BTC Regime Dashboard
Feed a 3-state HMM one strictly-causal number -- how far price sits below its running all-time high -- and the latent states line up with what a trader would call a bear, a chop, and a bull.
today · posterior P(regime)
tomorrow · one-step-ahead (π · A)
The regime tape
Log price over the model's posterior probability of each regime. Where one color floods the lower panel, the model is confident.
The observation
The only number the HMM sees: how far log-price sits below its running all-time high. Shallow excursions near zero are where the bull state lives; sustained moves below −0.5 are where the bear takes over.
Four episodes, same vocabulary
The model never sees these labels -- it reconstructs the cycle's chapters from the drawdown alone.
Episode composition
Share of each window the Viterbi path spends in each regime, and how often it switches.
The fitted states
The HMM learns two things over the full 2017→today window: per-state emission parameters (μ = mean drawdown, σ = spread -- bear sits deep below the high, bull hugs zero) and a transition matrix of one-step probabilities. The diagonal dominates -- each self-loop ≈ 0.99 -- which is what makes regimes persist for months rather than flicker daily.