EUVIMEDEUVIMED entdecken
Crossref

Price, Sentiment, and Activity in Bitcoin and U.S. Equities: A Comparative Machine-Learning and Predictive-Inference Framework, 2014 to 2026

Arnav Aggarwal

Student Journal of Business and Economics · 2026

DOI: 10.67521/sjbe.2026.011 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Student Journal of Business and Economics, 2026. Verfasst von Arnav Aggarwal. Die Arbeit nennt 30 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Bitcoin and U.S. equities are forecast in largely separate literatures, which leaves open whether the two markets respond to different forces or whether researchers simply study them differently. This paper imposes one framework on both: four feature blocks (Price/Technical, Sentiment/News, Activity/On-chain-or-Volume, Cross-market) evaluated across six classifier families under a walk-forward design, giving 114 model-block-asset comparisons for Bitcoin (4,346 daily returns) and for Tesla and the S&P 500 (2,338 common-calendar observations each). No comparison exceeds its majority-classbase rate by a statistically significant margin after Benjamini-Hochberg correction, so daily directional predictability is not detectable for either asset class. The fitted models nevertheless draw on different information. SHAP attribution assigns 60.3% of Bitcoin's fitted signal to activity and on-chain features and 5.6% to sentiment, against price and technical dominance for Tesla (55.0%) and the S&P 500 (53.5%). Bitcoin's largest single feature is a trading-volume-based illiquidity proxy rather than a blockchain-native metric, so the model's predictions rely on trading activity broadly rather than narrowly on on-chain metrics, and the shares are inconsistent with the claim that cryptoasset prices are driven mainly by hype; because SHAP describes a fitted model rather than the price process, this is evidence about predictive reliance, not causation. A HAC-robust, false-discovery-rate-corrected Granger battery finds that Bitcoin and Tesla returns predict next-day changes in the Crypto Fear & Greed Index (F = 157.72 and F = 15.34, p < 0.001) with no reverse effect. Since volatility and momentum carry half the weight of that index, the Bitcoin result is partly mechanical; the Tesla result cannot be, and is therefore the informative one, although a Granger test establishes predictive precedence rather than the underlying mechanism. GARCH(1,1) estimates place Bitcoin's volatility persistence (α+β = 1.000) and tail thickness (Student-tν = 3.20) above both equities.

Zitieren

APA

Aggarwal, A. (2026). Price, Sentiment, and Activity in Bitcoin and U.S. Equities: A Comparative Machine-Learning and Predictive-Inference Framework, 2014 to 2026. Student Journal of Business and Economics. https://doi.org/10.67521/sjbe.2026.011

Vancouver

Aggarwal A. Price, Sentiment, and Activity in Bitcoin and U.S. Equities: A Comparative Machine-Learning and Predictive-Inference Framework, 2014 to 2026. Student Journal of Business and Economics. 2026. doi:10.67521/sjbe.2026.011

BibTeX
@article{Aggarwal2026,
  author = {Aggarwal, Arnav},
  title = {Price, Sentiment, and Activity in Bitcoin and U.S. Equities: A Comparative Machine-Learning and Predictive-Inference Framework, 2014 to 2026},
  journal = {Student Journal of Business and Economics},
  year = {2026},
  doi = {10.67521/sjbe.2026.011}
}
Kostenlos starten →