Prediction of Daily Lognormal Returns for Bitcoin Based on LightGBM

Jiaxing Wei

2024

Abstract

With rapid development in Blockchain technologies, the security of cryptocurrencies like Bitcoin has been significantly improved. However, as the cryptocurrency with the largest traded volume per day, Bitcoin continuous to expose to volatile risk due to its intrinsic attributions, including non-supervisory and all-weather. This research utilizes neural network and tree-based models to predict the short-term future returns of Bitcoin. The Neural-Network-based models like Long-Short Term Memory (LSTM) and Transformer outperform with statistical significance. By introducing L2-regularzation, the research discovers an available approach to alleviate the short-term volatile risk for investors by proposing an embedding model to predict rapid changes from future returns. While leverages a R-squared that outperform the benchmark by 11%, the embedding model is verified to maintain efficiency with an enhanced convergence rate. The research analyses 4 commonly used Machine Learning models in financial time-series prediction and compares their performances with the calibrated embedding model. By contrasting the advantages and corresponding shortcomings, this research fills the gap in offering suggestions for investors to engage non-supervised market to decrease exposures in volatile risk.

Paper Citation

in Harvard Style

Wei J. (2024). Prediction of Daily Lognormal Returns for Bitcoin Based on LightGBM. In Proceedings of the 1st International Conference on E-commerce and Artificial Intelligence - Volume 1: ECAI; ISBN 978-989-758-726-9, SciTePress, pages 154-163. DOI: 10.5220/0013208400004568

in Bibtex Style

@conference{ecai24,
author={Jiaxing Wei},
title={Prediction of Daily Lognormal Returns for Bitcoin Based on LightGBM},
booktitle={Proceedings of the 1st International Conference on E-commerce and Artificial Intelligence - Volume 1: ECAI},
year={2024},
pages={154-163},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013208400004568},
isbn={978-989-758-726-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on E-commerce and Artificial Intelligence - Volume 1: ECAI
TI - Prediction of Daily Lognormal Returns for Bitcoin Based on LightGBM
SN - 978-989-758-726-9
AU - Wei J.
PY - 2024
SP - 154
EP - 163
DO - 10.5220/0013208400004568
PB - SciTePress