Return and risk in cryptocurrency portfolios: Evidence from competing allocation methods

We examine the performance of portfolio construction methods that incorporate cryptocurrencies, a market characterized by extreme volatility and rapid structural change. While investors increasingly seek to integrate digital assets into diversified portfolios, existing approaches often rely on traditional econometric techniques developed for more mature asset classes. At the same time, a growing literature proposes machine-learning-based allocation algorithms, though systematic comparisons between these methods and conventional econometric strategies remain limited. Using historical data on 134 cryptocurrencies, we conduct backtests to evaluate the relative performance of alternative portfolio allocation rules. Our findings shed light on the return and risk properties of cryptocurrency markets and provide evidence on the conditions under which machine-learning methods improve portfolio performance relative to traditional econometric approaches.