Daily trading activity may offer insight into changing cryptocurrency risk, but not necessarily into future returns. This study examines that distinction for Bitcoin, Ethereum, Solana and BNB using daily Binance data from January 2021 to September 2026. Two-state Gaussian Markov-switching regressions relate returns to lagged returns, trading volume and taker-buy pressure, and trading variables affect the probabilities of regime switching. Forecasts are compared against historical-mean and Gaussian linear benchmarks over January 2025-September 2026. The estimates identify lower-volatility and higher-volatility states, with shorter expected durations in the higher-volatility regime for all four assets. For Ethereum and BNB, volatility transitions are driven by trading volume, while for Bitcoin and Solana, they are driven by taker-buy pressure. Bitcoin also shows a stronger positive relationship between lagged buy pressure and returns under higher-volatility conditions. These in-sample relationships do not carry over into consistent improvements in return magnitude or directional forecasts. Overall, regime models improve predictive density scores over the Gaussian linear benchmark, but buy pressure provides no statistically reliable incremental density-forecasting benefit across the four assets. The results support an asset-specific evaluation of trading indicators and favour assessing regime models by their ability to describe changing return distributions, rather than assuming that regime identification leads to accurate return forecasts.

Keywords: Cryptocurrency; Markov-switching; Trading volume; Taker-buy pressure; Volatility regimes