pith. sign in

arxiv: 2606.27670 · v1 · pith:N3TWMPBZnew · submitted 2026-06-26 · 💻 cs.CE · q-fin.ST

CryptoGAT: Are Time Series Models Effective for Cryptocurrency Forecasting?

Yu Peng , Matloob Khushi , Josiah Poon This is my paper

Pith reviewed 2026-06-29 02:44 UTC · model grok-4.3

classification 💻 cs.CE q-fin.ST
keywords cryptocurrencyprice predictiongraph attention networktime series modelsvolatilitycross-asset dependenciesfinancial forecasting
0
0 comments X

The pith

Time series models struggle to learn from volatile cryptocurrency prices, while a graph attention network that models cross-asset dependencies outperforms them.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper questions the effectiveness of time series models such as LSTM, GRU, and Transformers for cryptocurrency price prediction using only price data. It argues that extreme volatility and wild swings in crypto markets prevent these models from extracting useful temporal patterns, unlike their performance on stock data. To test the claim, the authors introduce CryptoGAT, a lightweight graph attention network that reframes the task as learning relationships across multiple cryptocurrencies instead of modeling time sequences alone. Experiments on real benchmarks show CryptoGAT achieving higher accuracy than standard forecasting methods. The work also includes studies comparing signal predictability and cross-asset links in crypto versus stock markets.

Core claim

In pure price-based cryptocurrency prediction, time series models have difficulty learning effective information when faced with data exhibiting extreme volatility and wild swings. CryptoGAT, a lightweight Graph Attention Network, recasts cryptocurrency pure price prediction as a cross-asset graph problem rather than a temporal modeling task and outperforms various state-of-the-art forecasting methods. Comprehensive empirical studies reveal fundamental differences in the predictability of the signal and cross-asset dependencies between stock and cryptocurrency markets.

What carries the argument

CryptoGAT, a lightweight Graph Attention Network that models cryptocurrency prices as nodes in a graph to capture cross-asset dependencies.

If this is right

  • Cryptocurrency forecasting should prioritize cross-asset relationships over pure temporal modeling.
  • Time series methods require reevaluation when applied to assets with extreme volatility.
  • Graph-based approaches open new research directions for pure price prediction in cryptocurrencies.
  • Empirical differences in predictability between stocks and cryptos call for separate modeling strategies.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same graph reformulation could be applied to other noisy financial series such as commodities to test if cross-asset modeling helps there as well.
  • The results imply that simultaneous correlations across assets dominate sequential patterns within individual crypto prices.
  • Controlled tests that equalize compute and tuning across model families would clarify whether the graph component is the decisive factor.

Load-bearing premise

The performance gains of CryptoGAT arise specifically from modeling cross-asset dependencies via the graph structure rather than from differences in model capacity, hyperparameter tuning, or data selection.

What would settle it

An ablation where the graph structure is removed from CryptoGAT while keeping other components fixed, or where time series models receive the same cross-asset price inputs without graph attention, to check whether the performance margin disappears.

Figures

Figures reproduced from arXiv: 2606.27670 by Josiah Poon, Matloob Khushi, Yu Peng.

Figure 1
Figure 1. Figure 1: Left: Overview of the StockMixer. Right: The multi-head attention mechanism of the CryptoGAT model, where h1 obtains h ′ 1 by calculating the weights of its neighbors. C. A Simple Yet Effective Baseline In existing financial time-series forecasting solutions, all the compared baselines (e.g., LSTM, GRU, Transformer) model each cryptocurrency independently, which ignores the strong cross-asset correlations … view at source ↗
Figure 2
Figure 2. Figure 2: IC and annualized Sharpe ratio (Y-axis) across different lookback [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: The comparison of Correlation Strength, Graph Topological, and [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
read the original abstract

Cryptocurrency price prediction is a significant challenge in quantitative investment. In recent years, time series models have made significant progress in financial forecasting tasks, especially in the stock market. Despite the growing performance over the past few years, we question the validity of this line of research in cryptocurrency prediction. Specifically, time series models (e.g., LSTM, GRU, and Transformers) are effective at extracting temporal relationships in stock market data. However, in pure price-based cryptocurrency prediction, facing data with extreme volatility and wild swings, time series models have difficulty learning effective information. To validate our claim, we propose CryptoGAT, a lightweight Graph Attention Network that recasts cryptocurrency pure price prediction as a cross-asset graph problem rather than a temporal modeling task. Extensive experiments on real cryptocurrency benchmarks demonstrate that our proposed CryptoGAT outperforms various state-of-the-art forecasting methods with a notable margin. Moreover, we conduct comprehensive empirical studies to explore the fundamental differences exposed by time series models in stock and cryptocurrency prediction: differences in predictability of the signal and cross-asset dependencies. This finding opens up new research directions for the cryptocurrency pure price prediction task and inspires further graph-based exploration in the field. The source code is available at https://github.com/FanBroWell/CryptoGAT

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript claims that time series models (LSTM, GRU, Transformers) struggle to extract useful signals from pure price-based cryptocurrency data due to extreme volatility and wild swings, in contrast to their effectiveness in stock markets. It introduces CryptoGAT, a lightweight Graph Attention Network that reformulates cryptocurrency forecasting as a cross-asset graph problem rather than a temporal task, and reports that it outperforms various state-of-the-art methods with a notable margin on real benchmarks. The work also includes empirical studies exploring differences in signal predictability and cross-asset dependencies between stock and crypto markets, with source code released publicly.

Significance. If the reported gains are attributable to the graph attention structure after proper controls, the result would support shifting cryptocurrency pure-price forecasting research toward graph-based models that leverage inter-asset relations, while highlighting domain-specific differences from equities. The public availability of source code at the cited GitHub repository is a clear strength for reproducibility.

major comments (2)
  1. [Abstract and experimental results section] Abstract and experimental results section: the central claim that time series models 'have difficulty learning effective information' while CryptoGAT succeeds due to cross-asset graph edges requires controlled comparisons. The abstract labels CryptoGAT 'lightweight' and reports 'notable margin' gains, but provides no evidence that baselines were capacity-matched, re-tuned with equivalent hyperparameter search effort, or trained under identical schedules and asset universes; any systematic difference in these dimensions would explain the margin without validating the volatility-vs-graph narrative.
  2. [Empirical studies section] The assertion that the performance difference exposes 'fundamental differences' in predictability and cross-asset dependencies is load-bearing for the paper's narrative, yet without ablation studies isolating the graph edges from other architectural choices or data-handling differences, the empirical studies cannot rule out alternative explanations.
minor comments (2)
  1. [Abstract] The GitHub link is provided; this supports reproducibility and should be retained.
  2. [Experimental setup] Specify the exact cryptocurrency datasets, number of assets, prediction horizons, and evaluation metrics (e.g., RMSE, MAE) used in the benchmarks for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below with clarifications on our experimental setup and commit to revisions that strengthen the evidence for our claims.

read point-by-point responses

  1. Referee: [Abstract and experimental results section] Abstract and experimental results section: the central claim that time series models 'have difficulty learning effective information' while CryptoGAT succeeds due to cross-asset graph edges requires controlled comparisons. The abstract labels CryptoGAT 'lightweight' and reports 'notable margin' gains, but provides no evidence that baselines were capacity-matched, re-tuned with equivalent hyperparameter search effort, or trained under identical schedules and asset universes; any systematic difference in these dimensions would explain the margin without validating the volatility-vs-graph narrative.

    Authors: We agree that explicit documentation of controls is necessary to support the central claim. All models, including the time series baselines, were trained on identical asset universes and data splits with hyperparameter selection performed via the same validation-set grid search procedure and identical training schedules. CryptoGAT is labeled lightweight to emphasize its parameter efficiency relative to large transformers. In the revision we will add a dedicated subsection detailing the hyperparameter grids, reporting parameter counts for capacity matching, and confirming the shared experimental protocol. This will make the controlled nature of the comparisons transparent. revision: yes

  2. Referee: [Empirical studies section] The assertion that the performance difference exposes 'fundamental differences' in predictability and cross-asset dependencies is load-bearing for the paper's narrative, yet without ablation studies isolating the graph edges from other architectural choices or data-handling differences, the empirical studies cannot rule out alternative explanations.

    Authors: The empirical studies section reports head-to-head results of the same temporal architectures on stock versus cryptocurrency price series, exposing consistent differences in signal predictability. While these comparisons already control for model architecture and data-handling pipeline, we acknowledge that further isolation of the graph component would strengthen the argument. The revision will therefore incorporate ablation variants of CryptoGAT (e.g., with attention disabled or edges randomized) to quantify the contribution of the cross-asset graph structure and to address potential alternative explanations. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical comparison stands on reported experiments, not self-referential definitions or citations.

full rationale

The paper advances an empirical claim that time-series models struggle on volatile crypto price series while a graph-attention recasting succeeds. No equations, fitted parameters renamed as predictions, self-citation chains, or uniqueness theorems appear in the provided text. The outperformance is presented as an experimental result on benchmarks rather than a quantity forced by construction from the inputs. The central narrative therefore remains self-contained against external benchmarks and does not reduce to any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no free parameters, axioms, or invented entities can be identified from the provided text.

pith-pipeline@v0.9.1-grok · 5758 in / 1077 out tokens · 38918 ms · 2026-06-29T02:44:11.556225+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references

  1. [1]

    Bitcoin: A peer-to-peer electronic cash system,

    S. Nakamoto, “Bitcoin: A peer-to-peer electronic cash system,” white paper, 2008

    2008

  2. [2]

    Understanding cryptocurrencies,

    W. K. H ¨ardle, C. R. Harvey, and R. C. G. Reule, “Understanding cryptocurrencies,”J. Financial Econometrics, vol. 18, no. 2, pp. 181- 208, 2020

    2020

  3. [3]

    Blockchain governance—a new way of organizing collaborations?

    F. Lumineau, W. Wang, and O. Schilke, “Blockchain governance—a new way of organizing collaborations?”Organization Sci., vol. 32, no. 2, pp. 500-521, 2021

    2021

  4. [4]

    Cryptocurrencies as a financial asset: A systematic analysis,

    S. Corbet, B. Lucey, A. Urquhart, and L. Yarovaya, “Cryptocurrencies as a financial asset: A systematic analysis,”Int. Rev. Financial Anal., vol. 62, pp. 182-199, 2019

    2019

  5. [5]

    Blockchain technology in healthcare: A systematic review,

    C. C. Agbo, Q. H. Mahmoud, and J. M. Eklund, “Blockchain technology in healthcare: A systematic review,”Healthcare, vol. 7, no. 2, p. 56, 2019

    2019

  6. [6]

    Crypto sector breaches $4 trillion in market value during pivotal week,

    Reuters, “Crypto sector breaches $4 trillion in market value during pivotal week,”Reuters, Jul. 2025

    2025

  7. [7]

    Empirical asset pricing via machine learning,

    S. Gu, B. Kelly, and D. Xiu, “Empirical asset pricing via machine learning,”Rev. Financial Stud., vol. 33, no. 5, pp. 2223-2273, 2020

    2020

  8. [8]

    Machine learning stock market prediction studies: Review and research direc- tions,

    T. J. Strader, J. J. Rozycki, T. H. Root, and Y .-H. J. Huang, “Machine learning stock market prediction studies: Review and research direc- tions,”J. Int. Technology and Information Management, vol. 28, no. 4, pp. 63-83, 2020

    2020

  9. [9]

    Short and long term stock trend prediction using decision tree,

    R. A. Kamble, “Short and long term stock trend prediction using decision tree,” inProc. Int. Conf. Intelligent Computing and Control Systems (ICICCS), 2017, pp. 1371-1375

    2017

  10. [10]

    Decision support system for stock trading using multiple indicators decision tree,

    F. X. S. D. Nugroho, T. B. Adji, and S. Fauziati, “Decision support system for stock trading using multiple indicators decision tree,” in Proc. 1st Int. Conf. Information Technology, Computer, and Electrical Engineering, 2014, pp. 291-296

    2014

  11. [11]

    Semantic frames to predict stock price movement,

    B. Xie, R. J. Passonneau, L. Wu, and G. G. Creamer, “Semantic frames to predict stock price movement,” inProc. 51st Annu. Meeting Assoc. Computational Linguistics, vol. 1, 2013, pp. 873-883

    2013

  12. [12]

    Stock price prediction using k-nearest neighbor (kNN) algorithm,

    K. Alkhatib, H. Najadat, I. Hmeidi, and M. K. A. Shatnawi, “Stock price prediction using k-nearest neighbor (kNN) algorithm,”Int. J. Business, Humanities and Technology, vol. 3, no. 3, pp. 32-44, 2013

    2013

  13. [13]

    An innovative neural network approach for stock market prediction,

    X. Pang, Y . Zhou, P. Wang, W. Lin, and V . Chang, “An innovative neural network approach for stock market prediction,”J. Supercomput., vol. 76, no. 3, pp. 2098-2118, 2020

    2098

  14. [14]

    Financial time series forecasting with deep learning: A systematic literature review: 2005-2019,

    O. B. Sezer, M. U. Gudelek, and A. M. Ozbayoglu, “Financial time series forecasting with deep learning: A systematic literature review: 2005-2019,”Appl. Soft Comput., vol. 90, p. 106181, 2020

    2005

  15. [15]

    Simple technical trading rules and the stochastic properties of stock returns,

    W. Brock, J. Lakonishok, and B. LeBaron, “Simple technical trading rules and the stochastic properties of stock returns,”J. Finance, vol. 47, no. 5, pp. 1731-1764, 1992

    1992

  16. [16]

    Risks and returns of cryptocurrency,

    Y . Liu and A. Tsyvinski, “Risks and returns of cryptocurrency,”Rev. Financial Stud., vol. 34, no. 6, pp. 2689-2727, 2021

    2021

  17. [17]

    A critical investigation of cryptocurrency data and analysis,

    C. Alexander and M. Dakos, “A critical investigation of cryptocurrency data and analysis,”Quant. Finance, vol. 20, no. 2, pp. 173-188, 2020

    2020

  18. [18]

    Cryptocurrency price pre- diction algorithms: A survey and future directions,

    D. L. John, S. Binnewies, and B. Stantic, “Cryptocurrency price pre- diction algorithms: A survey and future directions,”Forecasting, vol. 6, no. 3, pp. 637-671, 2024

    2024

  19. [19]

    Nothing but noise? Price discovery across cryptocurrency exchanges,

    T. Dimpfl and F. J. Peter, “Nothing but noise? Price discovery across cryptocurrency exchanges,”J. Financial Markets, vol. 54, p. 100584, 2021

    2021

  20. [20]

    CryptoMixer: Fine-grained market information-aware MLP networks for individual cryptocurrency trading prediction,

    T. Feng, Z. Shen, X. Zhao, X. Lu, and Y . Zhou, “CryptoMixer: Fine-grained market information-aware MLP networks for individual cryptocurrency trading prediction,” inProc. 31st ACM SIGKDD Conf. Knowledge Discovery and Data Mining, 2025, pp. 603-614

    2025

  21. [21]

    Deep learning and NLP in cryptocurrency forecasting: integrating financial, blockchain, and social media data,

    V . Gurgul, S. Lessmann, and W. K. H ¨ardle, “Deep learning and NLP in cryptocurrency forecasting: integrating financial, blockchain, and social media data,”Int. J. Forecast., vol. 41, no. 4, pp. 1666-1695, 2025

    2025

  22. [22]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural Comput., vol. 9, no. 8, pp. 1735-1780, 1997

    1997

  23. [23]

    A time series is worth 64 words: Long-term forecasting with transformers,

    Y . Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam, “A time series is worth 64 words: Long-term forecasting with transformers,” inProc. 11th Int. Conf. Learning Representations (ICLR), 2023

    2023

  24. [24]

    iTrans- former: Inverted transformers are effective for time series forecasting,

    Y . Liu, T. Hu, H. Zhang, H. Wu, S. Wang, L. Ma, and M. Long, “iTrans- former: Inverted transformers are effective for time series forecasting,” inProc. 12th Int. Conf. Learning Representations (ICLR), 2024

    2024

  25. [25]

    Temporal graph networks for deep learning on dynamic graphs,

    E. Rossi, B. Chamberlain, F. Frasca, D. Eynard, F. Monti, and M. Bronstein, “Temporal graph networks for deep learning on dynamic graphs,” inProc. ICML Workshop Graph Represent. Learn. Beyond (GRL+), 2020

    2020

  26. [26]

    Do transformers really perform bad for graph representation?

    C. Ying, T. Cai, S. Luo, S. Zheng, G. Ke, D. He, Y . Shen, and T.-Y . Liu, “Do transformers really perform bad for graph representation?” inProc. 35th Advances in Neural Information Processing Systems (NeurIPS), vol. 34, 2021, pp. 28877-28888

    2021

  27. [27]

    MASTER: Market-guided stock transformer for stock price forecasting,

    S. Li, X. Liu, R. Shen, H. Wang, H. Chen, and J. Huang, “MASTER: Market-guided stock transformer for stock price forecasting,” inProc. 38th AAAI Conf. Artificial Intelligence, vol. 38, 2024, pp. 162-170

    2024

  28. [28]

    A distance measure for classifying ARIMA models,

    D. Piccolo, “A distance measure for classifying ARIMA models,”J. Time Series Anal., vol. 11, no. 2, pp. 153-164, 1990

    1990

  29. [29]

    Stock market trend prediction using ARIMA- based neural networks,

    J.-H. Wang and J.-Y . Leu, “Stock market trend prediction using ARIMA- based neural networks,” inProc. Int. Conf. Neural Networks (ICNN’96), vol. 4, 1996, pp. 2160-2165

    1996

  30. [30]

    Combining neural network model with seasonal time series ARIMA model,

    F.-M. Tseng, H.-C. Yu, and G.-H. Tzeng, “Combining neural network model with seasonal time series ARIMA model,”Technol. Forecast. Soc. Change, vol. 69, no. 1, pp. 71-87, 2002

    2002

  31. [31]

    Stock market’s price movement prediction with LSTM neural networks,

    D. M. Q. Nelson, A. C. M. Pereira, and R. A. De Oliveira, “Stock market’s price movement prediction with LSTM neural networks,” in Proc. Int. Joint Conf. Neural Networks (IJCNN), 2017, pp. 1419-1426

    2017

  32. [32]

    A dual-stage attention-based recurrent neural network for time series prediction,

    Y . Qin, D. Song, H. Chen, W. Cheng, G. Jiang, and G. Cottrell, “A dual-stage attention-based recurrent neural network for time series prediction,” inProc. 26th Int. Joint Conf. Artificial Intelligence (IJCAI), 2017, pp. 2627-2633

    2017

  33. [33]

    Forecasting stock prices from the limit order book using convolutional neural networks,

    A. Tsantekidis, N. Passalis, A. Tefas, J. Kanniainen, M. Gabbouj, and A. Iosifidis, “Forecasting stock prices from the limit order book using convolutional neural networks,” inProc. IEEE 19th Conf. Business Informatics (CBI), vol. 1, 2017, pp. 7-12

    2017

  34. [34]

    Stock price prediction using attention-based multi-input LSTM,

    H. Li, Y . Shen, and Y . Zhu, “Stock price prediction using attention-based multi-input LSTM,” inProc. Asian Conf. Machine Learning, 2018, pp. 454-469

    2018

  35. [35]

    Hierarchical multi-scale Gaussian transformer for stock movement prediction,

    Q. Ding, S. Wu, H. Sun, J. Guo, and J. Guo, “Hierarchical multi-scale Gaussian transformer for stock movement prediction,” inProc. 29th Int. Joint Conf. Artificial Intelligence (IJCAI), 2020, pp. 4640-4646

    2020

  36. [36]

    Hierarchical adaptive temporal- relational modeling for stock trend prediction,

    H. Wang, S. Li, T. Wang, and J. Zheng, “Hierarchical adaptive temporal- relational modeling for stock trend prediction,” inProc. 30th Int. Joint Conf. Artificial Intelligence (IJCAI), 2021, pp. 3691-3698

    2021

  37. [37]

    Temporal relational ranking for stock prediction,

    F. Feng, X. He, X. Wang, C. Luo, Y . Liu, and T.-S. Chua, “Temporal relational ranking for stock prediction,”ACM Trans. Inf. Syst., vol. 37, no. 2, pp. 1-30, 2019

    2019

  38. [38]

    Efficient integration of multi-order dynamics and internal dynamics in stock movement prediction,

    T. T. Huynh, M. H. Nguyen, T. T. Nguyen, P. L. Nguyen, M. Weidlich, Q. V . H. Nguyen, and K. Aberer, “Efficient integration of multi-order dynamics and internal dynamics in stock movement prediction,” inProc. 16th ACM Int. Conf. Web Search and Data Mining, 2023, pp. 850-858

    2023

  39. [39]

    StockMixer: A simple yet strong MLP-based architecture for stock price forecasting,

    J. Fan and Y . Shen, “StockMixer: A simple yet strong MLP-based architecture for stock price forecasting,” inProc. 38th AAAI Conf. Artificial Intelligence, vol. 38, no. 8, 2024, pp. 8389-8397

    2024

  40. [40]

    Enhancing stock movement prediction with adversarial training,

    F. Feng, H. Chen, X. He, J. Ding, M. Sun, and T.-S. Chua, “Enhancing stock movement prediction with adversarial training,” inProc. 28th Int. Joint Conf. Artificial Intelligence (IJCAI), 2019, pp. 5843-5849

    2019

  41. [41]

    Learning phrase representations using RNN encoder-decoder for statistical machine translation,

    K. Cho, B. Van Merri ¨enboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y . Bengio, “Learning phrase representations using RNN encoder-decoder for statistical machine translation,” inProc. Conf. Empirical Methods in Natural Language Processing (EMNLP), 2014, pp. 1724-1734

    2014

  42. [42]

    Trading and arbitrage in cryptocurrency markets,

    I. Makarov and A. Schoar, “Trading and arbitrage in cryptocurrency markets,”Journal of Financial Economics, vol. 135, no. 2, pp. 293-319, 2020. APPENDIX A. Baseline methods To rigorously evaluate the efficacy of our proposed archi- tecture, we benchmark it against a diverse set of state-of-the- art methods, ranging from fundamental sequence models to adva...

    2020

  43. [43]

    We conduct experiments using the following parameter settings as shown in Table VI

    Parameter settings:To ensure the fairness and consis- tency of the experiment, we used a uniform 30-day lookback window length and consistently set the learning rate to 0.00025 and Dropout to 0 throughout the 100 training epochs. We conduct experiments using the following parameter settings as shown in Table VI. TABLE VI DEFAULT PARAMETER SETTINGS Paramet...