EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Sentiment Trumps Fundamentals: Machine-Learning Evidence on the Drivers of Cryptocurrency Returns

Min Zhang, Liangbo Zheng, Jing Chen

Journal of Business and Data Analytics · 30.09.2026 · 4 (9) · 149-170

DOI: 10.63646/yyor9860 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Business and Data Analytics, online erschienen am 30.09.2026. Verfasst von Min Zhang und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Whether investor sentiment is more informative than network fundamentals for cryptocurrency returns remains an empirical question. This study evaluates that proposition using a transparent, time-ordered forecasting design for Bitcoin and Ether. Daily prices and on-chain activity are drawn from Coin Metrics community data, and the Bitcoin-oriented Crypto Fear & Greed Index is obtained from Alternative.me. The index is a composite market-mood proxy rather than a pure measure of emotion because its documented components include volatility and market momentum. Daily log returns are forecast using market-history features alone or augmented with sentiment, blockchain-activity proxies, or both. Ridge regression and histogram-based gradient boosting are specified with fixed hyperparameters. Models are developed on data from June 2018 to December 2021, assessed on 2022, and refitted through 2022 before testing on 2023–2024. After common complete-case filtering, each asset has 723 test observations. In the test period, the gradient-boosting market-only root mean squared errors are 2.546 percentage points for Bitcoin and 2.967 for Ether, compared with 2.558 and 2.967 when the sentiment block is added. The sentiment block does not consistently improve forecasting relative to the market-only or network-augmented models, and all gradient-boosting variants have negative out-of-sample R² relative to a zero-return forecast. Paired block-bootstrap intervals show substantial uncertainty in the incremental comparisons. The title’s proposition is therefore not supported as a general conclusion by this restricted test. The results illustrate the importance of release timing, overlapping information, simple benchmarks, and reproducible validation when interpreting claims about the drivers of cryptocurrency returns.

Zitieren

APA

Zhang, M., Zheng, L., & Chen, J. (2026). Sentiment Trumps Fundamentals: Machine-Learning Evidence on the Drivers of Cryptocurrency Returns. Journal of Business and Data Analytics, 4(9), 149-170. https://doi.org/10.63646/yyor9860

Vancouver

Zhang M, Zheng L, Chen J. Sentiment Trumps Fundamentals: Machine-Learning Evidence on the Drivers of Cryptocurrency Returns. Journal of Business and Data Analytics. 2026;4(9):149-170. doi:10.63646/yyor9860

BibTeX
@article{Zhang2026,
  author = {Zhang, Min and Zheng, Liangbo and Chen, Jing},
  title = {Sentiment Trumps Fundamentals: Machine-Learning Evidence on the Drivers of Cryptocurrency Returns},
  journal = {Journal of Business and Data Analytics},
  year = {2026},
  volume = {4},
  number = {9},
  pages = {149--170},
  doi = {10.63646/yyor9860}
}
Kostenlos starten →