Implementation of Machine Learning-Based Portfolio Optimization Methods: HRP, HCAA, and HERC in Cryptocurrency Portfolios

Hierarchical clustering is an unsupervised learning method that organizes assets into a tree-like structure based on a similarity or distance measure, which can be applied in portfolio optimization. Hierarchical clustering-based portfolio construction is designed to overcome the limitations of simple optimization methods in dealing with correlated assets, as commonly observed in the cryptocurrency market. Therefore, this study applies several hierarchical portfolio construction techniques, such as hierarchical risk parity (HRP), hierarchical clustering-based asset allocation (HCAA), and hierarchical equal risk contribution (HERC). Several metrics are applied to evaluate the constructed portfolios, including the adjusted Sharpe ratio, certainty-equivalent return, maximum drawdown, and average turnover. In this research, these methods are tested on twenty cryptocurrencies with the largest market capitalizations, which tend to have lower risk and higher liquidity. Among those assets there is stable coin (USDT) that minted with US dollar as the underlying, which can influence portfolio constructions. The study shows that the hierarchical portfolio weighting approach exhibits heterogeneous performance in terms of return and risk. For portfolio returns, the HERC method with a single link and the CDaR risk metric achieved the highest return of 46.08%, followed by the HCAA method with a Ward link at 27.21%, while the HRP method recorded a significantly lower return of 0.03%. From a risk perspective, HERC demonstrated superior ability to combine risk factors and control extreme losses, as reflected in its certainty-equivalent returns and favorable diversification measures. Meanwhile, HCAA achieved the highest adjusted Sharpe ratio of 0.1603 and demonstrated strong diversification properties. These findings suggest that the hierarchical portfolio construction method provides an effective alternative for cryptocurrency markets characterized by high correlations and extreme risk dynamics.