How Realized-Volatility Regimes Are Classified

The published implementation converts one of four realized-volatility estimators into historical rank, fixed tiers, an estimator-specific size override, term structure, and acceleration.

Estimator · Rank · Absolute size · Term structure

The practical question
Is the reported tier coming from historical rank, the absolute-size override, or both?

Auction Foundry publication capture from TradingView. It is an example chart, not live market data.

The regime-classification contract

A regime label is the last step, not the first. Reproduce the estimator and its windows before interpreting the percentile, expansion ratio, or tier.

Estimator: Choose close-to-close, Parkinson high-low, Garman-Klass OHLC, or normalized true range. Each route measures a different feature of the same bars.
Window set: Fast uses 5/15/30 bars, Normal 10/30/60, and Slow 20/60/120. The short-to-long ratio remains one-to-six across presets.
Historical rank: The current short-window estimate is ranked against 400 short-window observations, including the current value; ties count at or below the observation.
Classification: Percentile thresholds create QUIET, NORMAL, ELEVATED, and EXTREME tiers, while an estimator-specific short-over-long ratio can independently qualify EXTREME.

Start with the estimator’s information set

Close-to-close uses the population standard deviation of log returns. Parkinson takes the square root of the mean high-low variance terms. Garman-Klass combines the high-low range with the open-to-close move. The ATR route averages each bar’s true range divided by that bar’s own close.

The ATR route is a simple N-bar mean, not Wilder’s recursive smoothing, and none of the four routes is annualized in the published implementation. Close-to-close and ATR can exclude the first intraday reading that spans a session boundary; Parkinson and Garman-Klass use only information inside the bar and are unaffected by that setting.

Build comparable short, mid, and long estimates

The sensitivity preset changes timescale while preserving a six-to-one long-versus-short span: Fast is 5/15/30 bars, Normal is 10/30/60, and Slow is 20/60/120. Range-based estimators use the root of mean variance terms; the ATR route uses a plain mean because its input is already a range level.

Short-over-long expansion is the short estimate divided by the long estimate. The script’s two-point term-structure state compares those endpoints with a 10% tolerance; it is a window-shape label, not a futures-curve or options-surface measurement. Acceleration separately compares the current short-window estimate with its value one short window earlier and applies a 5% deadband.

Rank only after the full historical sample exists

The percentile ranks the current short-window estimate against the trailing 400 short-window values, current bar included. It stays undefined until all 400 observations exist; ranking a partial sample would be a different statistic. A zero-volatility long window also suppresses classification so a series of zero values cannot rank as EXTREME.

The raw percentile tiers are QUIET below 25, NORMAL from 25 to below 75, ELEVATED from 75 to below 85, and EXTREME at 85 or higher. A one-point regime band adds hysteresis around transitions so a marginal cross does not immediately flip the reported tier back and forth.

Treat EXTREME by size as a separate route

A percentile is self-normalizing: once a volatile stretch fills the reference window, rank can decline even while absolute movement remains large. The published tool therefore permits EXTREME when short-window volatility reaches an estimator-specific multiple of long-window volatility, regardless of rank.

The published multiples are 2.00 for close-to-close, 1.35 for Parkinson, 1.30 for Garman-Klass, and 1.40 for ATR. This route can classify EXTREME before the 400-observation percentile is available. It is an implementation threshold calibrated by estimator, not a universal volatility boundary.

Use the tier as a conditioning variable

QUIET, NORMAL, ELEVATED, and EXTREME describe the magnitude of movement already observed. Expansion describes the short-versus-long relation; acceleration describes change in the short estimate. None contains sign, trend direction, expected move, entry, target, or position size.

A professional review record should preserve symbol, venue or feed, session, timeframe, estimator, sensitivity preset, gap treatment, sample count, percentile, raw and reported tier, size-override status, expansion ratio, and acceleration state. Evaluate those fields as conditioning variables against a separately defined setup and risk process.

BEFORE YOU USE THE IDEA

5 questions for your chart

01
Which estimator and sensitivity preset produced the short, mid, and long values?
02
Was the session-opening observation excluded, and does that setting affect this estimator?
03
Is the 400-observation rank available, or did the absolute-size rule classify the tier first?
04
What are the raw tier, reported tier, expansion ratio, and acceleration state?
05
Does the interpretation keep volatility magnitude separate from directional bias?

WHAT IT CAN CLARIFY

Makes the estimator, fixed window family, full sample count, and tier thresholds explicit.
Separates historical rank, absolute expansion, term structure, and acceleration.
Uses estimator-specific size thresholds instead of pretending unlike measurements share one scale.

LIMITATIONS TO RESPECT

The 400-observation lookback, thresholds, deadbands, and size multiples are implementation choices.
A percentile is relative to this symbol and sample; it is not a probability, implied volatility, or expected move.
Active-bar measurements can change before confirmation, and alert delivery depends on the user’s TradingView alert and data configuration.

CONTINUE YOUR RESEARCH

Continue with first-party material

These links document the underlying market mechanics, calculation conventions, or platform behavior. They are starting points for independent study, not endorsements.

01Michael Parkinson · Journal of Business archival copy at CME GroupThe Extreme Value Method for Estimating the Variance of the Rate of ReturnThe primary 1980 paper behind the high-low range estimator used as one available measurement route.02Mark B. Garman and Michael J. Klass · Journal of Business archival copy at CME GroupOn the Estimation of Security Price Volatilities from Historical DataThe primary 1980 paper behind the OHLC estimator available in the published classifier.03TradingView Help CenterAverage True RangeDocuments true range and TradingView’s built-in RMA smoothing, useful when comparing implementations.04Cboe Global IndicesVIX methodologyShows that VIX is an options-implied forward-looking measure, distinct from historical realized-volatility estimators.05TradingView Pine Script documentationExecution modelExplains bar-by-bar calculation, realtime updates, rollback, confirmation, and why active-bar values can change.

The regime-classification contract

Start with the estimator’s information set

Build comparable short, mid, and long estimates

Rank only after the full historical sample exists

Treat EXTREME by size as a separate route

Use the tier as a conditioning variable

5 questions for your chart

See the framework in a published script.

Volatility Regime Classifier

Volatility Regime Range Map

Range Budget

Continue with first-party material