Volatility Regime Classification

What is Volatility Regime Classification?

Volatility regime classification assigns each period a discrete volatility state, most commonly low, normal, and high, sometimes with an extra crisis tier, instead of leaving volatility as a continuous number. The simplest classifiers threshold a measure such as ATR or its volatility percentile; more sophisticated ones let the data draw the boundaries, using unsupervised grouping methods such as k-means on volatility features, or regime-switching statistical models that infer hidden states.

The premise is the persistence of volatility, documented by Benoit Mandelbrot in his 1963 cotton-price study: large changes tend to be followed by large changes, of either sign, and small changes by small changes. Because volatility is sticky, a regime identified today usually persists long enough to act on. The label is descriptive, not predictive: it says what conditions are, not when they will change, and the boundaries between states are modeling choices, so readings near a boundary can flip between adjacent regimes.

The idea carries serious econometric pedigree. Robert Engle's ARCH model (1982) and Tim Bollerslev's GARCH extension (1986) treated volatility as a process with memory, work that earned Engle a share of the 2003 Nobel Memorial Prize; James Hamilton's 1989 Markov-switching framework added the notion of hidden states with estimated transition probabilities. Chart-facing classifiers compress that machinery into something tradable: a colored background, a state label, or a stepped line that other rules can key off.

Implementations differ in input and boundary. The measure can be realized volatility on returns, ATR in price units, or a width gauge such as BandWidth from Bollinger Bands; boundaries can be multiples of a long average, rolling percentiles, or fitted by the model. Careful classifiers add hysteresis so borderline readings do not flip the state every bar. The payoff is discipline: a few named states can be backtested separately and wired into rules, which a raw continuous line resists.

How to identify volatility regimes on a chart

A workable classifier can be approximated with standard indicators before any code is written.

1. 1Plot a volatility series, such as 14-period ATR or 20-day realized volatility, in a pane below price.
2. 2Add a long baseline, for example a 100-period average of that series, and read the gap or ratio between the two.
3. 3Draw provisional boundaries: comfortably below the baseline is low, near it normal, well above it high; percentile bands over a one-year window make the same cut with more rigor.
4. 4Mark the crossings and check dwell time; states that flip every few bars mean the boundaries sit too close together.
5. 5Confirm against price character: low states should coincide with tight candles and narrow bands, high states with wide ranges and gaps.

How traders use it

• As a strategy switch: mean-reversion tactics are typically favored in low-volatility states, breakout and trend tactics as volatility expands, and reduced or no exposure in crisis states.
• As a risk control: volatility-targeted sizing shrinks position size as the regime rises, and stop distances widen so that normal noise in a hot regime does not force exits.
• As a parameter map: adaptive indicators re-tune lookbacks or multipliers per regime, running tighter settings in quiet states and looser settings in volatile ones.
• As a backtest lens: splitting performance by regime shows whether a strategy's edge concentrates in one state, which guides when the system should simply be switched off.
• As squeeze context: binary compression flags such as the TTM Squeeze or a Bollinger Squeeze classify only the quiet end; a fuller regime model grades how unusual the compression is against longer history.

Volatility regime classification vs. related tools