Constructing a Cryptocurrency Investment Portfolio Using Machine Learning Algorithms

Keywords

cryptocurrencies
investment portfolio
machine learning
CVaR
XGBoost
risk management
walk-forward

How to Cite

Алиев, Т. А. (2026). Constructing a Cryptocurrency Investment Portfolio Using Machine Learning Algorithms. Risk Managmenent, 4. https://www.riskmanagementjournal.ru/index.php/rmj/article/view/565

Abstract

The article evaluates whether combining a machine-learning return forecast with Mean-CVaR optimization yields a statistically significant improvement in the risk-adjusted performance of a crypto-asset portfolio relative to the Hierarchical Risk Parity (HRP) benchmark. Using daily data on twenty USDT cryptocurrency pairs from the Binance exchange for 2018–2025, three strategies are compared under an expanding walk-forward scheme: Mean-CVaR with XGBoost forecasts, Mean-CVaR on historical means, and HRP. CVaR optimization is shown to significantly outperform HRP by the Sharpe ratio, whereas adding the ML forecast provides no statistically significant value. Conclusions on the practical applicability of the approach for retail and institutional investors are formulated.