Volatility: ATR, Bollinger, Keltner and the RVI

Volatility tools answer questions about distance: how far this thing typically travels in a bar, how far it has strayed from its own centre, how wide to expect the next hour to be. Those are questions with real answers, and they feed directly into decisions you have to make anyway — where a stop has to sit to survive normal noise, and therefore how large a position can be.

They say nothing at all about direction. Every misuse in this family is some version of forgetting that.

Average True Range

Wilder again, 1978, and the most quietly valuable thing on this page. It starts by fixing a flaw in the obvious measurement.

The obvious measurement is high minus low. The problem is that it ignores what happened between sessions. If a market closes at 100 and opens at 94, and then trades in a tight one-point range all morning, «high minus low» reports a very calm day. Six points of movement just vanished from the record.

True range fixes it by taking the largest of three distances: today's high to today's low, today's high to yesterday's close, and today's low to yesterday's close. The gap is now inside the number. Average that over fourteen bars and you have ATR.

ATR earns its place because it converts into decisions without any interpretation. If normal movement in this instrument is currently X, a stop closer than X will be taken out by normal movement, and a position sized as though movement were half of X is twice the risk you think it is. Both of those are arithmetic, not opinion, which is rare on this list.

Two cautions. ATR is an average, so it lags a change in conditions — on the day everything changes, it is still describing last week. And it is in the instrument's own units, so comparing the ATR of two different markets tells you nothing until you convert both into money, which is what ticks, points and what a move is worth is about.

Bollinger Bands

John Bollinger, early 1980s. A 20-period moving average with a band above and below at two standard deviations of the closing prices in that same window.

Standard deviation is a measure of dispersion, so the bands widen when recent closes have been scattered and narrow when they have been clustered. The useful consequence: the bands adapt on their own. You are not choosing a distance; the market's recent behaviour is.

The misreading is universal enough to be worth stating plainly, and its author has said so himself for decades:

Touching the upper band is not a sell signal. In a genuine trend price will ride the band for a long time — that is what a trend looks like through this particular lens. The bands describe where price is relative to its recent scatter. They do not contain it.

What the bands are genuinely good at is the opposite state: a squeeze, when they contract to unusually narrow. That says recent closes have been tightly clustered. It is one of the few readings on this page with a decent record of preceding something, and it comes with an important blank: it gives no hint of direction. A squeeze is a statement about energy, not about which way it will be spent.

Keltner Channels

Chester Keltner described a version of this in 1960; the form nearly everyone uses today — an exponential moving average with bands set a multiple of ATR away — is a later refinement.

On a chart the two look like cousins. The difference in the arithmetic is the point:

Bollinger — standard deviation of closes

Reacts to how scattered the closing prices have been. A session of wild swings that keeps closing in the same place barely widens them.

Keltner — a multiple of average true range

Reacts to how much ground each bar covered, gaps included. The same wild session widens these immediately.

Neither is more correct. They are measuring two different ideas of «volatile», and knowing which one you are looking at is the difference between a tool and a decoration. Because Bollinger bands react to a narrower thing, they contract more readily — and the observation that they sometimes contract inside the Keltner channel was built into a well-known setup, the «squeeze» popularised by John Carter. Worth naming so you know where it comes from when you meet it.

The Relative Volatility Index

Donald Dorsey, Stocks & Commodities, June 1993, with a revision of his own in 1995. This is the second of the two indicators that share the initials RVI; the Relative Vigor Index is a different thing by a different author, and they are constantly confused.

The construction is elegant and easy to state: it is RSI, with standard deviation substituted for price change. Where RSI asks «were the recent moves mostly up or mostly down», Dorsey's index asks «was volatility mostly rising or mostly falling», and puts the answer on the same 0–100 scale.

His stated reason for building it is the most interesting part and is worth taking seriously. Almost every indicator in this section is ultimately derived from price change. Stacking several of them and calling the agreement «confirmation» is, as the first page puts it, asking one witness twice. Dorsey's point was that a measure built on a genuinely different input is the only kind that can add anything — so it was designed as a companion, not a replacement.

The practical caution is the same as for its parent: rising volatility is not rising price. A falling market is a volatile market.

What none of them tell you

• Direction. Worth repeating once more, because every expensive mistake in this family is a version of reading a distance as a forecast.
• Why volatility changed. A scheduled release, a thin overnight session and a large participant leaving all register as the same widening. The calendar, not the indicator, distinguishes them — see how the NQ reacts to news.
• Whether the next move will be normal. Every one of these is an average of the recent past. The move that hurts you is, by definition, the one that was not like the recent past.