HyBiLSTM: Multivariate Bitcoin Price Forecasting Using Hybrid Time-Series Models With Bidirectional LSTM
Despite their growing popularity in recent research, most hybrid models that harness the strengths of both classical time-series analysis and deep learning models have been explored within the univariate forecasting context. In the econometric domain, where exogenous factors play a crucial role; the...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2024-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10494511/ |
| _version_ | 1797206146084241408 |
|---|---|
| author | Anny Mardjo Chidchanok Choksuchat |
| author_facet | Anny Mardjo Chidchanok Choksuchat |
| author_sort | Anny Mardjo |
| collection | DOAJ |
| description | Despite their growing popularity in recent research, most hybrid models that harness the strengths of both classical time-series analysis and deep learning models have been explored within the univariate forecasting context. In the econometric domain, where exogenous factors play a crucial role; there is a pressing need for more studies focusing on multivariate forecasting. This paper introduces a novel hybrid model, HyBiLSTM. It integrates an ARIMAX GARCHX model for initial forecasting, followed by a second forecasting phase that addresses the residuals using a bidirectional long short-term memory model optimized through grey wolf optimization algorithm. The final forecast is a composite derived from both models. Three quantitative metrics (mean absolute error, root mean square error, and mean absolute percentage error) assessed the model performance using data that spanned social and economic variables from July 1, 2019, to December 31, 2022. The results revealed several key findings: 1) The addition of exogenous factors improved the performance of the ARIMA and GARCH models. 2) The BiLSTM variant outperformed other LSTM variants when combined with the ARIMAX GARCHX model. 3) An analysis using Shapley additive explanations indicated that bitcoin price was influenced by stock prices, Twitter volume, gold prices, and the Twitter sentiment index. 4) The presence of a structural break had a significant effect on the model’s forecasting accuracy. Beyond expanding the academic literature on hybrid models within a multivariate context, this offers valuable practical insights for investors. Specifically, it analyzes various factors that could serve as early indicators of bitcoin price fluctuations. |
| first_indexed | 2024-04-24T09:02:22Z |
| format | Article |
| id | doaj.art-ee47a1e52c65449b854df36367344b48 |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-04-24T09:02:22Z |
| publishDate | 2024-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-ee47a1e52c65449b854df36367344b482024-04-15T23:00:32ZengIEEEIEEE Access2169-35362024-01-0112507925080810.1109/ACCESS.2024.338602910494511HyBiLSTM: Multivariate Bitcoin Price Forecasting Using Hybrid Time-Series Models With Bidirectional LSTMAnny Mardjo0https://orcid.org/0000-0002-8678-8278Chidchanok Choksuchat1https://orcid.org/0000-0002-8241-7090College of Digital Science, Prince of Songkla University, Hat Yai, ThailandDivision of Computational Science, Faculty of Science, Prince of Songkla University, Hat Yai, ThailandDespite their growing popularity in recent research, most hybrid models that harness the strengths of both classical time-series analysis and deep learning models have been explored within the univariate forecasting context. In the econometric domain, where exogenous factors play a crucial role; there is a pressing need for more studies focusing on multivariate forecasting. This paper introduces a novel hybrid model, HyBiLSTM. It integrates an ARIMAX GARCHX model for initial forecasting, followed by a second forecasting phase that addresses the residuals using a bidirectional long short-term memory model optimized through grey wolf optimization algorithm. The final forecast is a composite derived from both models. Three quantitative metrics (mean absolute error, root mean square error, and mean absolute percentage error) assessed the model performance using data that spanned social and economic variables from July 1, 2019, to December 31, 2022. The results revealed several key findings: 1) The addition of exogenous factors improved the performance of the ARIMA and GARCH models. 2) The BiLSTM variant outperformed other LSTM variants when combined with the ARIMAX GARCHX model. 3) An analysis using Shapley additive explanations indicated that bitcoin price was influenced by stock prices, Twitter volume, gold prices, and the Twitter sentiment index. 4) The presence of a structural break had a significant effect on the model’s forecasting accuracy. Beyond expanding the academic literature on hybrid models within a multivariate context, this offers valuable practical insights for investors. Specifically, it analyzes various factors that could serve as early indicators of bitcoin price fluctuations.https://ieeexplore.ieee.org/document/10494511/LSTMARIMAXGARCHXbitcoinSHAPbidirectional LSTM |
| spellingShingle | Anny Mardjo Chidchanok Choksuchat HyBiLSTM: Multivariate Bitcoin Price Forecasting Using Hybrid Time-Series Models With Bidirectional LSTM IEEE Access LSTM ARIMAX GARCHX bitcoin SHAP bidirectional LSTM |
| title | HyBiLSTM: Multivariate Bitcoin Price Forecasting Using Hybrid Time-Series Models With Bidirectional LSTM |
| title_full | HyBiLSTM: Multivariate Bitcoin Price Forecasting Using Hybrid Time-Series Models With Bidirectional LSTM |
| title_fullStr | HyBiLSTM: Multivariate Bitcoin Price Forecasting Using Hybrid Time-Series Models With Bidirectional LSTM |
| title_full_unstemmed | HyBiLSTM: Multivariate Bitcoin Price Forecasting Using Hybrid Time-Series Models With Bidirectional LSTM |
| title_short | HyBiLSTM: Multivariate Bitcoin Price Forecasting Using Hybrid Time-Series Models With Bidirectional LSTM |
| title_sort | hybilstm multivariate bitcoin price forecasting using hybrid time series models with bidirectional lstm |
| topic | LSTM ARIMAX GARCHX bitcoin SHAP bidirectional LSTM |
| url | https://ieeexplore.ieee.org/document/10494511/ |
| work_keys_str_mv | AT annymardjo hybilstmmultivariatebitcoinpriceforecastingusinghybridtimeseriesmodelswithbidirectionallstm AT chidchanokchoksuchat hybilstmmultivariatebitcoinpriceforecastingusinghybridtimeseriesmodelswithbidirectionallstm |