Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

MambaStock: Selective state space model for stock prediction (2402.18959v1)

Published 29 Feb 2024 in cs.CE and q-fin.ST

Abstract: The stock market plays a pivotal role in economic development, yet its intricate volatility poses challenges for investors. Consequently, research and accurate predictions of stock price movements are crucial for mitigating risks. Traditional time series models fall short in capturing nonlinearity, leading to unsatisfactory stock predictions. This limitation has spurred the widespread adoption of neural networks for stock prediction, owing to their robust nonlinear generalization capabilities. Recently, Mamba, a structured state space sequence model with a selection mechanism and scan module (S6), has emerged as a powerful tool in sequence modeling tasks. Leveraging this framework, this paper proposes a novel Mamba-based model for stock price prediction, named MambaStock. The proposed MambaStock model effectively mines historical stock market data to predict future stock prices without handcrafted features or extensive preprocessing procedures. Empirical studies on several stocks indicate that the MambaStock model outperforms previous methods, delivering highly accurate predictions. This enhanced accuracy can assist investors and institutions in making informed decisions, aiming to maximize returns while minimizing risks. This work underscores the value of Mamba in time-series forecasting. Source code is available at https://github.com/zshicode/MambaStock.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (18)
  1. J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” CoRR, 2018.
  2. A. Gu and T. Dao, “Mamba: Linear-time sequence modeling with selective state spaces,” arXiv preprint arXiv:2312.00752, 2023.
  3. C. Jin, J. Gao, Z. Shi, and H. Zhang, “Attcry: Attention-based neural network model for protein crystallization prediction,” Neurocomputing, vol. 463, pp. 265–274, 2021.
  4. C. Jin, Z. Shi, C. Kang, K. Lin, and H. Zhang, “Tlcrys: Transfer learning based method for protein crystallization prediction,” International Journal of Molecular Sciences, vol. 23, p. 972, 1 2022.
  5. C. Jin, Z. Shi, W. Li, and Y. Guo, “Bidirectional lstm-crf attention-based model for chinese word segmentation,” arXiv preprint arXiv:2105.09681, 2021.
  6. C. Jin, Z. Shi, K. Lin, and H. Zhang, “Predicting mirna-disease association based on neural inductive matrix completion with graph autoencoders and self-attention mechanism,” Biomolecules, vol. 12, no. 1, p. 64, 2022.
  7. C. Jin, Z. Shi, H. Zhang, and Y. Yin, “Predicting lncrna–protein interactions based on graph autoencoders and collaborative training,” in IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Houston, USA, 9-12 December, 2021.
  8. R. E. Kalman, “A new approach to linear filtering and prediction problems,” Journal of Fluids Engineering, vol. 82, no. 1, pp. 35–44, 1960.
  9. D. P. Kingma and J. Ba, “Adam: a method for stochastic optimization,” in 3rd International Conference for Learning Representations(ICLR), 2015.
  10. K. Lin, X. Quan, C. Jin, Z. Shi, and J. Yang, “An interpretable double-scale attention model for enzyme protein class prediction based on transformer encoders and multi-scale convolutions,” Frontiers in Genetics, vol. 13, p. 885627, 2022.
  11. Z. Shi, “Incorporating transformer and lstm to kalman filter with em algorithm for state estimation,” arXiv preprint arXiv:2105.00250, 2021.
  12. ——, “Differential equation and probability inspired graph neural networks for latent variable learning,” arXiv preprint arXiv:2202.13800, 2022.
  13. Z. Shi, Y. Hu, G. Mo, and J. Wu, “Attention-based cnn-lstm and xgboost hybrid model for stock prediction,” arXiv preprint arXiv:2204.02623, 2022.
  14. Z. Shi and B. Li, “Graph neural networks and attention-based cnn-lstm for protein classification,” arXiv preprint arXiv:2204.09486, 2022.
  15. Z. Shi, H. Zhang, C. Jin, X. Quan, and Y. Yin, “A representation learning model based on variational inference and graph autoencoder for predicting lncrna-disease associations,” BMC Bioinformatics, vol. 22, p. 136, 2021.
  16. I. Sutskever, O. Vinyals, and Q. V. Le, “Sequence to sequence learning with neural networks,” in NeurIPS, 2014.
  17. A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in NeurIPS, 2017.
  18. G. Zhang, “Time series forecasting using a hybrid arima and neural network model,” Neurocomputing, vol. 50, pp. 159–175, 2003.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (1)
  1. Zhuangwei Shi (7 papers)
Citations (9)
View on Semantic Scholar

Summary

Overview of "MambaStock: Selective State Space Model for Stock Prediction"

The paper "MambaStock: Selective State Space Model for Stock Prediction" introduces a sophisticated approach to stock market prediction, addressing the limitations of traditional time series models in capturing the inherent nonlinearity and volatility of stock prices. The paper focuses on leveraging an advanced sequence modeling architecture, the Mamba model, to enhance prediction accuracy without the necessity of handcrafted feature engineering.

Introduction and Motivation

The stock market's complex behavior poses significant challenges in predicting stock prices. Traditional models like ARIMA have limitations due to their predisposition to linear assumptions, making them less effective in environments characterized by nonlinear patterns. In contrast, neural networks, with their powerful nonlinear generalization capabilities, provide a robust alternative for modeling such data. The Mamba model expands on this through its innovation in sequence modeling, specifically designed to handle the peculiarities of time series data like stock prices more effectively.

Methodology

The MambaStock model builds on the Mamba framework, which incorporates a structured state space sequence model with selection and scan modules, collectively known as S6. This architecture allows dynamic adaptation to input sequences and efficiently models temporal dependencies with a linear-time complexity. The model processes a range of financial indicators alongside historical stock price data to predict future stock movements. It makes use of the hyperbolic tangent activation function to ensure the predicted movement rates fall within a practical range, complementing the temporal dependencies captured by the model.

Empirical Findings

Empirical evaluations were conducted using stock data from entities such as China Merchants Bank and the Agricultural Bank of China. The MambaStock model demonstrated superior performance over traditional models and several advanced machine learning baselines, including LSTM, BiLSTM, and Transformer, by achieving lower Mean Absolute Error (MAE) and higher R-squared values. These results highlight its efficacy in accurately capturing the dynamics of stock price movements, thereby providing a reliable tool for investors looking to optimize their decision-making process while mitigating risks.

Comparative Analysis

When compared to a suite of existing methodologies—including the Kalman Filter, ARIMA, XGBoost, TL-KF, and AttCLX—MambaStock consistently showed enhanced prediction accuracy. Its success is attributed to its advanced state space model capable of capturing complex correlations within the data, setting it apart from simpler linear or hybrid models. The comparative superiority of MambaStock against both standalone and hybrid approaches underscores its potential as a leading technique in financial time-series forecasting.

Implications and Future Work

The paper underscores the importance of advanced sequence modeling techniques such as Mamba in financial forecasting. By eliminating the need for extensive preprocessing, the MambaStock model streamlines the predictive process, offering a practical tool for institutional and individual investors alike. Future research could explore the adaptability of the Mamba model to other financial instruments or integrate macroeconomic indicators to further enhance prediction robustness. Additionally, the scalability of the model to real-time stock trading scenarios presents a promising area for further investigation.

In conclusion, the research presented indicates that approaches like MambaStock, which leverage structured and selective state spaces, hold significant promise for improving the accuracy of stock market predictions, thus contributing to the broader field of financial analytics and artificial intelligence-driven decision support systems.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
Github Logo Streamline Icon: https://streamlinehq.com

GitHub

  1. GitHub - zshicode/MambaStock: MambaStock: Selective state space model for stock prediction (156 stars)
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets