Keltner Channel: ATR Bands vs. Bollinger Bands
A Keltner Channel is a 20-period EMA with bands set a multiple of ATR above and below it. The formula, the two different versions that share the name, the settings that scale down for day trading, and the one structural difference that separates it from Bollinger Bands.
A Keltner Channel is three lines on the price chart: a moving average down the middle, and a band above and below it set by the average true range. That description sits close enough to Bollinger Bands that plenty of traders plot one, conclude it is the other with extra steps, and move on. It is not. The two indicators disagree about what volatility even is, and that single disagreement changes the shape of the bands, what a band touch means, and which kind of trade the tool is any good at. There is also a wrinkle in the history worth knowing before you quote anyone on it, because the formula almost everybody runs today is not the one Chester Keltner published, and the person who wrote the modern version is not the person the indicator is named after.
Quick Answer
A Keltner Channel is a volatility envelope made of three lines. The middle line is an exponential moving average, conventionally 20 periods. The upper band is that EMA plus a multiple of Average True Range and the lower band is the EMA minus the same multiple, with 2.0 the standard multiplier. Because the width is set by ATR rather than by the standard deviation of price, the channel is smoother and slower to flare than Bollinger Bands, which is why it gets pointed at trends and breakouts rather than at mean reversion.
What Is a Keltner Channel?
It is an overlay, not an oscillator. Nothing about it lives in a sub-panel and nothing about it is bounded between two numbers. The channel is drawn on the price chart itself, and it answers one question: how far is price from its own recent average, measured in units of how much this instrument typically moves in a bar?
That last clause is the part that matters and the part most write-ups skate past. Two dollars away from a 20 EMA means something completely different on a name that swings four dollars a day and on one that swings forty cents. ATR is the unit conversion. By expressing distance-from-average in multiples of average bar range, the channel gives you a reading that is comparable across instruments and across volatility regimes on the same instrument, which a fixed-percentage envelope cannot do.
Practically you get three usable objects out of it. The centerline is a trend reference. The outer bands are a stretch reference. And the width between them is a volatility readout in its own right, independent of direction, which widens when the tape gets violent and narrows when it goes to sleep.
What Is the Keltner Channel Formula?
Middle line = EMA of price, conventionally 20 periods
Upper band= Middle line + ( multiplier × ATR )
Lower band= Middle line − ( multiplier × ATR )
The standard multiplier is 2.0. The same number is used above and below, so the channel is always symmetrical around the EMA.
Run it on numbers. Say a stock's 20-period EMA sits at 48.20 and its ATR reads 0.85. The upper band is 48.20 plus 1.70, which is 49.90. The lower band is 48.20 minus 1.70, which is 46.50. The channel is 3.40 wide, or four ATRs from edge to edge, which is what a 2.0 multiplier always gives you.
Now hold price completely still and double the ATR to 1.70. The bands move to 51.60 and 44.80 and the channel is suddenly 6.80 wide, with the EMA unchanged. Nothing directional happened. That is the useful mental model: the centerline carries direction and the width carries volatility, and they move for entirely separate reasons.
One setting to check on your own platform before you compare charts with anyone. The EMA period and the multiplier are both shown in every settings panel, but the ATR look-back is not always exposed, and implementations disagree about it. Some match the ATR window to the EMA period, others run a shorter one. Neither is wrong, and the difference is small on a daily chart and not small on a 1-minute one. The StockCharts ChartSchool page on Keltner Channels walks the calculation through a worked series if you want to watch each input move bar by bar, and the thinkorswim study reference documents what one major platform ships as its defaults.
Two Formulas, One Name
This is the piece of the story that gets flattened into a single sentence almost everywhere, and flattening it makes the sentence wrong.
Chester W. Keltner described the concept in his 1960 book How to Make Money in Commodities. His version had nothing to do with EMAs and nothing to do with ATR. The centerline was a 10-day simple moving average of the typical price, meaning the average of the high, the low and the close. The bands sat a 10-day simple moving average of the high-low range above and below that centerline. Same three-line shape, entirely different arithmetic.
The version everyone actually runs today, the 20-period EMA with ATR-scaled bands, came out of the 1980s and is attributed to Linda Bradford Raschke. There is a clean structural reason it could not have been the original: Average True Range was not published as an indicator until nearly two decades after Keltner's book, so the 1960 formula could not have used it even in principle. The Wikipedia entry on the Keltner channel carries both versions side by side, which is the fastest way to see how far apart they are.
Why this is worth two minutes rather than being pub trivia: a small number of charting packages still ship the original high-low-range construction, or offer it as an option, and it does not behave like the ATR version. True range accounts for the gap from the prior close and a plain high-low range does not, so on a gappy instrument the two channels can be visibly different widths on the same bars. If your channel looks nothing like the one in a tutorial, this is one of the candidate explanations before you start blaming your settings.
How Is a Keltner Channel Different From Bollinger Bands?
Structurally there is one difference and everything else is downstream of it. Keltner sets its width from Average True Range. Bollinger sets its width from the standard deviation of price. Both then add and subtract that width from a moving average, and both use 2.0 as the conventional multiplier, which is exactly why they look like siblings.
The two inputs are measuring different things. Standard deviation is a statistic about closing prices, describing how scattered the closes in the window are around their own mean. ATR is a statistic about bar size, averaging how much ground each bar covered including any gap from the previous close. A market can run hard in one direction with consistent bar sizes and steadily advancing closes, and its standard deviation will stay fairly moderate while the move is obvious to look at. A market can also chop violently around an unchanged level and blow its standard deviation out while going nowhere.
Because of that, Bollinger Bands flare and pinch dramatically while Keltner Channels breathe. ATR is an average of bar ranges, so a single big bar moves it by roughly one nth of its look-back and no more. Standard deviation reacts harder to outliers because the deviations are squared before they are averaged. Same chart, same 2.0 multiplier, and one envelope is visibly more nervous than the other.
The second structural difference is the centerline. Keltner conventionally rides an exponential moving average, which weights recent bars more heavily and turns sooner. Bollinger conventionally rides a simple moving average, which gives every bar in the window equal weight. If you already run an EMA stack on your intraday charts, a Keltner centerline will feel familiar immediately, because it is the same construction with error bars attached.
| Field | Keltner Channel | Bollinger Bands |
|---|---|---|
| What it measures | Distance from a moving average in units of average bar range | Distance from a moving average in units of statistical dispersion of closes |
| Width mechanic | Average True Range (ATR) | Standard deviation of price |
| Moving-average type | Exponential moving average, recent bars weighted more heavily | Simple moving average, every bar in the window weighted equally |
| Reactivity | Smoother. ATR is a steadier input, so the channel breathes | Sharper. Standard deviation spikes and collapses, so the bands flare and pinch |
| How a band close is usually read | Strength. Price is extending in the direction of the move | Stretch. Price is far from its mean and may revert |
| Gap handling | True range includes the gap from the prior close, so an overnight gap widens the channel | Standard deviation of closes never sees the gap itself, only where the closes landed |
| Best-suited use | Trend identification and breakout confirmation | Mean reversion and volatility-compression setups |
| Typical hold horizon | Longer. The bands lag, so a signal stays valid for more bars | Shorter. The bands react faster, so a signal expires faster |
| Default settings | 20-period EMA, bands at 2.0 x ATR | 20-period SMA, bands at 2.0 standard deviations |
The only row in that table you have to internalise is the width mechanic. Every other row is a consequence of it. Pick the channel whose definition of volatility matches the trade you are trying to describe: if you want to know whether price is extending, ATR is the better unit, and if you want to know whether closes have scattered away from their mean, standard deviation is.
ATR sets a width that breathes. Standard deviation sets one that collapses and flares.
What Is the Bollinger-Inside-Keltner Squeeze?
The best-known use of the two indicators together falls straight out of the reactivity difference. Plot both on the same chart with standard settings. When the tape goes quiet, the Bollinger Bands contract faster and further than the Keltner Channel does, because standard deviation collapses harder than ATR does. Push that far enough and the Bollinger pair ends up sitting entirely inside the Keltner pair.
That nesting is the squeeze. It is a statement that volatility has compressed to an unusual degree relative to the instrument's own recent bar ranges, which historically precedes expansion more often than it precedes more of the same. The concept was popularised by John Carter and shows up under various names on various platforms. The mechanic is worth understanding even if you never trade it, because it is the clearest demonstration that the two envelopes are measuring different things. If they measured the same thing they could not cross.
Two honest limits. The squeeze has no directional content whatsoever, so it tells you a move is more likely and nothing about which way, which is why every serious treatment of it pairs the signal with something that does carry direction. And the release can be a head fake that reverses within a few bars, which is a failure mode covered in more depth in the write-up on how a Bollinger Band squeeze behaves on momentum names. This is also a technique you set up on your own charting platform, not something a chart-reading tool infers for you, a distinction worth keeping straight and one I come back to at the end.
Price closed outside the band. That still leaves the question of whether the rest of the chart supports the trade.
Upload the screenshot and SnapPChart reads that one image against a fixed rubric, then returns a setup grade, an entry, a stop with the reasoning behind the level, targets, and the reward-to-risk those levels imply. The grade is the part worth arguing with before money is on it.
Grade this chartHow Do You Use Keltner Channels to Trade a Trend or a Breakout?
The conventional read is direct. A close above the upper band is bullish, a close below the lower band is bearish. Price has travelled more than two average bar ranges past its own EMA and finished there, which is not something a directionless market produces very often.
The part that costs people money is what happens next. In a real trend price does not tag the band and come back. It rides it, closing outside or right on the upper band for bar after bar, and every one of those bars looks like a fresh overextension reading. It is not. A sustained stretch against the band is a description of trend consistency, not a queue of reversal warnings, and fading the third or fourth one because the first two felt like too much is the single most reliable way to lose money with this indicator. Nothing in the calculation has any notion of due.
The centerline is where the trades usually are. Both the EMA and the outer bands function as dynamic support and resistance, and inside an established trend the common re-entry is a pullback to the centerline that holds. You are buying a retracement into a moving reference rather than chasing a band close, which gives you something the band close never does: a defined invalidation just below the centerline, close enough to size against. Those pullbacks are considerably more trustworthy when the centerline lands on top of a horizontal level you had already drawn for reasons that have nothing to do with the channel.
On breakouts, insist on the close. A wick through the band is not a band close, and the distinction is the entire filter. Wicks outside the channel happen constantly on any timeframe with reasonable volatility. A bar that finishes outside it is a much smaller population. Most tactical treatments add volume confirmation on top, on the reasoning that an expansion bar without participation behind it is a candidate to be given back.
Where the whole framework falls apart is a range-bound, low-conviction chart. In chop, price pings between the bands, closes outside them on noise, returns to the centerline six times a session, and every rule above fires repeatedly and means nothing. The channel cannot tell you which regime you are in, because width is a volatility reading and not a trend reading. That is the argument for pairing it with a trend-strength measure such as ADX, which is computing something genuinely unrelated and will happily tell you the tape is going nowhere while the bands look interesting.
What Settings Should You Use for Day Trading?
The published default is a 20-period EMA with a 2.0 ATR multiplier, and it is a daily-chart default. Dropped unchanged onto a 5-minute chart it tends to be too wide to be useful, because the channel is calibrated to contain a daily bar and an intraday leg often finishes before a close clears it.
The intraday adjustment goes in one direction on both dials. Faster EMA, roughly 10 to 15 periods, and a tighter multiplier, roughly 1.0 to 1.5. The trade-off is fixed and there is no setting that escapes it: a wider multiplier gives fewer signals of higher average quality, a narrower one gives more signals and more noise. You are choosing where on that line to sit, not optimising your way off it.
| Setting | What it is | What it does | What it costs you |
|---|---|---|---|
| 20 EMA, 2.0 x ATR | The published default on most platforms | On a daily chart the bands contain nearly all of the bar range. A close outside is genuinely uncommon | Ported straight onto a 5-minute chart the channel is wide enough that an intraday leg can finish before a close clears it |
| 13-15 EMA, 1.5 x ATR | The common intraday tune | Tighter channel, faster centerline. Closes outside the band happen several times in a session | A larger share of those closes are noise, so the thing that filters them has to come from somewhere other than the channel |
| 10 EMA, 1.0 x ATR | Scalping territory | On a 1-minute chart price is outside one band or the other most of the time | The band close stops being an event. You are watching a fast EMA with decoration attached |
| 20 EMA, 3.0 x ATR | Widening the multiplier and leaving the centerline alone | Far fewer band closes, and each one represents a considerably larger move | Late by construction. The signal prints after most of the leg is behind you |
| 50 EMA, 2.0 x ATR | Slowing the centerline instead of the width | Centerline pullbacks become rarer and deeper, which suits a multi-day hold | On an intraday chart the centerline barely moves, so the pullback entry stops appearing at all |
| ATR period matched to EMA period | The consistency tweak rather than a speed change | One look-back governs both the centerline and the width, so the whole channel responds to a single time horizon | Nothing, except that your chart no longer matches a platform shipping a shorter ATR window by default |
The thing the table cannot show is that the period interacts with the clock, not just with the bar count. A 20-period EMA on a 5-minute chart spans 100 minutes, which is a reasonable window on a liquid name. The same 20 periods on a 1-minute chart during the opening range spans a stretch where the ATR input is dominated by the two or three widest bars of the day, so the channel is enormous for the next twenty minutes regardless of what price is doing. Think in minutes and hours before you think in periods.
Where the Keltner Channel Fits on a Real Chart
It is a context layer, not a trigger. The order that works is regime, then direction, then level, then timing. Establish whether the tape is trending or rotating, which the channel cannot do for you. Pick a side. Mark the price you would act at and the price that proves you wrong, both of which have to come from structure. Only then does the channel earn its place, and what it contributes is scale: it tells you whether the distance between your entry and your stop is large or small relative to how much this instrument actually moves.
That sequencing, and the broader case for treating overlays as a layer on top of structure rather than a replacement for it, is the spine of the technical analysis overview. The related question of how many independent reads you want agreeing before you commit is handled in the piece on how much confluence is enough, and there is a genuine wrinkle here worth naming. Keltner and Bollinger really do count as two measurements, because their width inputs are computed from different data. Keltner and a bare ATR reading do not, because one is the other with a moving average wrapped around it.
What a chart grader can and cannot see here
Worth being straight about, since this site sells a tool. SnapPChart has no Keltner Channel field. It does not calculate an EMA, it does not calculate ATR, it does not construct bands, and it carries no dedicated state for a channel the way it does for the moving average stack, the VWAP relationship, or the MACD cross. What it does is read a chart screenshot you upload, so if you plot a Keltner Channel on your own platform before taking that screenshot, the visible bands are part of what the analysis sees, as geometry drawn on the price chart. That is a picture of a channel, not a verified computed value. It cannot confirm that your multiplier is 2.0, it cannot tell an ATR channel from the original high-low-range version, and it emphatically does not detect a Bollinger-inside-Keltner squeeze, because that requires tracking two computed widths over time and it is tracking neither. Which indicator states a screenshot-based read actually carries and which it only infers from shape is the subject of the wider guide to how AI reads a chart, and a neutral description of what a single chart read covers sits on the AI chart analysis page. If a band close is the reason you are taking the trade, verifying that band close stays your job.
A Keltner Channel is a 20-period EMA with bands at 2.0 times ATR. The original 1960 formula was different arithmetic entirely and the modern one dates from the 1980s. Against Bollinger Bands the only structural difference is the width input, ATR against standard deviation, and everything else follows from it: Keltner is smoother, sits on an EMA, and suits trends and breakouts, while Bollinger reacts harder and suits mean reversion. Require a close outside the band rather than a wick, never fade a band that price is riding, and take centerline pullbacks over band chases. For intraday work drop to roughly a 10 to 15 period EMA and a 1.0 to 1.5 multiplier, and accept that the tighter you go the more noise you buy.
Frequently Asked Questions
Can you run Keltner Channels and Bollinger Bands on the same chart?
Yes, and unlike a lot of indicator stacking this one is not redundant. Two oscillators that both measure range position are the same series drawn twice. Keltner and Bollinger are not, because one sets its width from average true range and the other from the standard deviation of closes, so the two widths genuinely move apart from each other. That gap is the whole point of the squeeze setup. What you should not do is count a touch of the Keltner upper band and a touch of the Bollinger upper band as two independent confirmations of direction. They are two readings of volatility taken at the same moment, and neither one is telling you which way price goes next.
Why does my Keltner Channel look different from someone else's screenshot?
Four settings can differ and most platforms only show you two of them by default. The EMA period is the obvious one, 20 being the common default but 10 and 50 both shipping somewhere. The ATR multiplier is the second, usually 2.0 but frequently tuned down for intraday charts. The third is the ATR look-back, which platforms do not agree on, with some matching it to the EMA period and others using a shorter window. The fourth is the one people miss entirely: a handful of implementations still ship the original high-low-range version rather than the ATR version, and that channel behaves differently enough that comparing it to an ATR channel is comparing two indicators. Read the settings panel before you read the bands.
Do Keltner Channels work on low-volume stocks and low timeframes?
Mechanically yes, since the calculation only needs a high, a low and a close per bar. Practically the ATR input degrades in exactly the places you would expect. On a thin name a single wide bar, or one gap on no real volume, sets a true range that the average carries for the whole look-back window, so the channel stays inflated long after the tape went quiet again. On a 1-minute chart in the dead middle of the session the opposite happens: the ranges are so compressed that the channel pinches to almost nothing and every small push closes outside it. The indicator is not broken in either case. It is faithfully reporting a volatility estimate built from bars that were not worth measuring.
Is a touch of the upper Keltner band a sell signal?
Usually it is the opposite read. A close above the upper band is conventionally taken as strength, because it means price moved more than a couple of average bar ranges above its own EMA, and that is what the start of a leg looks like. The mean-reversion instinct comes from Bollinger Bands, where a band touch does get read as stretch, and importing that habit into a Keltner Channel is the most common way people lose money with it. There is a narrow case where fading a band touch is defensible: a market you have already established is rotating in a range, where the band touch lines up with a horizontal level you drew independently. In that case the level is the trade and the band is the timing. In a trend, the band touch is the trend.
If you only change one Keltner setting, which should it be?
The ATR multiplier. Moving the EMA period changes two things at once, how the centerline tracks price and where the bands sit, so you end up with a different-looking chart and no clean read on which change caused what. The multiplier only scales the width, which means it does one job: it sets how often price closes outside the channel. Raise it and you get fewer band closes, each representing a bigger move, at the cost of being later to every one of them. Lower it and you get more of them, a larger share of which are noise. That is a single, legible trade-off, and it is the one worth tuning first on a timeframe you have not used the indicator on before.
This article is for educational and informational purposes only and is not investment, financial or trading advice. The Keltner Channel formula, the 20-period EMA and 2.0 ATR multiplier defaults, the attribution of the original 1960 construction to Chester W. Keltner and of the modern EMA and ATR version to Linda Bradford Raschke, and the attribution of the Bollinger-inside-Keltner squeeze to John Carter are the conventional published formulations reproduced by charting platforms and reference sources. The alternative EMA periods and ATR multipliers quoted are configurations commonly cited in that literature rather than tested or recommended parameters, and no combination of them is claimed to be profitable. Nothing here is backtested performance and no win rate, edge or result is claimed or implied. Indicator readings describe what price has already done and do not predict what it will do next. Day trading and active trading carry a substantial risk of loss and are not suitable for every investor. SnapPChart grades a static chart screenshot you upload and returns a setup grade, entry, stop, targets and reasoning for that single image; it does not calculate a Keltner Channel, an EMA, an ATR value, a Bollinger Band or any other indicator or band value itself, does not detect a squeeze between two channels, does not track indicator states it has not been shown, does not scan the market, and does not track your account, positions or P&L. It can only account for indicators that are visibly drawn on the image you upload. Do your own analysis, size positions so that being wrong is survivable, and consider speaking to a licensed financial professional about your own circumstances before trading.
Writes about AI-assisted day trading, technical analysis, and the systems traders actually use to stay disciplined.
The channel tells you how far price is from its average. It has no opinion on whether the chart is worth trading.
A close above the upper band looks the same on a clean continuation and on the last gasp of a move that has already run too far. The bands cannot separate them, because all they know is an EMA and an average bar range. Upload the screenshot and SnapPChart reads that one image against a fixed rubric, then returns a setup grade, an entry, a stop with the reasoning behind the level, targets, and the reward-to-risk those levels imply.