Regime Misclassification

Regime misclassification is the failure mode in which an HMM (or any regime model) assigns the wrong hidden state to a period — labelling a calm market as turbulent or vice versa. It is the dominant source of trading losses in regime strategies: Bulla et al. (2011) note that a wrong regime call is “detrimental” rather than merely sub-optimal, since it allocates in the contrary direction to the neutral position. Misclassification is worst at the start and end of estimation windows (error rates spike to ~10%) and during oscillating markets; HMM signals also lag turning points by a median of around 25 days. It appears in this vault as the core reason HMM regime trading delivers weak standalone alpha after costs.

Connections

• Hidden Markov Model Regime Detection — suffers_overfitting_risk, source: https://arxiv.org/html/2402.05272v2
• Shu Yu and Mulvey 2024 — contradicts, source: https://arxiv.org/html/2402.05272v2
• Regime Classification — relates, source: https://www.nber.org/system/files/working_papers/w17182/w17182.pdf