How to Use Keltner Channels
To use Keltner channels, treat the middle line as the trend and the bands as a measure of how far price normally travels from it. Because the width comes from the average true range, the channels expand and contract more smoothly than deviation-based bands.
Keltner channels draw a moving average with two bands offset by a multiple of the average true range. The construction is what distinguishes them: they measure how far price typically travels, not how far it has deviated from an average.
Before you start
A decision about whether you are trading touches of the bands or breaks of them. These are opposite trades and the chart looks identical for both.
The instrument’s average true range, since that sets the channel width. On this site’s shared series the fourteen-period average true range has a median of 0.5994 and a ninetieth percentile of 0.7954.
An understanding that these are not Bollinger bands and behave differently. Both draw a line with two bands. The similarity ends at the picture.
The steps
1. Read the middle line as the trend
It is an exponential moving average. Its slope is the directional information, and the bands describe normal travel away from it rather than an extreme.
2. Set the multiplier, then the period
The multiplier decides how many average true ranges the bands sit away. It is the dominant input, and it is the one that determines how often price reaches a band at all.
3. Choose your interpretation before you look
Touch as exhaustion in a range, or break as continuation in a trend. Deciding after seeing the chart means the chart decides, and it will support whichever reading you brought to it.
4. Do not treat a band touch as a signal
In a trending market price sits against the upper or lower band for many consecutive bars. Every one of those is a touch and almost none of them is a reversal.
5. Watch the width rather than the level
A narrowing channel means the average true range is falling, which means the market has gone quiet. That is a more useful observation than any single touch.
6. Compare against deviation-based bands
When Bollinger bands contract inside these channels, deviation has fallen faster than range. That is a specific, nameable condition, and it is what most people are looking for when they add either indicator.
7. Take the stop from structure
The opposite band is a distance, not a place anybody is defending. Where the idea is wrong is structural, and that distance sets the position size.
How to tell it worked
The interpretation was written down before the indicator went on the chart.
0 trades were taken on a band touch alone, without a structural reason.
The multiplier was set deliberately, with a stated reason rather than a default.
And each stop came from the chart, so 1 adverse move costs your intended risk.
What the construction means
Average true range includes gaps; standard deviation does not. A market that gapped and then moved quietly produces a wide channel here and narrow deviation bands, and that difference is real information about what happened.
Which also means an illiquid instrument produces tight channels. Small range, narrow bands, frequent touches — none of it describing an active market.
Against Bollinger bands, specifically
Bollinger bands widen when returns become dispersed. A few large moves in either direction expand them sharply, because standard deviation is sensitive to outliers.
Keltner channels widen when the typical bar gets bigger. That responds more slowly and more smoothly, and it does not care whether the movement was one-directional.
The disagreement between them is the technique. Deviation bands inside the channels means recent moves have been unusually orderly relative to their size, which is a condition worth naming rather than a line to trade against.
Setting the two inputs
The moving average period sets how quickly the centre line responds. A shorter period tracks price closely and the channel follows it around; a longer one holds a direction and lets price travel further from it before the bands are reached.
The multiplier sets how far the bands sit. Doubling it roughly halves how often price reaches one, which changes the indicator’s entire character without changing anything about the market.
Match the pair to your holding period, not to the chart’s appearance. A configuration that looks tidy across the visible window has been chosen by that window, which is the same fitting problem that applies to every parameter on every indicator.
And keep them fixed once chosen. Adjusting the multiplier because price reached a band is deciding that the band was wrong, which removes the only thing the indicator was contributing.
The original data
Of the 24,971 unique videos in research/search-study-corpus.jsonl, 126 mention these channels in the
title, at a median of 3,163 views across 101 channels, and 65% of those titles are instruction-shaped.
One trend indicator appears in 122 at 19,638 and a five-component system in 151 at 10,245. The counts
come from site/corpus_count.py.
126 videos at a 3,163 median — heavy coverage and one of the smallest audiences per video measured here. A great deal of explanation reaching very few people, which usually means the subject is being taught as a feature rather than as a decision.
The answer to the question on that chart is that five consecutive touches describe a trend. On this site’s series price traded through 85% of 39 twenty-bar breakout levels, so persistent contact with an upper band is the ordinary behaviour of a market going somewhere, not a queue of rejected attempts.
When it fails
The failure is fading band touches in a trend, and it produces a run of losses that each looked like a good entry. Price reaches the upper band, which reads as extended, so a short goes on. Price stays against the band for another eight bars. Each individual touch was a real touch and the indicator reported it accurately — the question being asked of it, whether the move is finished, is one a distance measurement has never been able to answer.
The second failure is deciding the interpretation after looking. The chart then chooses.
A third is confusing these with deviation bands. They respond to different things.
A fourth is stopping on the opposite band. It is a statistic, not a level.
A fifth is leaving the multiplier at a default. It is the dominant input.
And a sixth is reading tight channels as calm in a thin market. Small range is not stability.
Related
Keltner channels covers the calculation itself. Average true range is what sets the width. And Bollinger bands is the deviation-based alternative worth comparing against.
The use I actually get value from is comparing these against Bollinger bands rather than trading either alone. When the deviation-based bands contract inside these ones, something specific has happened to the character of the market, and that is a condition I can name — unlike a band touch, which happens constantly.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.