Volatility compression and expansion

Compression means movement has become unusually quiet

Volatility compression is a state in which recent price movement is small relative to a chosen reference. Expansion is a later increase in movement.

The word relative matters. A daily range of 1% may be quiet for one market and extreme for another. It may also be ordinary at midday and unusual near the open. A compression signal must therefore state both what it measures and what history defines normal.

Compression does not predict direction. It can precede an upward break, a downward break, or more quiet trading. A directional strategy usually uses compression as an eligibility condition and a separate price event to choose direction.

This differs from a breakout signal. Compression describes the environment before a potential move; the breakout defines the event that leaves a range. Keeping them separate lets a researcher test whether the quiet-state filter adds value beyond the breakout alone.

Why volatility changes in clusters

Large changes tend to be followed by other large changes, and quiet observations often cluster as well. Information arrival, changing participation, leverage, liquidity, and feedback from risk controls can all contribute to this behavior. Conditional-volatility models formalize this changing variance; they do not imply that a quiet state predicts the direction of the next move.

A quiet period can reflect balanced trading, uncertainty before an event, or positions accumulating without moving the visible price much. A later shock or imbalance can produce expansion. But the same quiet period can persist far longer than expected. “Coiling energy” is a metaphor, not a mechanism that forces release.

The research question is conditional: does a predeclared quiet state change the distribution of a later, separately defined trade?

Different measures describe different kinds of quiet

A global intraday threshold can label the lunch period “compressed” every day simply because activity is normally lower then. A same-time-of-day reference asks the stronger question: is this interval unusually quiet for this interval?

From state to strategy

An illustrative daily rule might be:

1. Calculate 10-day realized volatility after each completed daily bar.
2. Compare it with the distribution of 10-day volatility over the prior 252 completed sessions.
3. Mark the market compressed when the current value is below the 20th percentile.
4. While compressed, freeze the previous 10-day high and low as directional boundaries.
5. Enter at the next eligible event after a completed close beyond either boundary.
6. Exit on a return inside the range, an opposite break, or a declared timeout.

If compression occurs but no boundary breaks, there is no trade. If the market gaps beyond the boundary, the fill must reflect the first modeled opportunity, not the skipped line.

The percentile, lookbacks, and release event are all separate parameters. Changing them together makes it impossible to learn which component mattered.

Design a comparison that isolates the gate

Run the directional breakout without compression first. Then add exactly the same breakout, size, execution, and cost model with the compression gate. Report:

• total and eligible opportunity counts;
• how long compression states persist;
• delay and price distance between compression and release;
• frequency of immediate false releases;
• gross and net return distributions;
• behavior by direction and time of day; and
• sensitivity to neighboring percentiles and lookbacks.

The gate can improve average trade results simply by removing many trades. That is not enough: ask whether the remaining sample is large enough, whether capital sits idle, and whether the improvement survives later periods.

For volatility-scaled positions, freeze or update the risk estimate according to a declared rule. Expansion after entry can shrink the next position, but it cannot retroactively change the exposure already held.

Interpreting the result

A coherent compression measure should identify quieter paths, not merely profitable ones. Check its state against realized ranges before examining strategy P&L.

If the gate is useful, results should not depend on one isolated percentile. The effect should also survive a signal lag and reasonable transaction costs. Expansion magnitude alone is not profit: direction, entry price, exit, and slippage determine whether the move was captured.

Important failure modes

• Quiet conditions persist and tie up attention without producing a trade.
• A scheduled announcement causes a gap that skips the modeled entry.
• Intraday seasonality is mistaken for unusual compression.
• A backward-looking volatility estimate remains low just as the state changes.
• The chosen percentile is fitted to a few famous releases.
• Adding a directional filter, volume filter, and confirmation simultaneously hides which component contributed.

Try it in Arizmic

Further reading

• Engle, “Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation” (1982) — Introduces conditional variance that changes with past shocks. It supplies the foundation for treating volatility as a state, not evidence that a quiet state predicts breakout direction.
• Andersen, Bollerslev, Diebold, and Labys, “Modeling and Forecasting Realized Volatility” (2003) — Connects high-frequency observations to realized-volatility measurement and documents persistent volatility dynamics in exchange rates, clarifying what a compression measure is trying to estimate.
• Corsi, “A Simple Approximate Long-Memory Model of Realized Volatility” (2009) — Models daily, weekly, and monthly volatility components together and shows why the horizon used to define “quiet” matters. Its volatility forecasts do not supply a directional entry signal.