Paper Type
Complete
Paper Number
1511
Description
Recognizing the need for greater technology-driven innovation in finance, this study introduces a pioneering Neural Ordinary Differential Equations (NODE) model, engineered explicitly for stock price forecasting. Unlike discrete-time neural networks, NODE interprets data transformations in a continuous manner, enabling the efficient handling of irregular time series and continuous-time dynamics. By incorporating Phase Space Reconstruction, NODE then transforms original time series data into a higher-dimensional space, unveiling complex temporal relationships that conventional models often miss. The empirical evidence, obtained from applying NODE to various stock sectors, illustrates a significant average reduction in MAE and RMSE evaluation metrics by over 76% compared to baseline models. Likewise, NODE consistently showcases strong adaptability to diverse chaotic conditions over DNN-based architectures in capturing long-term dependencies, and seasonal or cyclical patterns. The accuracy of NODE with chaotic systems, such as Lorenz and Mackey-Glass, finally demonstrates its potential applicability across a broader spectrum of time-series forecasting fields.
Recommended Citation
Nguyen, Anh Quoc; Ha, Son; and Thai, Hieu, "Phase Space Reconstructed Neural Ordinary Differential Equations Model for Stock Price Forecasting" (2024). PACIS 2024 Proceedings. 4.
https://aisel.aisnet.org/pacis2024/track01_aibussoc/track01_aibussoc/4
Phase Space Reconstructed Neural Ordinary Differential Equations Model for Stock Price Forecasting
Recognizing the need for greater technology-driven innovation in finance, this study introduces a pioneering Neural Ordinary Differential Equations (NODE) model, engineered explicitly for stock price forecasting. Unlike discrete-time neural networks, NODE interprets data transformations in a continuous manner, enabling the efficient handling of irregular time series and continuous-time dynamics. By incorporating Phase Space Reconstruction, NODE then transforms original time series data into a higher-dimensional space, unveiling complex temporal relationships that conventional models often miss. The empirical evidence, obtained from applying NODE to various stock sectors, illustrates a significant average reduction in MAE and RMSE evaluation metrics by over 76% compared to baseline models. Likewise, NODE consistently showcases strong adaptability to diverse chaotic conditions over DNN-based architectures in capturing long-term dependencies, and seasonal or cyclical patterns. The accuracy of NODE with chaotic systems, such as Lorenz and Mackey-Glass, finally demonstrates its potential applicability across a broader spectrum of time-series forecasting fields.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
AI