Intermediate3 min read

Keltner and Donchian Channels

Two envelope constructions that answer different questions: one is built from average range, the other from the highest high and lowest low over a window.

MadStockAlerts Research · Updated August 28, 2026

What to take away

  • Keltner channels are a moving average plus and minus a multiple of ATR.
  • Donchian channels are simply the highest high and lowest low of a lookback.
  • Donchian is the original breakout construction, and it is entirely objective.
  • Keltner is smoother than Bollinger because ATR is smoother than standard deviation.
  • The Bollinger-inside-Keltner condition is a common definition of a squeeze.

MAD Academy Training Video · 0:45

Three Channels, Three Different Questions

Bollinger, Keltner and Donchian look alike and are built from completely different measurements — which is why they disagree.

This lesson is part of a Stock Alerts + Tools plan.

See the library

The two constructions

Keltner = EMA(n) +/- k x ATR(n)

  • n is conventionally 20, and k conventionally 2
  • the width responds to average range rather than to dispersion around the mean

Donchian = highest high over n periods, lowest low over n periods

  • no averaging and no parameters beyond the lookback
  • the upper band is by definition the level at which a new n-period high occurs

Donchian is the more objective of the two because it involves no smoothing at all. The upper band is a price at which the security has not traded for n periods, which is a fact rather than a calculation.

Where each one is used

KeltnerDonchianBollinger
Width driven byAverage true rangeThe extremes of the windowStandard deviation of closes
Responds to gapsYes, ATR includes themYes, extremes include themOnly through the closes
SmoothnessSmootherSteps, rather than curvesNoisier
Conventional useTrend envelopesBreakout definitionsVolatility and mean reversion
An envelope built from average range
An envelope built from average range41.047.053.059.065.0VolumePrice riding the upper band isordinary in a trend

Scroll the chart sideways to see all of it.

  • EMA 20
  • Bands
Because ATR is smoother than the standard deviation of closes, the bands widen and narrow more gradually than Bollinger bands do on the same series. Illustrative, not live data.

The squeeze definition

A widely used definition of a volatility squeeze compares the two: when Bollinger bands sit entirely inside Keltner channels, the dispersion of closes has fallen below the average range, which is a defined and objective condition.

The construction is worth knowing because it replaces a visual judgement with a comparison of two numbers. As with every squeeze definition, it says a move may be larger and says nothing about direction.

Donchian and the breakout tradition

The Donchian channel is the construction behind the best-known mechanical breakout rules: buy an n-period high, exit on an m-period low. Its appeal is that it has one parameter and no judgement.

  • The rule is entirely reproducible: two people applying it reach the same answer.
  • It is trend-following by construction, so it performs poorly in ranges by construction.
  • The lookback is the only choice, and optimising it across history is a textbook case of fitting.
  • It has been widely published for decades, which is the persistence question in the technical analysis article.

Choosing a lookback, and the fitting trap

Both constructions have one substantive parameter, and the temptation to optimise it against history is the clearest small-scale example of the overfitting problem in this whole library.

  • A shorter lookback produces more signals, earlier, and more of them wrong.
  • A longer one produces fewer, later, and captures less of each move.
  • Testing every value and selecting the best is fitting the parameter to the sample.
  • A parameter chosen for a reason and left alone is testable; one chosen by search is a description of the past.

The sensitivity check from the overfitting article applies directly. If a lookback of 20 works and 18 and 22 do not, the result is about the number rather than about the market.

The conventional values persist for the same reason the moving average conventions do: they are widely used, which gives them a coordination property that a privately optimised value does not have.

Educational content only. MadStockAlerts provides market commentary, research, and educational content. It is not personalized investment advice, and nothing here is a recommendation to buy or sell any security. Trading and investing involve substantial risk, including loss of capital. See the Risk Disclosure and Customer Agreement.