HMM Regime Filter | Market State

The HMM Regime Filter indicator classifies current market conditions into one of three hidden states—Trend Up, Range, or Trend Down—using a Bayesian forward filtering model to provide real-time context for strategy selection.

Usage

The Usage section describes how the script can be used to distinguish between trending and mean-reverting environments. It serves as a context layer rather than a trade trigger; users can use the identified regime to determine which technical tools to prioritize.

• Trend Up / Trend Down: In these regimes, trend-following logic such as pullbacks, breakouts, and momentum continuation is typically more effective.
• Range: When the model identifies a ranging state, mean-reversion strategies, such as fading price extremes or using oscillators, are generally preferred.
• Uncertain: If no regime meets the dominance threshold, the model reports an uncertain state, suggesting a period of indecision where traders might reduce position sizes or stand aside.

The indicator provides visual feedback through a probability ribbon in a separate pane, bar coloring on the main chart, and a status dashboard. The dashboard tracks the duration of the current regime ("Held for"), allowing users to gauge the stability of the current market state.

Details

The HMM Regime Filter utilizes a Hidden Markov Model (HMM) to infer market states that are not directly observable. It updates the probability of each regime on every bar close through two primary components:

1. Transition Model: This defines the probability of moving between states. The model is structured so that trends typically decay into a Range state before reversing, reflecting the natural cycle of most financial markets.
2. Gaussian Emissions: The model compares current price behavior (normalized returns) against expected behavior for each state. These emissions scale automatically to the instrument's volatility using a lookback window.

To ensure stability and prevent "flickering" (rapid switching between states), the script implements three mechanisms:

• Stickiness: A setting that dictates the model's preference to remain in its current state unless significant evidence suggests a shift.
• Confirmation Rule: Requires a new regime to maintain its lead for a specific number of consecutive bars before the official state updates.
• Dominance Threshold: A confidence requirement that prevents the model from committing to a regime if the leading probability is too low.

Settings

Model

• Observation smoothing: EMA length applied to normalized returns before processing. Higher values reduce noise on fast timeframes.
• Scaling window: The lookback period used to calibrate the model to the specific instrument's volatility.
• Regime separation: Determines the distance between the Trend Up and Trend Down means in standard deviations. Higher values result in more "Range" classifications.

Stability

• Stickiness (stay probability): Controls the diagonal of the transition matrix. Higher values create more persistent, stable regimes.
• Confirmation bars: The number of consecutive bars a new regime must be the most likely before it is officially confirmed.
• Dominance threshold: The minimum probability required for a regime to be considered active; otherwise, the state is marked as "uncertain."

Display

• Probability ribbon: Toggles the visibility of the flowing probability streams in the indicator pane.
• Ribbon smoothing: Applies visual smoothing to the ribbon streams without affecting the underlying logic.
• Dark pane background / Tint pane: Controls the aesthetic background and regime-based tinting of the indicator pane.
• Color price bars: Colors the main chart candles based on the confirmed market regime.
• Show dashboard: Toggles the data table containing live probabilities and regime duration.

FAQ

How do I interpret the "Held for" counter in the dashboard? The "Held for" counter indicates how many bars the current regime has been official. High numbers suggest a stable, established environment, while frequent resets indicate choppy or transitional market conditions.

Why does the indicator stay on "uncertain" for several bars? The model requires a warm-up period defined by the "Scaling window" to calibrate to the instrument. It may also show "uncertain" if the "Dominance threshold" is set too high for the current market conviction.

How do I access the HMM Regime Filter | Market State? You can get access on the LuxAlgo Library for charting platforms like TradingView, MetaTrader (MT4/MT5), and NinjaTrader for free.