Volatility Bands and Squeezes

Two channels, one idea

A Bollinger Band is a moving average plus and minus a multiple — usually two — of the standard deviation of the closes in that window. A Keltner Channel is a moving average plus and minus a multiple — usually one and a half — of the average true range. They look similar on a chart and they are built from different quantities: standard deviation measures how spread out the closing prices are, while ATR measures how large the typical bar has been, including its gaps and wicks. That distinction is the reason traders look at both: on a stock that is drifting smoothly, the two channels are close together; on a stock that is gapping every few sessions, the ATR-based channel is noticeably wider, because ATR counts the overnight jump while a close-to-close standard deviation partly does not. Worked on the same $64.00 stock: a standard deviation of $1.20 puts the upper band at $66.40 and the lower at $61.60, a channel $4.80 wide, or a bandwidth of 7.50%. A 10-period ATR of $2.60 with a 1.5 multiplier puts the Keltner half-width at $3.90, a channel of $7.80, or 12.19% of price. The Bollinger channel is therefore inside the Keltner channel — the standard definition of a squeeze — and the ratio of the two, 7.50 ÷ 12.19 = 61.5%, quantifies how far inside. Comparing the bandwidth with the instrument’s own six-month average of 14.80% gives the second and more useful reading: today is 50.7% of normal, which says the market is unusually quiet for this stock rather than that it is quiet in some absolute sense. What the arithmetic does not contain is a direction. Neither channel knows anything about which side will break, and the two edges are not levels in the structural sense — they are statistical descriptions, roughly two standard deviations from the mean, so price touching the upper band is an observation that the session is far from the average rather than a rejection by buyers. That is the single most common misuse: treating a band touch as support or resistance. It is a measure, and a reliable one — around ninety-five per cent of observations fall inside two standard deviations if the distribution behaves — but a measure of where price is relative to its own average, not of where it will go next. The same stock, two channels — Bollinger: $64.00 ± 2 × $1.20: Upper $66.40 · lower $61.60 · bandwidth 7.50% ← · Keltner: $64.00 ± 1.5 × $2.60: Half-width $3.90 · channel 12.19% of price · Bollinger inside Keltner: Squeeze is on — 61.5% of the Keltner width · Against its own six-month average: 7.50% ÷ 14.80% = 50.7% of normal — unusually quiet for this instrument ← Bandwidth has no absolute meaning, exactly like implied volatility. A 7.5% channel is tight for this stock and loose for another, so the only comparable figures are the instrument’s own past readings.

Why the squeeze is a sizing problem first

The claim behind a squeeze is empirical and modest: volatility clusters, so periods of low realised movement are more likely to be followed by high movement than by more of the same. It is one of the better-attested regularities in market data — large moves arrive in batches, and calm does not persist indefinitely. What the claim does not include is a sign. A squeeze resolves in either direction about as often as the market’s own drift suggests, and the break can also be a fake that stops out both sides before the real move. So a trader who responds to a squeeze by picking a direction has replaced a statistical regularity with a guess; a trader who responds by preparing has used it. Preparation has three parts and they are all sizing. First, the stop budget: with the ATR at $2.60 and a channel of ±$3.90, a stop at one ATR is inside the noise of the expansion that is being anticipated, so the position has to be sized from a stop of about two ATR — $5.20 — which for a fixed dollar risk means roughly half the shares a one-ATR stop would have bought. Second, the trigger: the break of the compression range, defined in advance, with the direction taken from the break rather than predicted, and both edges marked so neither is missed. Third, the expectation: an expansion after a squeeze commonly overshoots the channel within a few sessions, which is a reason to have a plan for the move rather than a target for it, and the honest way to hold a position into that is with the size the stop allows and a trail behind the structure. There is a fourth consideration that belongs to the account rather than the trade. Because squeezes occur across many instruments at the same time — the same market-wide calm compresses every correlated channel together — a trader who takes a squeeze in six correlated names has not made six independent bets but one large bet on the expansion, with six times the cost if it goes the wrong way. The position limit that governs correlated exposure applies here in an unusually literal way: the expansion arrives everywhere at once, and the sizes add up on the bad day exactly as they do on the good one. • Volatility clusters: calm periods precede directional ones — a claim about magnitude, not sign. • Size from a stop that survives the expansion (about two ATR), not from the calm one-ATR range. • Define both break levels in advance and take the direction from the break rather than from a forecast. • Expect overshoot beyond the channel, and manage with a structure trail rather than a fixed target. • Correlated squeezes resolve together: six names is one bet on the expansion, not six bets. The most expensive version of this mistake is selling options, or writing covered calls, into a squeeze because the premiums look attractive and the market looks calm. Compressed volatility is *cheap* volatility, and selling it immediately before a known expansion is the trade that defines a generation of blow-ups — the income is modest and capped, and the loss arrives all at once when the break comes. A squeeze is a reason to buy optionality or to trade the break with a wide stop, not a reason to sell the calm.

Reading the band, not just the price

A channel has two edges, and a second set of measurements is built from them that describes the state of the instrument rather than the location of price. The first is where price sits inside the channel, expressed as a percentage of the way from the lower edge to the upper one. At zero price is on the floor, at one hundred it is on the ceiling, and around fifty it is at the average. That single figure takes the average out of the question: a reading of ninety-five says the last close was near the top of the range whatever the average did, and it makes two instruments comparable in a way their raw prices never are. The second is the width itself over time, which is the quantity the squeeze is defined by and which has its own recognisable history. A contraction is a run of declining peaks in width; an expansion is the reverse. The transitions matter more than either state, because they are the only moments the channel says something new: after a sequence of lower readings in width, the first higher one is the market announcing that the quiet is ending, and it arrives before any trend filter accepts the new direction. The third is the count of touches. Two edges told you where a multiple of the noise sits; how often price has tested each edge tells you whether the market is accepting that range or leaning against one side. Several taps of the upper edge with shallow pullbacks is a different market from a single spike and an immediate return, and the channel alone cannot separate them — the count and the depth of the returns carry that information. Two readings should be treated as unreliable. A single bar closing outside the channel is an observation about that bar rather than about the trend; the useful version is a cluster of closes outside it, or a close that holds outside on the following bar. And a channel that widens while price goes nowhere is an expansion of noise rather than of direction, which changes the stop distance and says nothing about the next move. Both are the same mistake in different clothes: reading a statistical description as though it were a signal. • Where price sits in the channel: zero at the floor, one hundred at the ceiling, fifty at the average. • Width over time: falling peaks of width is contraction, rising peaks is expansion. • The transition is the information — a contraction that stops contracting is the announcement, not the continuation. • Count the touches and the depth of the returns to tell acceptance apart from a single spike. • One close outside the band is a fact about a bar; a cluster that holds is a change of state. A band is a description, so the readings taken from it are descriptions too. They become decisions only after a trigger — a break, a close, a failed retest — has been attached to them in advance.

Sources

• Bollinger Bands: construction, bandwidth and the squeezeBollinger, “Bollinger on Bollinger Bands”
• Keltner Channels and the volatility squeezeKeltner / Chester Keltner channel literature and the TTM Squeeze definition
• Volatility clustering and why calm precedes movementEngle, ARCH and the volatility-clustering literature