A new hybrid machine learning model for predicting the bitcoin (BTC-USD) price
Pavan Kumar Nagula and
Christos Alexakis
Additional contact information
Pavan Kumar Nagula: ESC Rennes School of Business - ESC [Rennes] - ESC Rennes School of Business
Christos Alexakis: ESC Rennes School of Business - ESC [Rennes] - ESC Rennes School of Business
Post-Print from HAL
Abstract:
Several machine learning techniques and hybrid architectures for predicting bitcoin price movement have been presented in the past. Our paper proposes a hybrid model encompassing classification and regression models for predicting bitcoin prices. Our analysis found that the automated feature interactions learner (deep cross networks) error performance using a plethora of technical indicators, including crypto-specific technical indicator difficulty ribbon compression and control variables such as Metcalfe's value of bitcoin, number of unique active addresses, bitcoin network hash rate, and S&P 500 log returns, in a hybrid architecture is better than the single-stage architecture. The hybrid model predicted a 100% directional hit rate and maintained steady volatility in returns for the out-of-sample period. Our paper concludes that in terms of risk (Sharpe ratio 1.03) and profitability (260% and 82%), the hybrid model's bitcoin futures strategy performed better than the deep cross network regression and buy-and-hold benchmark strategies.
Date: 2022-12
References: Add references at CitEc
Citations: View citations in EconPapers (2)
Published in Journal of Behavioral and Experimental Finance, 2022, 36, pp.100741. ⟨10.1016/j.jbef.2022.100741⟩
There are no downloads for this item, see the EconPapers FAQ for hints about obtaining it.
Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.
Export reference: BibTeX
RIS (EndNote, ProCite, RefMan)
HTML/Text
Persistent link: https://EconPapers.repec.org/RePEc:hal:journl:hal-03877093
DOI: 10.1016/j.jbef.2022.100741
Access Statistics for this paper
More papers in Post-Print from HAL
Bibliographic data for series maintained by CCSD (hal@ccsd.cnrs.fr).