Hidden regimes (Markov)

Also seen as: hidden Markov model, HMM, Markov regime switching, regime-switching model

What is a hidden-regime model?

A hidden-regime model assumes the market moves between a small number of states that you never observe directly — calm and turbulent — each with its own typical behaviour, and that it tends to stay in one state for a while before moving to another. The states are "hidden" because only the returns are visible; the model infers from them how likely each state is on a given day. Moving between states follows a Markov chain: tomorrow's state depends only on today's, which is where the name comes from.

The rule asks for Risk-Off on day $t$ when

where $S_t$ is the hidden state on day $t$, $r_1, \dots, r_t$ are the series' returns up to and including that day, and $q$ is the Probability threshold. In words: once the model judges the turbulent state at least $q$ likely, given everything seen so far, the strategy runs Risk-Off.

What do the number of states and the probability threshold change?

Number of states sets how many regimes the model looks for: 2 (the app labels it "2 · calm / turbulent") or 3. With 3 states the model also finds a middle regime, and only the most turbulent state counts towards Risk-Off — the app's note reads "With 3 states, only the most turbulent one counts."

Probability threshold sets how sure the model must be. The app puts the trade-off in one line: "Risk-Off when the turbulent state is at least this likely. Higher: fewer false alarms, but it reacts later." A lower threshold switches earlier and more often; a higher one waits for stronger evidence and switches less.

How does Fincanva handle it?

• The series is any instrument you pick under Instrument; a new rule starts on the S&P 500. The app's hint: "The model reads the returns of this series." There is no indicator, operator or pair of thresholds to set, unlike a Single series or Double series condition — see condition types.
• Number of states is 2 or 3, default 2. Probability threshold runs from 50% to 95%, default 70%.
• The model uses only history available on each day. It is recalibrated as the backtest moves forward, and a day's probability never draws on data from after that day.
• It stays Risk-On until it has enough history to learn from, and it stays Risk-On when the history shows no clearly distinct calm and turbulent states.
• One threshold means no built-in hysteresis. A Single series condition has separate thresholds to leave and re-enter Risk-On; this rule has one, so the confirmation delay is what damps whipsaw. The app's hint on that field: "This rule has a single threshold: the delay is your brake against switching too often. 0 = immediate." Auto-rebalance works as it does on any condition.
• Hidden regimes (Markov) is included from the Advanced plan, at the strategy level and inside a Combined alike; Free and Starter do not offer it. See what each plan includes.
• In the list of risk conditions a saved rule shows its short name, "Hidden regimes", with its series, its threshold as a comparison such as "≥ 70%", and its state count, such as "2 states".

What does it look like in practice?

You add a Hidden regimes (Markov) rule on the S&P 500 with 2 states, a probability threshold of 70% and a confirmation delay of 1 week. Through a quiet stretch the model puts the turbulent state at 10%–30%, so the strategy stays Risk-On. Returns then turn large and erratic, and the probability climbs to 55%, then 74%. At 74% the rule asks for Risk-Off; one week later, the reading still above 70%, the strategy switches to its Risk-Off allocation.

A threshold of 90% on the same history would have waited longer, and might never have switched if the probability peaked at 85%. A threshold of 55% would have switched a step earlier — and would also have fired on more short bursts that faded. The numbers here are illustrative, not a suggested setting.