Volatility based stop placement

Two answers to one question 

Structure based placement, the subject of the previous lesson, answers a question about the chart: at what price would the reading that motivated this position no longer hold. The level is found by looking, and the distance is whatever falls out of where it sits. Volatility based placement never asks that question. It asks a narrower one: how far does this instrument travel in an ordinary period of the length being traded, so that the distance is not sitting inside the movement a quiet day produces.

The two are also denominated differently, which is why they so rarely agree. A structural level is a price, found on the chart and then measured back to the entry. A volatility derived level is a distance, calculated first and then subtracted from an anchor to produce a price. Neither construction contains the other, so a chart on which both are drawn shows two levels rather than one level and a version of it.

How the distance is derived 

The derivation takes three inputs, and trading literature is usually explicit about the first two and silent about the third. The first is a measure of range. Average true range is the conventional choice, for the two properties set out in the charts module: it is expressed in the instrument's own price units, and it carries no direction, so it describes how far rather than which way. The second is a multiple, chosen by the practitioner. Multiplying the reading by the multiple returns a distance in price units, which is a number of points.

The third input is the anchor, and it is the one most often left implicit even though it moves the resulting price as much as the multiple does. A distance is not a level until it is subtracted from something. Three anchors circulate: the price at which the position was opened, the closing price of the bar that preceded it, and the extreme of that bar, meaning its low for a long position and its high for a short one. One reading and one multiple therefore produce three different prices depending only on which is used. Anchoring at the bar's extreme places the level beyond a price the instrument has already traded at; anchoring at the entry ignores where the bar has already been.

A fourth choice hides inside the measure itself: the length of the bars it is computed on. Average true range is an average of bars, so a reading taken from five minute bars and one taken from daily bars, on the same instrument at the same moment, are different quantities, and the distances derived from them differ accordingly. The convention most commonly reported is to measure on the bar length the position is held over, on the reasoning that a distance derived from far shorter bars describes a stretch of price the position is expected to sit through. That is a convention rather than a result.

The same chart, both ways 

Drawing both levels on one chart produces two distances, and the informative part is which of them is larger. When the structural level is the nearer of the two, the price the chart offers sits inside the range the instrument covers in an ordinary period, so it is a price the market reaches without anything of note having happened. When it is the further, the measured distance rests short of the feature it was meant to sit beyond, so the instrument can trade through the level and back without the chart reading it was placed against having changed at all.

The consequence runs past the level and into the size, because the distance is the denominator of the sizing chain set out earlier in this module. Two distances on one chart are also two position sizes for one risk amount. The method that produces the wider distance produces the smaller position, and the amount at stake if the distance is reached is identical in both cases, since that amount was fixed before either level was drawn.

The last sentence of that note belongs in the body as well. No figure appears on this page for how often either distance is reached, because no such figure exists independent of the bar length, the lookback, the smoothing, the multiple and the anchor, all fixed before any counting could begin. A number produced after those five choices is a description of the choices.

The choice between the two distances is resolved in three named ways, each carrying the objection the other two make of it. One convention takes the wider of the two distances, on the argument that a level inside ordinary travel is reached by the instrument doing nothing in particular; the objection is that it commits a position past the price at which the chart reading was already contradicted. A second takes the structural level regardless of the measure, since the chart is where the reading came from; the objection is that a level inside the ordinary range makes the placement a function of how quiet the instrument happened to be that week. A third places the level at the structural feature plus a fraction of the measured range as a buffer, and is criticised by both camps for importing the arbitrary input of each method instead of choosing between them.

What a change in regime does to it 

A measure moves. A level, once resting as an instruction, does not. That asymmetry is the whole of what happens when volatility changes, and it is easy to miss because nothing on the screen announces it. The distance was derived from a reading taken at one moment and then converted into a fixed price, while the reading carries on being recalculated on every new bar. Through a stretch of expanding range the same untouched level is a smaller and smaller multiple of the current reading, until a distance that was two ranges wide is one range wide, and then less.

Two accounts of what follows are in circulation, and the arithmetic above is what makes the argument between them sharp rather than academic. One holds that a level derived from a reading which no longer applies is no longer the level the method specifies, so the derivation is repeated on each new bar and the level tracks the measure. The other holds that moving a level further away enlarges the amount at stake without a single new fact about the position having arrived, as the two adverse rows show, and that a rule capable of moving an exit further away is not a limit on anything.

The ratcheting conventions, described in trading literature as trailing volatility stops or as chandelier constructions, attempt to hold both positions at once: the distance is re-derived on every bar, but the level is only ever carried in the direction that reduces the distance still to be travelled, never in the direction that enlarges it. That construction has its own known failure, stated by its own practitioners. A level that only ever tightens converges on the price, so in a long enough quiet stretch it ends up inside the range the instrument covers in an ordinary period, which is precisely the condition the measure was introduced to detect. The ratchet does not remove the problem. It relocates it to the far end of the position.

One property of the measure sharpens the whole argument. An average of periods that have closed cannot describe a period that has not, so the reading is smallest immediately before range expands, which means the narrowest level the method produces is the one standing when it does.

What neither approach removes 

Both methods answer where an instruction rests. Neither answers what price it is filled at, and those are different questions with different answers. A stop is an instruction that becomes an order once the market reaches the stated level, and the order is then filled at the next price available. Where the market is trading through every price in between, the two are close together. Where it is not, because a session has closed and reopened elsewhere or because a release has moved the quote in one step, the order is filled beyond the level by whatever distance the market skipped.

There is an irony in the volatility method specifically, and it is worth naming. The phenomenon average true range was constructed to measure, price moving between bars rather than within them, is the same phenomenon by which an instruction resting inside a range of prices is filled outside it. A large reading is a description of the market that produces gaps, not a defence against one. Widening a distance changes where the instruction rests. It changes nothing about whether the market trades at every price on the way there.

A smaller mechanism sits in the same gap. A chart is drawn from one side of the quote or from a mid, while the instruction is triggered by whichever side an instrument's contract specifications name, so the distance a chart displays and the distance an instruction is measured against are not automatically the same distance.

Where practitioners disagree 

The oldest objection to a derived level is that it lands wherever the arithmetic puts it, at a price nothing on the chart distinguishes from any other, and is arbitrary in a way that is merely harder to see than a level chosen by eye. The volatility traditions answer that a chart feature is not free of arbitrariness either, since which feature counts is a judgement made after the price has printed, and that a measured distance at least states its inputs in advance. A sharper version of the same argument holds that obvious chart features are where instructions cluster, so a derived level scattered away from them rests in thinner company. The counter is that the same reasoning also puts it in thinner conditions when it is reached. Neither half has ever been settled, and no figure on this page settles it.

The second disagreement is whether one multiple should travel across instruments at all. One tradition holds that it must, because the point of stating a distance in units of measured range is that the statement converts itself: one multiple lands at a different number of points on every chart without anyone deciding on either. The other holds that instruments differ in how their ranges are shaped and not only in how large they are, so a multiple calibrated on a market that grinds describes something else on a market that jumps. That argument turns on a property the measure does not capture, which is why reading it off the measure cannot resolve it.

What is not in dispute is narrower than the argument around it, and it is the honest summary of the method. Deriving a distance from a measured range makes that distance mean the same thing on every chart and in every regime, which a number of points chosen by hand does not. That is a claim about units and comparability. It is not a claim that the level is a better one, that it is reached less often, or that a measure built from bars that have closed describes the bar that has not.