HMM Math (Conceptual)

The HMM detector models regimes as hidden states that generate observed feature vectors.

Model Diagram

flowchart LR
  A[Hidden States] --> B[Observations]
  B --> C[State Probabilities]
  C --> D[Regime State]

Intuition

• Each hidden state has its own statistical profile.
• Observed features are explained by a state-specific distribution.
• Transition probabilities define how regimes shift over time.

Implementation Notes

RegimeFlow uses rolling windows of features and optional normalization. The model supports optional Kalman smoothing for regime probabilities.

See guide/regime-detection.md for configuration.