Authors :
Ashraf Nidhal Mohammed Mohammed
Volume/Issue :
Volume 9 - 2024, Issue 12 - December
Google Scholar :
https://tinyurl.com/2t47kb5w
Scribd :
https://tinyurl.com/2jx37djn
DOI :
https://doi.org/10.5281/zenodo.14540291
Abstract :
The cryptocurrency market, characterized by
extreme volatility and complex dynamics, presents
significant challenges for accurate price prediction. This
study introduces a novel hybrid predictive model that
integrates Exponential Generalized Autoregressive
Conditional Heteroscedasticity (EGARCH) with Long
Short-Term Memory (LSTM) networks, augmented by
Explainable AI (XAI) techniques such as SHAP (SHapley
Additive exPlanations). The EGARCH model captures
statistical properties like volatility clustering and leverage
effects, while the LSTM component effectively models
nonlinear dependencies and long-term temporal patterns.
SHAP enhances interpretability by quantifying feature
contributions, offering actionable insights into the
decision-making process.
Using historical data from major cryptocurrencies
such as Bitcoin and Ethereum, the hybrid model
outperforms standalone EGARCH and LSTM models
across diverse market conditions, achieving up to 95%
accuracy. Visual analyses, including performance
comparisons and SHAP-based feature importance
graphs, provide clarity on prediction outcomes and error
patterns. By addressing limitations in both accuracy and
transparency, this research advances hybrid financial
modeling methodologies and offers practical tools for
traders, investors, and policymakers. These findings
enable data-driven decision-making and effective risk
management in highly volatile markets.
Keywords :
Hybrid Models, EGARCH, LSTM, Explainable AI (XAI), SHAP.
References :
- García-Medina, A., & Aguayo-Moreno, E. (2023). LSTM–GARCH hybrid model for the prediction of volatility in cryptocurrency portfolios. Computational Economics, 63, 1511–1542. https://doi.org/10.1007/s10614-023-10373-8
- Vidal, T., & Kristjanpoller, W. (2022). Forecasting Bitcoin volatility using hybrid GARCH models with machine learning. Risks, 10(12), 237. https://doi.org/10.3390/risks10120237
- Zhang, Y., et al. (2021). A hybrid model integrating LSTM and GARCH for Bitcoin price prediction. In 2021 IEEE International Conference on Big Data (Big Data) (pp. 1933–1942). https://doi.org/10.1109/BigData52589.2021.9671860
- Arsenault, P.-D., Wang, S., & Patenaude, J.-M. (2024). A survey of explainable artificial intelligence (XAI) in financial time series forecasting. arXiv preprint arXiv:2407.15909. https:// arxiv.org/abs/2407.15909
- Zhang, Y., et al. (2024). ShapTime: A general XAI approach for explainable time series forecasting. In Intelligent Systems and Applications (pp. 659–673). Springer. https://doi.org/ 10.1007/978-3-031-47721-8_45
- Ade, M., et al. (2023). Explainable AI in financial time series forecasting: Interpretability of deep learning models compared to traditional techniques. ResearchGate. https:// www.researchgate.net/publication/384665323.
- Misheva, B. H., & Osterrieder, J. (2023). A hypothesis on good practices for AI-based systems for financial time series forecasting: Towards domain-driven XAI methods. arXiv preprint arXiv:2311.07513.https://arxiv.org/abs/2311.07513
- TensorFlow Developers. (2023). TensorFlow: Open source machine learning platform. Retrieved from https://www.tensorflow.org
- SHAP Documentation. (2023). Explainable AI with SHAP. Retrieved from https:// shap.readthedocs.io
- Bui, C. (2021). Bitcoin volatility forecasting: GARCH and multivariate LSTM models. GitHub Repository. Retrieved from https://github.com/chibui191/bitcoin_volatility_forecasting Springer. (2023). A hybrid LSTM-GARCH model for cryptocurrency volatility prediction.
- Springer Computational Finance, 12(2), 123–145. https://doi.org/10.1007/s10614-023-10373-8
- Wang, J., & Wu, H. (2023). The role of deep learning in financial forecasting: A review. Journal of Financial Data Science, 5(1), 55–75. https://doi.org/10.3905/jfds.2023.1
- MDPI. (2022). Advancements in hybrid financial forecasting models: Integrating GARCH and deep learning. MDPI Risks, 10(3), 45. https://doi.org/10.3390/risks10030045
- Hoelli, J. (2022). Awesome time series explainability: A curated list of XAI for time series data. GitHub Repository. Retrieved from https://github.com/JHoelli/Awesome-Time-Series- Explainability
- Le Menestrel, T. (2021). A Python implementation of a hybrid LSTM-GARCH model for volatility forecasting. GitHub Repository. Retrieved from https://github.com/tlemenestrel/ LSTM_GARCH
- Saluja, R., et al. (2021). Towards a rigorous evaluation of explainability for multivariate time series. arXiv preprint arXiv:2104.04075. https://arxiv.org/abs/2104.04075
- Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press. Retrieved from https://www.deeplearningbook.org/
- Misra, D., et al. (2023). The application of explainable AI in crypto-finance: A roadmap. Journal of Blockchain Research, 4(2), 89–104. https://doi.org/10.1108/JBR202304
- CoinMarketCap. (n.d.). Cryptocurrency market data. Retrieved from https:// www.coinmarketcap.com
- Yahoo Finance. (n.d.). Historical market data. Retrieved from https://www.finance.yahoo.com
- Emekss. (2022). Ethereum price prediction using hybrid LSTM and GRU models. GitHub Repository. Retrieved from https://github.com/emekss/ Ethereum_Price_Prediction_Using_Hybrid_LSTM_and_GRU
- Emrah, A., et al. (2022). Advances in financial volatility modeling using hybrid machine learning techniques. Computational Statistics, 40(4), 567–589. https://doi.org/10.1007/s11222-022-10005-9
- SHAP Research Group. (2023). Advances in explainable AI for financial applications. Journal of Financial Machine Learning, 6(3), 10–32. https://doi.org/10.1007/s12056-023-2045
The cryptocurrency market, characterized by
extreme volatility and complex dynamics, presents
significant challenges for accurate price prediction. This
study introduces a novel hybrid predictive model that
integrates Exponential Generalized Autoregressive
Conditional Heteroscedasticity (EGARCH) with Long
Short-Term Memory (LSTM) networks, augmented by
Explainable AI (XAI) techniques such as SHAP (SHapley
Additive exPlanations). The EGARCH model captures
statistical properties like volatility clustering and leverage
effects, while the LSTM component effectively models
nonlinear dependencies and long-term temporal patterns.
SHAP enhances interpretability by quantifying feature
contributions, offering actionable insights into the
decision-making process.
Using historical data from major cryptocurrencies
such as Bitcoin and Ethereum, the hybrid model
outperforms standalone EGARCH and LSTM models
across diverse market conditions, achieving up to 95%
accuracy. Visual analyses, including performance
comparisons and SHAP-based feature importance
graphs, provide clarity on prediction outcomes and error
patterns. By addressing limitations in both accuracy and
transparency, this research advances hybrid financial
modeling methodologies and offers practical tools for
traders, investors, and policymakers. These findings
enable data-driven decision-making and effective risk
management in highly volatile markets.
Keywords :
Hybrid Models, EGARCH, LSTM, Explainable AI (XAI), SHAP.