Forecasting Implementation of Hybrid Time Series and Artificial Neural Network Models

Abstract

This study implemented AR, SARIMA, and SETAR models and their hybrid with ANN using the Canadian lynx data. Implementing a SETAR-ANN has been shown to be successful in generating up to 10-step forecasts. The forecasting capability of AR-ANN and SARIMA-ANN applied to the same data are also investigated. The study shows that implementing hybridization does not guarantee better forecasting performance and accuracy, except for the case of SETAR-ANN. Despite results showing that the efficiency of the hybrid models may be worse than its components models, this study exhibits the promising predictive capability of time series, ANN, and its hybrid models.