Model Hybrid for Sales Forecast for the Housing Market of São Paulo

Abstract

Abstract This research proposes a combined model of time series for forecasting housing sales in the city of São Paulo. We used data referring to the time series of sales of residential units provided by SECOVI-SP. The Exponential Softening, Box-Jenkins and Artificial Neural Networks models are individually modelled, later these are combined through five forecast combination techniques. The techniques used are Arithmetic Mean, Geometric Mean, Harmonic Mean, Linear Regression and Principal Component Analysis. The measures of accuracy to measure the results obtained and to select the best model are the RMSE, MAPE and UTheil of forecast. The results showed that Linear Regression with an independent variable, being a combination of the SARIMA model (2,0,0)(2,0,0)12 and MLP/RNA (12,10,1), provided a satisfactory performance, with an RMSE of 368.74, MAPE of 19.2% and UTheil of 0.315. The combination of time series models allowed a significant increase in forecast performance. Finally, the model was validated, using it to predict housing sales. The results show that the model has a good fit, thus demonstrating that using a housing sales forecasting model helps industry professionals minimize error and make sales and launch decisions.