Abstract
This paper examines the use of machine learning methods in modeling and forecasting time series with long memory through GARMA. By employing rigorous model selection criteria through simulation study, we find that the hybrid GARMA-LSTM model outperforms traditional approaches in forecasting long-memory time series. This characteristic is confirmed using popular datasets such as sunspot data and Australian beer production data. This approach provides a robust framework for accurate and reliable forecasting in long-memory time series. Additionally, we compare the GARMA-LSTM model with other implemented models, such as GARMA, TBATS, ARIMA, and ANN, highlighting its ability to address both long-memory and non-linear dynamics. Finally, we discuss the representativeness of the datasets selected and the adaptability of the proposed hybrid model to various time series scenarios.
Document Type
Journal Article
Volume
13
Issue
1
School
School of Business and Law
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Publisher
MDPI
Identifier
David E. Allen: https://orcid.org/0000-0001-7782-0865
Comments
Gadhi, A. H. A., Peiris, S., Allen, D. E., & Hunt, R. (2025). Optimal time series forecasting through the GARMA model. Econometrics, 13(1). https://doi.org/10.3390/econometrics13010003