Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CryptoMamba: Leveraging State Space Models for Accurate Bitcoin Price Prediction

Mohammad Shahab Sepehri, Asal Mehradfar, Mahdi Soltanolkotabi, Salman Avestimehr

TL;DR

CryptoMamba introduces a Mamba-based State Space Model designed for financial time-series forecasting, addressing Bitcoin price volatility and regime shifts. By stacking C-Blocks of CMBlocks with input-dependent dynamics and a Merge head, it captures long-range dependencies more efficiently than traditional RNNs or Transformers. In experiments against LSTM, Bi-LSTM, GRU, iTransformer, and S-Mamba, CryptoMamba achieves superior predictive accuracy and, when coupled with trading strategies, translates forecasts into robust real-world returns, notably with the volume-enhanced variant achieving RMSE $1598.1$, MAPE $2.034$, and MAE $1120.7$. The architecture is compact ($\approx 136k$ parameters) and computationally efficient (training $\approx 29$ minutes; inference $\approx 1.17$ ms/sample), supporting deployment in latency-sensitive environments and suggesting broad applicability to other assets and trading settings.

Abstract

Predicting Bitcoin price remains a challenging problem due to the high volatility and complex non-linear dynamics of cryptocurrency markets. Traditional time-series models, such as ARIMA and GARCH, and recurrent neural networks, like LSTMs, have been widely applied to this task but struggle to capture the regime shifts and long-range dependencies inherent in the data. In this work, we propose CryptoMamba, a novel Mamba-based State Space Model (SSM) architecture designed to effectively capture long-range dependencies in financial time-series data. Our experiments show that CryptoMamba not only provides more accurate predictions but also offers enhanced generalizability across different market conditions, surpassing the limitations of previous models. Coupled with trading algorithms for real-world scenarios, CryptoMamba demonstrates its practical utility by translating accurate forecasts into financial outcomes. Our findings signal a huge advantage for SSMs in stock and cryptocurrency price forecasting tasks.

CryptoMamba: Leveraging State Space Models for Accurate Bitcoin Price Prediction

TL;DR

CryptoMamba introduces a Mamba-based State Space Model designed for financial time-series forecasting, addressing Bitcoin price volatility and regime shifts. By stacking C-Blocks of CMBlocks with input-dependent dynamics and a Merge head, it captures long-range dependencies more efficiently than traditional RNNs or Transformers. In experiments against LSTM, Bi-LSTM, GRU, iTransformer, and S-Mamba, CryptoMamba achieves superior predictive accuracy and, when coupled with trading strategies, translates forecasts into robust real-world returns, notably with the volume-enhanced variant achieving RMSE , MAPE , and MAE . The architecture is compact ( parameters) and computationally efficient (training minutes; inference ms/sample), supporting deployment in latency-sensitive environments and suggesting broad applicability to other assets and trading settings.

Abstract

Predicting Bitcoin price remains a challenging problem due to the high volatility and complex non-linear dynamics of cryptocurrency markets. Traditional time-series models, such as ARIMA and GARCH, and recurrent neural networks, like LSTMs, have been widely applied to this task but struggle to capture the regime shifts and long-range dependencies inherent in the data. In this work, we propose CryptoMamba, a novel Mamba-based State Space Model (SSM) architecture designed to effectively capture long-range dependencies in financial time-series data. Our experiments show that CryptoMamba not only provides more accurate predictions but also offers enhanced generalizability across different market conditions, surpassing the limitations of previous models. Coupled with trading algorithms for real-world scenarios, CryptoMamba demonstrates its practical utility by translating accurate forecasts into financial outcomes. Our findings signal a huge advantage for SSMs in stock and cryptocurrency price forecasting tasks.
Paper Structure (16 sections, 3 equations, 4 figures, 5 tables, 3 algorithms)

This paper contains 16 sections, 3 equations, 4 figures, 5 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background
  4. Methodology
  5. Dataset
  6. CryptoMamba Architecture
  7. Experiments
  8. Setup
  9. Evaluation
  10. Efficiency
  11. Application in Real-World Trading
  12. Conclusion and Future Work
  13. Metrics
  14. Regression Metrics.
  15. Maximum Drawdown (MDD).
  16. ...and 1 more sections

Figures (4)

  • Figure 1: CryptoMamba model consists of several C-Blocks followed by a Merge block. In each C-Block, we have several CMBlock and an MLP at the end.
  • Figure 2: Forecasting results for all models (a) CryptoMamba, (b) LSTM, (c) Bi-LSTM, (d) GRU, (e) iTransformer, and (f) S-Mamba on the training, validation, and test sets without using volume data. Non-Mamba models struggle to capture large price fluctuations, often underperforming during periods of high volatility.
  • Figure 3: Forecasting results for all models (a) CryptoMamba, (b) LSTM, (c) Bi-LSTM, (d) GRU, (e) iTransformer, and (f) S-Mamba on the training, validation, and test sets with volume data included as an additional feature. While volume data helps non-Mamba models slightly, they still face challenges in accurately predicting large price changes, especially during volatile periods.
  • Figure 4: Net worth over the test period using three trading strategies, Vanilla, Smart, and Extended Smart, under both volume-exclusive (top row) and volume-inclusive (bottom row) setups. CryptoMamba consistently achieves the highest final balance across all configurations.