Bollinger Bands vs Donchian Channels

Michael Whitman · Trading since 2017, full-time since 2020 · Published 2026-09-08 · Updated 2026-10-19 · Reviewed 2026-10-04

Both draw lines above and below price. One draws a statistic and the other draws the highest high and lowest low of a period, which is a real price rather than a calculation. On thirty-two years of SPY daily bars, that difference meant the two rarely flagged the same day.

What each one is

Bollinger Bands plot a statistical envelope. John Bollinger’s default is a 20-day simple average of closes with lines two standard deviations of those 20 closes either side, and his own site says the standard deviation is the population form. Bollinger Bands covers it.

Donchian channels trace the extreme high and low of a fixed run of past sessions. StockCharts, which credits the idea to Richard Donchian, draws the 20-day version from the 20 sessions before the current one, so today’s high can break above the line. Nothing is smoothed. Donchian channels covers it.

One line is a number and the other is a price. That difference decides whether the level means anything to anybody but you.

Where they differ

Whether the line is a real price. A Donchian edge is a price the market actually traded at. A Bollinger edge is a computed distance from an average.

How the line moves. The statistical envelope adjusts every session. A Donchian edge holds still on most days, steps up when a new high prints, and steps down when an old high drops out of the 20-session window, which on SPY happened on more than one session in twelve.

What counts as an event. A new twenty-bar high is a named, countable event. A close above the upper band depends on a multiplier somebody chose, and on SPY the two coincided far less often than their shared reputation as breakout tools suggests.

How wide each one is. Neither is reliably the narrower. The Donchian channel was narrower than the bands on a bit over a third of sessions and wider on the rest, so a rule written for one does not transfer to the other.

Why a twenty-day high and a band close disagree

Because they read different parts of the bar. The Donchian break here is a high above the prior 20 highs; the band reading is a close above a line built from closes. A session can poke a new high intraday and settle back, which is a Donchian event and not a band event.

Because steady trends shrink the bands. When closes rise in even steps, their standard deviation stays small and the upper band hugs price, while each new high is still only a new high. When one violent close widens the bands, price can make a new 20-day high well inside them.

And because the band moves while the channel waits. The upper band is recomputed from every close; the channel only changes when a high enters or leaves the window. On SPY its upper line was unchanged from the previous session about seven days in ten.

Which one to use: Bollinger Bands or Donchian Channels?

Use Donchian channels when the rule is a breakout. A new extreme over a stated period is an event anybody can check against the price record, so its history on a given instrument can be counted before any money depends on the answer.

Use Bollinger Bands when the width itself is the reading. Narrowing dispersion is a genuine feature of the statistical version and it does not exist on the extremes-based channel.

Use Donchian channels when you want a level others share. The highest high of the last 20 sessions is the same number on every platform, which a band whose multiplier you set is not.

And when you want a stop that steps rather than drifts, use Donchian, with the warning that it also steps down as old highs leave the window. A trailing exit at the opposite edge is one of the plainest exits available.

Settings to fix before a channel or band rule

The lookback period. Twenty and fifty-five are the conventional Donchian settings and both were chosen elsewhere; on SPY the 55-day high still broke on about one session in six.

The multiplier, on the statistical version. It decides how often a close lands outside, which is the whole reading.

Whether a wick or a close counts. On SPY, 44.07% of the sessions that pushed above the old 20-day high closed back at or below it the same day. A rule must say which event it means.

And what the stop is. Neither channel supplies one, and it has to be written down before entry, sized for the widest bars the instrument has recently printed.

The original data

The sample. Yahoo Finance daily SPY bars, highs, lows and closes as traded, from 29 January 1993 to 2 October 2026, measured from 3 January 1994: 8,243 sessions. Opens are not used. Bollinger: the 20-close simple average with lines two population standard deviations above and below it. Donchian: the highest high and lowest low of the 20 sessions before the current one (55 for the longer version); a break up is a high above that line.

How often each fired. The high topped the prior 20-day high on 1,720 sessions (20.87%) and the prior 55-day high on 1,318 (15.99%). The low broke the prior 20-day low on 732 (8.88%). The high reached above the upper Bollinger Band on 929 sessions (11.27%).

How often they agreed. Of the 1,720 new 20-day highs, 407 (23.66%) also closed above the upper band, which is 4.94% of all sessions. The overlap runs one way: 92.08% of closes above the upper band were also new 20-day highs, so the band close is close to a subset of the channel break. The upper Donchian line rose on 20.87% of sessions, fell on 698 (8.47%) and was unchanged on 5,825 (70.67%). Its median width was 5.71% of the close, from 1.53% on 29 June 2017 to 49.34% on 15 October 2008, and it was narrower than the bands on 3,042 sessions (36.90%).

What came next. “Higher” means the close 10 sessions later was above the close on the day, price only. Any day with 10 sessions to come: 5,005 of 8,233 (60.79%) were higher, median change +0.65%. Each kind of day is tested against all the other days with a Fisher exact test.

The one low p-value is not a finding. New 20-day highs cluster: 1,556 of the 1,719 gaps between them were 10 sessions or fewer, so their 10-session windows overlap and the test treats one run of highs as hundreds of separate cases, which makes p too small. Counting only fresh highs, whose windows never overlap, the difference vanished (p = 1.00). Seven comparisons were run: the four in the table, the 20-day low and closes beyond either band (59.28% higher after a close above the upper band, p = 0.52; every other p above 0.38). Run seven tests when nothing is going on, and a p-value this low turns up in at least one of them roughly one time in five.

Longer did not mean better. After a 55-day high SPY was higher 10 sessions later 62.37% of the time, against 63.02% for the 20-day version; 63.20% and 63.78% respectively closed above the broken level 10 sessions on. Every session’s lines are in the SPY Bollinger and Donchian file, and the table in the 10-session summary.

When it fails

The failure is fading a band close during a breakout, and the two tools are pointing opposite ways. Price makes a new twenty-bar high, which the Donchian channel reports as a break, while the Bollinger edge reads as stretched. Fading the band means shorting a new extreme that other participants can see, and on SPY the days after a close above the band went much like any other days, so the stretch reading did not supply the reversal the short needs.

The second failure is treating a new high as a forecast. On SPY the 10 sessions after a fresh 20-day high were no different from any 10 sessions.

A third is using a lookback you did not check. The 55-day break fired almost as often as the 20-day one on this index.

A fourth is trusting a wick that pokes past the line. More than two in five of SPY’s new 20-day highs closed back at or below the old high the same day.

A fifth is resting a stop right at the channel line. The edge is the most visible price on the chart, and one oversized bar can run straight through it.

And a sixth is running both as confirmation. They read the same bars and still disagreed on most breakout days.

Related

Bollinger Bands covers the statistical envelope. Donchian channels covers the highest-high-lowest-low version. And breakout covers the event the second one is built to mark.

This page is educational, not financial advice. Test every idea on your own charts before risking money.