Modelo híbrido de séries temporais para previsão de demanda do mercado imobiliário de São Paulo
Ano de defesa: | 2017 |
---|---|
Autor(a) principal: | |
Orientador(a): | |
Banca de defesa: | |
Tipo de documento: | Dissertação |
Tipo de acesso: | Acesso aberto |
Idioma: | por |
Instituição de defesa: |
Universidade Federal de Santa Maria
Brasil Engenharia de Produção UFSM Programa de Pós-Graduação em Engenharia de Produção Centro de Tecnologia |
Programa de Pós-Graduação: |
Não Informado pela instituição
|
Departamento: |
Não Informado pela instituição
|
País: |
Não Informado pela instituição
|
Palavras-chave em Português: | |
Link de acesso: | http://repositorio.ufsm.br/handle/1/18271 |
Resumo: | One of the biggest problems associated with the use of demand forecasts at the support the decision-making is the choice of forecasting method to be implemented. In the context, present a behavior different from other sectors, the real estate market has difficulty in finding correct methods to predict its demand, indeed, due to the significant time interval between the project decision making, investment, and the actual entry of the enterprise in the market dispute. This complexity leads to the choice of wrong methods, resulting in large inventories of residential units, generating high costs for builders and incorporators, as it has since 2014 in São Paulo, the most representative real estate market of Brazil. Therefore, this research aims to propose a hybrid model of time series for forecasting demand of real estate in the city of São Paulo. For this, will be used data referring to the time series of residential units sales, provided by SECOVI-SP. At first, the Exponential Smoothing, Box-Jenkins, Conditional Heteroskedasticity and Artificial Neural Networks models are modeled individually, posteriorly these are combined by means of six forecast combining techniques. The techniques used are Arithmetic Mean, Geometric Mean, Harmonic Mean, Minimum Variance, Linear Regression and Principal Component Analysis. The measures of accuracy to measure the obtained results and to select the best model are the RMSE, MAPE, MAE and UTheil of forecast. The results showed that the Linear Regression with independent variable being the combination of the SARIMA(2,0,0) (2,0,0)12 and MLP/RNA(18,25,1) models through Principal Component Analysis provided a performance satisfactory prediction, with RMSE of 349.21, MAPE of 17.1%, MAE of 287.62 and UTheil of 0.298. Thus, demonstrating that the combination and hybridization of time series models allowed a significant increase in prediction performance. Finally, we used the proposed model to forecast the demand of real estate between July 2016 and December 2017. The results were in agreement with estimates of specialists in the area, stating that in 2017 the real estate market will recover, however while these estimate that the market grows 10% in 2017, the model shows a growth of 19%. |