Document Type : Research Paper
Authors
1
Ph.D. Student, Department of Economic, Faculty of Management and Economics, Islamic Azad University, Science and Research Branch, Tehran, Iran.
2
Prof. Department of Economic, Faculty of Economics, Shahid Beheshti University, Tehran, Iran.
3
Assistant Prof, Department of Economic, Faculty of Management and Economic, Science and Research Branch, Islamic Azad University, Tehran, Iran.
4
Assistant Prof, Department of Economic, Faculty of Humanities, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran
10.30465/jnet.2025.47420.2096
Abstract
This study aims to predict the volatility of cryptocurrencies, which is a crucial and difficult task. Considering the nonlinear characteristics and time-varying features of various factors that affect the price of cryptocurrencies, this study uses a novel method that combines two powerful techniques: the GARCH model and the LSTM network. The GARCH model captures the statistical patterns of price fluctuations and provides GARCH forecasts. The second technique is machine learning models. Previous studies have shown that combining GARCH models and machine learning can improve the volatility prediction in various markets, such as energy, metals and stocks markets. This study tests this hypothesis in the cryptocurrency market by using different LSTM models to predict the volatility of a selected cryptocurrency. It also creates hybrid models that feed the outputs of different GARCH variants, with three different assumptions for the residual distribution, to the LSTM network. In other words, the GARCH models act as feature extractors and the LSTM models use the extracted features as input to generate future volatility. The results show that the LSTM models alone outperform the GARCH models with any residual distribution, and that the GARCH models as features can enhance the prediction performance of the LSTM models.
Keywords
)