Regime-aware volatility forecasting for bitcoin: A hybrid EGARCH–hidden Markov model approach

Abstract

This study develops a regime-aware hybrid volatility framework for daily Bitcoin returns from January 2020 to May 2026 by integrating EGARCH with a three-state Gaussian Hidden Markov Model (HMM). EGARCH was selected as the best single-regime model, while the HMM identified quiet (1.0% volatility; 43.2% of observations), normal (3.4%; 43.4%), and crisis (6.0%; 13.4%) regimes, with crisis periods lasting about 9.6 days. Bitcoin returns display pronounced non-normality, negative skewness, autocorrelation, and ARCH effects, supporting the use of fat-tailed and regime-switching methods. In a strict out-of-sample evaluation over 465 test days, the Hybrid EGARCH+HMM outperformed standalone EGARCH, reducing RMSE by 31.2%, MAE by 47.2%, and QLIKE by 19.6%, and consistently achieving better monthly forecasts. Although an MS-EGARCH+HMM model improved accuracy, it remained inferior to the simpler hybrid approach. Results were robust across alternative volatility measures, subsamples, and Monte Carlo simulations, supporting practical Bitcoin risk management and forecasting applications.