ATR, and the Unit Problem

Charts lie about scale, and they lie in a way that is almost impossible to see, because every chart you look at has been auto-scaled to fill its own window.

Two instruments can produce pictures that are indistinguishable — same swings, same shapes, same apparent volatility — while one of them moves four times as far per day as the other. The chart cannot show you this, because the chart's vertical axis was rescaled to make the picture fit.

So a stop "two dollars below the low" is not a risk decision. It is a risk decision on one instrument and a rounding error on another, and you cannot tell which by looking.

ATR converts every instrument, and every regime, into the same currency: how far this thing normally moves.

The arithmetic, which matters in one specific way

True range is the largest of three numbers: today's high minus today's low, today's high minus yesterday's close, or yesterday's close minus today's low. Then average it — fourteen periods is conventional.

The reason for those second and third terms is the only interesting part: they capture gaps. A stock that closes at 50 and opens at 44 has moved six dollars, but its high-minus-low that day might be one. Plain range would call that a quiet session. True range knows better, and that is exactly the kind of day that hurts people who sized as though the instrument were quiet.

Exhibit 1 · The same picture, four times the movementAuto-scaled, as every chart is

This is why "use a two-dollar stop" or "risk fifty points" are not rules. They are rules plus an unstated assumption about volatility that was true when the rule was written and stops being true the moment the regime changes — which it does, constantly, on the same instrument.

Express the stop in ATR rather than in currency, and the position size falls out of it. Three numbers, no judgement required once you have chosen the two inputs.

1. Decide what one trade may cost you. A fixed fraction of the account — one percent is a common choice, and the exact number matters far less than it being fixed.
2. Set the stop in ATR. Two ATR is a reasonable default for a swing position; the right multiple depends on how long you intend to hold, not on how confident you feel.
3. Divide. Position size equals the risk budget divided by the stop distance in currency. That is the whole calculation.

Exhibit 2 · Same risk, three instruments1% of a 100,000 account, 2 ATR stop

Note what this does to the question "how much should I buy." It stops being a question. Once the risk budget and the stop multiple are fixed, position size is an output, and the only decisions left are the ones worth spending attention on.

The claim in the introduction to this series was that a tool which fixes sizing beats a tool that picks entries. That is testable, and the mechanism is variance rather than expectancy.

Take one strategy with a genuine edge and trade it two ways. Both take exactly the same trades, with the same win rate and the same reward-to-risk. The only difference is that one sizes positions in fixed share counts and the other sizes them so that each trade risks the same amount of money.

Fixed share counts mean your dollar risk per trade moves with volatility — in a quiet regime each trade barely matters, in a wild one a single trade can cost several times what you intended. Same edge, same trades, radically different experience.

Exhibit 3 · Identical trades, two sizing rulesSimulated paths · same edge in both panels

The point is not that volatility-scaled sizing earns more. On average it does not. The point is that it narrows the distribution of outcomes, and a narrower distribution is what lets a real edge survive long enough to be realised. Most strategies do not fail because they were wrong. They fail because a run of oversized losses ended them before the arithmetic arrived.

Entry quality changes your expectancy a little. Sizing changes whether you are still there to collect it.

Beyond sizing, ATR is a decent description of which volatility state a market is in — and volatility, unlike price, has properties you can rely on.

Two of them are about as close to established facts as market statistics get. Volatility clusters: quiet periods follow quiet periods and violent ones follow violent ones. And it mean-reverts: extremes in either direction tend to resolve back toward normal.

Put together, those say something useful. A market whose ATR has compressed to the bottom of its own range is not a market that will stay quiet — it is a market where a volatility expansion is overdue. That is the statistical version of what the triangle and wedge pages describe structurally as a coil, and the two readings usually agree.

What it does not say is direction. Compressed volatility tells you a large move is likely and nothing whatever about which way. Every use of ATR on this page is direction-blind, which is the reason it combines well with everything else rather than competing with it.

• It is backward-looking and it lags. A fourteen-period average takes fourteen periods to fully register a regime change. In the first days of a volatility shock your ATR is describing the calm that just ended, and that is precisely when the sizing error is largest.
• It is not a forecast. ATR tells you what recently happened, not what the market expects. For a forward-looking measure you need implied volatility from the options market, which is a different instrument answering a different question.
• It says nothing about direction, and nothing about the tails. An average of ranges does not describe gap risk, and it cannot tell you about the day that moves ten ATR. Sizing rules built on it still need a hard limit for the outlier.
• ATR stops are not magic stop placement. A two-ATR stop is a statement about noise, not about where your idea is wrong. The structure pages are where invalidation levels come from; ATR tells you how big the position can be once you have one.
• Beware of double-counting. If your stop is already at a structural level, adding an ATR buffer on top and then sizing off the total quietly halves your position for no reason. Pick one method of setting the stop.

• Express every stop in ATR before you express it in money. "1.8 ATR" is comparable across instruments and across years. "Two dollars" is comparable to nothing.
• Make position size an output. Risk budget divided by stop distance. If you are still deciding how much to buy after you have chosen the stop, something has gone wrong.
• Recompute when the regime changes. The same instrument in a quiet month and a violent one needs different position sizes for the same idea. That is the entire point.
• Compare ATR as a percentage of price when you are looking across instruments. Raw ATR is in the instrument's own currency and is not comparable either.
• Watch ATR against its own history, not against an absolute number. A percentile over the last year tells you whether this is a compressed or expanded regime; the raw value tells you nothing on its own.
• Do not turn it into a signal. ATR crossing a threshold is not an entry. It is a measurement instrument, and the moment you trade it directly you have lost the thing that made it valuable.

A description of market mechanics, not trading advice. The simulation in Exhibit 3 illustrates the variance argument; it is not a claim about any real strategy's returns.