CryptoGAT

Are Time Series Models Effective for Cryptocurrency Forecasting?

Official implementation of CryptoGAT, a lightweight Graph Attention Network for cryptocurrency forecasting.

Overview

We introduce a set of simple yet effective models named CryptoGAT as a new baseline for comparison. CryptoGAT uses an attention mechanism architecture to process graph-structured data, allowing each node to focus on its neighbors to compute hidden representations and assign different weights. To handle cryptocurrency markets with different characteristics, we further introduce a variant, named FGAT. Now we have a CryptoGAT family:

• GAT: A simple yet effective graph-based model via a graph attention mechanism.
• FGAT: FeatureMixer Graph Attention Network (FGAT) combines the Indicator mixing mechanism with the graph attention architecture.

Paper

Our paper is available on arXiv:

CryptoGAT: Are Time Series Models Effective for Cryptocurrency Forecasting?
Yu Peng, Matloob Khushi, Josiah Poon
arXiv:2606.27670, 2026
Paper | PDF

Citation

If you find this repository or paper useful, please cite our work.

BibTeX

@misc{peng2026cryptogat,
  title = {{CryptoGAT}: Are Time Series Models Effective for Cryptocurrency Forecasting?},
  author = {Peng, Yu and Khushi, Matloob and Poon, Josiah},
  year = {2026},
  eprint = {2606.27670},
  archivePrefix = {arXiv},
  primaryClass = {cs.CE},
  doi = {10.48550/arXiv.2606.27670},
  url = {https://arxiv.org/abs/2606.27670}
}

APA

Peng, Y., Khushi, M., & Poon, J. (2026). CryptoGAT: Are Time Series Models Effective for Cryptocurrency Forecasting? arXiv. https://doi.org/10.48550/arXiv.2606.27670

IEEE

Y. Peng, M. Khushi, and J. Poon, "CryptoGAT: Are Time Series Models Effective for Cryptocurrency Forecasting?," arXiv:2606.27670, 2026. doi: 10.48550/arXiv.2606.27670.

Plain Text

Yu Peng, Matloob Khushi, and Josiah Poon. CryptoGAT: Are Time Series Models Effective for Cryptocurrency Forecasting? arXiv:2606.27670, 2026. https://arxiv.org/abs/2606.27670

Getting Started

Environment Requirements

• tqdm==4.64.1
• pandas==2.0.3
• pyyaml==6.0
• numpy==1.22.1
• matplotlib==3.5.1
• torch>=1.9.0

Dataset and Preprocessing

In order to improve file reading speed, we process the raw data to generate corresponding .pkl or .npy files. Datasets are provided in the dataset folder.

CRYPTO_1D_ALL

# Process the data into CRYPTO_1D_ALL
cd src
python process_crypto_ALL.py

CRYPTO_1D_ENHANCED

# Process the data into CRYPTO_1D_ENHANCED for FGAT
cd src_Feature
python process_crypto_enhanced.py

Running the Code

Train GAT Model

cd src
python train_gat_crypto.py

Train FGAT Model

cd src
python train_gat_enhanced.py

Evaluation Metrics

• MSE: Mean Squared Error
• IC: Information Coefficient
• ICIR: Information Coefficient Information Ratio
• Prec@10: Precision at Top 10
• Sharpe: Sharpe Ratio

Dataset

The CryptoGAT Crypto 1D dataset used in this project is available on the Hugging Face Hub:

CharlieYPeng/cryptogat-crypto-1d