State‐Dependent Market (In)Efficiency in Cryptocurrency Markets

This paper develops a meta‐learning framework to test whether cryptocurrency market inefficiency is static or state‐dependent. To resolve competing claims between the Efficient and Adaptive Market Hypotheses (AMH), we construct an adaptive trading strategy using high‐frequency data for 50 major cryptocurrencies. The framework is unique in its integration of four distinct information sets: on‐chain metrics, behavioral trend asymmetries derived from Directional Change, strategy meta‐information, and market statistics. Our out‐of‐sample results support the AMH: the Adaptive Meta‐Learning Strategy achieves a Sharpe ratio of 1.34, more than double the best static benchmark (0.59), while reducing maximum drawdown by 37.5%. Crucially, feature attribution analysis reveals that outperformance is regime‐dependent; the model learns economically intuitive rules, prioritizing trend asymmetries during high‐volatility periods and fundamental metrics during stable markets. These findings demonstrate that alpha in cryptocurrency markets is a time‐varying reward that depends on using the appropriate information sources in each market regime.