Machine Learning and the Cross-Section of Cryptocurrency Returns

We employ a repertoire of machine learning models to explore the cross-sectional return predictability in cryptocurrency markets. While all methods generate substantial economic gains, those that account for nonlinearities and interactions fare the best. The return predictability derives mainly from a handful of simple features—such as idiosyncratic volatility, past alpha, or maximum daily return—and is likely driven by mispricing. Accordingly, abnormal returns originate predominantly from short positions, concentrate in hard-to-arbitrage assets, and gradually decline over time. Despite a high portfolio turnover, machine learning strategies remain a profitable net of trading costs. However, they critically depend on shorting small cryptocurrencies, which may pose challenges in practice