizado de máquina

A more precise real estate appraisal generates benefits for all stakeholders involved in real estate value chain: customers, government and companies. Therefore, the capacity to construct comparative models capable of, with real market observations, predict more precisely the value of real estate assets influences positively a economic sector with large presence on national economy. Literature reveals that the use of methods capable of modelling non-linear relationships between variables, methods that incorporate the effects of special autocorrelation and emsemble of different models results with the purpose of reducing prediction errors. The main objective of the study is to build a mass real estate appraisal spatial hedonic model for the city of Fortaleza, Ceará, derived from techniques that allows exploring potential non-linear spatial autocorrelation effects. However, it is known that the modelling process is swayed by randomness, demanding that the betterment observed in error metrics results are produced by the changes applied by researchers. Thus, the performance of the proposed model will be compared with referential models built with well-established methods through statistical tests that can certify the significance of the difference between error metrics obtained. Statistical testing reveal that single error metrics comparison are not suited for evaluate model performance in relative and absolute terms. The study provide evidence that models constructed with ensembles (namely Random Forest) and models exploring spatial autocorrelation phenomenon (Spatial Autoregressive Models) are relativily more precise for real estate mass appraisal problems, although no found no evidence to state that there are significant difference between the performance of Random Forest and Autoregressive Spatial Models. Results point to advantages in the use of spatial autoregressive model, since in addition to its overall performance, such models are easily interpreted.