Exploring the predictability of cryptocurrencies via Bayesian hidden Markov models

•We explore the predictive power of hidden Markov models in cryptocurrency returns.•The 4-state non-homogeneous hidden Markov model has the best forecasting performance.•Based on profits and risks, the model distinguishes bull, bear, and calm regimes.•The model identifies predictors with state-dependent, linear and non-linear effects.•The most common predictors are series momentum, VIX and US Treasury Yield.