Akademik Veri Yönetim Sistemi
Annals of Financial Economics, 2025 (ESCI)
The autoregressive integrated moving average (ARIMA) model is widely used in time series analysis. However, this model is not able to capture nonlinear structures. The hybrid model brings different perspectives to forecasting analysis. This model uses a combination of ARIMA models and machine learning methods to overcome the deficiencies of single linear or nonlinear models. This paper aims to investigate whether hybrid models increase the forecasting accuracy compared to single models such as the ARIMA, random forest (RF), support vector regression (SVR) and artificial neural networks (ANNs). To this aim in this study, the Turkish Residential Property Price Index series were analyzed with the ARIMA, RF, SVR and ANN single models and the ARIMA-RF, ARIMA-SVR and ARIMA-ANN hybrid models. The methods were compared using forecasting evaluation criteria such as the RMSE and MAE. The findings proved that the ARIMA-RF has the highest out-of-sample forecasting accuracy compared with other models. Also, the Diebold–Mariano (DM) test was used for the comparison. DM test findings show that ARIMA-SVR also has significant prediction accuracy compared to other models. The results of the study suggest that hybrid models can be useful for time series forecasting.