Bollinger Bands® vs Keltner Channels: What Actually Differs

Bollinger Bands® and Keltner Channels sit next to each other in almost every indicator list, and they look alike on a chart: a center line with an envelope above and below it. The usual comparison describes them by temperament. Keltner is smoother, Bollinger reacts faster, and the reader is told to pick whichever suits their style.

That description is not wrong, but it skips the reason. The two bands differ in exactly two arithmetic choices, and once those are visible, the smoothness, the reactivity and the arguments about which is "better" all become easier to read. This article sets out those two choices, what each does on a chart, and what one published test found when both bands were scored at their own default settings.

What each band computes

A standard Bollinger Band takes a simple moving average of closing prices, usually over 20 bars, and draws lines a set number of standard deviations of those same 20 closes above and below it. The common default is two standard deviations, written (20, 2).

A modern Keltner Channel takes an exponential moving average of price as its center, commonly 20 bars, and offsets the lines by a multiple of Average True Range, commonly a 10-bar ATR at 1.5 or 2 times. Settings vary more across platforms for Keltner than for Bollinger, so the numbers on any given chart are worth checking in the indicator's settings.

Both are price-space bands: both are built on price levels rather than on percentage moves. That shared property matters later. The differences are the two ingredients inside them.

Difference one: the center line

A simple moving average weights every bar in its window equally. An exponential moving average weights recent bars more heavily, so in a steady trend it sits closer to the current price.

That is a real difference, but a limited one. Both are averages of past price levels, and in a trend any average of past levels trails the current price. How that lag pulls a price-space band off-center is covered in the Bollinger Band drift problem. Swapping the average softens the effect without removing it.

Difference two: the width

This is the larger difference, and it is where most of the behavior people notice comes from.

Bollinger's width is the standard deviation of recent closes around their own average. Two things follow. Deviations are squared, so a single large close has outsized leverage on the width. And in a trending window, the closes spread out simply because price moved in one direction, so the trend itself is counted as dispersion. The working paper calls this width inflation.

Keltner's width is Average True Range: the average distance price travelled within each bar, measured from high to low, and extended to include any gap from the prior close. It measures how far price moves per bar, not how far closes have spread around an average. It does not count a steady trend as dispersion, and because it averages absolute ranges rather than squared deviations, one extreme bar moves it less. Why a close-to-close measure and a high-low measure see different things is covered in why closing prices matter.

What that looks like on a chart

Put the two ingredients together and the familiar temperaments appear. A Bollinger Band tends to balloon after a sharp move and in a sustained trend, then pinch in quiet stretches. A Keltner Channel's width changes more gradually, because bar ranges change more gradually than the dispersion of closes.

Some traders plot both at once and watch for the Bollinger Band narrowing inside the Keltner Channel. Mechanically, that only means the dispersion of recent closes has fallen low relative to the average bar range. What, if anything, tends to follow is not something Oisigma has published.

Why "2" does not mean the same thing in both

The multiplier is where head-to-head comparisons most often go astray. Two standard deviations carries a textbook meaning: under a normal distribution with known parameters, about 95.45% of observations fall within it. Two ATRs carries no equivalent target. ATR is measured in its own units, and a multiple of it has no built-in claim about what share of closes should fall inside.

So a Bollinger (20, 2) and a Keltner at 2×ATR are not two settings of the same instrument. One implies a coverage level by construction; the other does not state one. That is why "which is tighter" or "which is more accurate" has no answer until someone picks a target and measures both against it.

What one published test found

The working paper behind the Behavioral Transform Model did that. It scored both bands at their practitioner defaults — Bollinger (20, 2), and Keltner with an EMA-20 center and 2×ATR-10 width — on the same-bar basis chartists actually plot, where each bar is judged against a band whose window includes that bar. The target was the 95.45% a two-standard-deviation band nominally implies, and the sample was S&P 500 daily data from mid-1982 to 2024, restricted to bars with genuine intraday ranges.

On that basis, Keltner's containment sat 16.9 percentage points below the 95.45% target, and Bollinger's sat 4.7 points below, on the same sample. These are historical measurements on one index, and past behavior is not a guarantee of future results.

The fair reading is narrow. Keltner Channels were never designed to contain 95.45% of closes, so the gap measures distance from a target Keltner does not claim, not a failure of what it sets out to do. What the result does show is that reading a 2×ATR Keltner as if it were a two-standard-deviation envelope would have overstated how often closes stayed inside it by a wide margin, historically.

The paper also scored both constructions a stricter way, as one-step prediction intervals at a matched 60-bar window, and crossed the two ingredients to see which one drove the gap. The width basis mattered far more than the center line. Those results are set out in what we found when we tested the popular bands; the conclusion there, scoped to that comparison, was that none of the price-space constructions tested behaved like a calibrated prediction interval.

Where the evidence stops

The same-bar comparison covers one pair of indicators, at one pair of default settings, on one index, on daily bars, and one question: containment. It says nothing about breakouts, reversals, squeezes, or which band suits a trading style, and nothing published by Oisigma scores either band on those. The paper validates calibration only; whether any use of either band, or of Oisigma's own, delivers value after costs is an open question.

The takeaway

Bollinger Bands® and Keltner Channels differ in a center line and a width basis. The center line changes the lag a little; the width basis changes the band's character a lot, because the standard deviation of closes and the average bar range measure different things. Their multipliers are not like-for-like, so the question of which is "better" only becomes answerable once a target is chosen. Scored against the two-standard-deviation target at their own defaults on the same sample, Bollinger landed closer to it than Keltner did, historically — a result about calibration to that target, not about trading with either. The broader point is the one behind comparing any two bands: state the claim, then count.

Frequently asked questions

What are the default Keltner Channel settings on TradingView? At the time of writing, TradingView's built-in Keltner Channels uses a 20-period length, a multiplier of 2, an exponential moving average as the center and a 10-bar ATR for the width — the same practitioner default the working paper scored. Other platforms ship different defaults, often a 1.5 multiplier or a 20-bar ATR, so the published comparison applies only to that configuration.

Do Bollinger Bands® or Keltner Channels repaint? Both include the current bar in their calculation, so on the live bar they update with every tick until that bar closes; once it closes, its values stay fixed as long as the settings are unchanged. Changing the length or multiplier recalculates every historical bar, which is a recomputation under new settings rather than repainting. Oisigma's own indicator behaves the same way for closed bars, as described on How It Works.

Are Keltner Channels better for trending markets? Nothing Oisigma has published scores either band on trend-following suitability. What is published is the mechanism: both are centered on an average of past prices, so both trail a trend, and Bollinger's width also absorbs the trend as dispersion. How price-space and return-space bands behaved in trending versus sideways stretches is covered in do volatility indicators work in trending markets.

Is a Keltner Channel the same as the ATR indicator? No. ATR on its own is a single number per bar, plotted in a separate pane in the units of price. A Keltner Channel takes a multiple of that number and draws it around a moving average on the price chart itself. Why some volatility tools overlay the chart and others sit below it is covered in how to add a volatility indicator to TradingView.

Oisigma's own band makes the same kind of claim in return space, and its calibration record, misses included, is on the Proof page. If you'd like to see it next to the bands you already use, you can start a free 30-day trial.