Donchian Channels vs Standard Deviation
Michael Whitman
·
Trading since 2017, full-time since 2020
·
Published 2026-09-09
·
Updated
·
Reviewed
A Donchian channel is drawn from the top and bottom prints of a lookback, so both edges are prices that traded. Standard deviation measures how widely recent prices or moves spread around their average. It is a statistic, and reading it as a rarity scale assumes a bell-shaped spread that daily moves lack.
Both put a boundary around recent price. One takes the boundary from what happened: the highest and lowest prices traded. The other computes it from a spread statistic and, when read as a rarity scale, borrows an assumption about how often big moves come. On SPY that assumption fails at the edges.
What each one is
A Donchian channel is the range itself. Its upper line is the top print of the lookback and its lower line the bottom one, so both are prices someone paid. Donchian channels covers it.
Standard deviation measures how widely values spread around their own average. Applied to price it is the basis of bands around a moving average; applied to daily moves it is the usual measure of volatility. Standard deviation covers it, and Bollinger Bands covers the best-known band.
One is observed and one is computed. A Donchian edge is a print; a deviation edge is a distance from an average, and no trade need ever have happened there.
Where they differ
Whether the edge is a real price. The channel’s edge is the extreme itself. The deviation band’s edge is an estimate that moves every session even when no new extreme is made.
What each assumes. The channel assumes nothing about shape. Reading two or three deviations as rare assumes a bell curve, in which three-deviation moves are about one in 370.
How each reacts to one wild session. A single huge bar sets a Donchian edge and holds it there for the whole lookback. The deviation absorbs that bar into its average of squared distances, so it widens and then decays as the bar ages out.
How each behaves when nothing happens. A quiet stretch leaves the channel at old extremes while the deviation shrinks steadily, which makes the next ordinary move look large in deviation units.
Which one to use: Donchian Channels or Standard Deviation?
Use the channel when you want a level others can see. A 20-session high is a print every platform shows identically.
Use standard deviation when you want a live volatility reading. It updates every session, whereas the channel can sit unchanged for weeks.
Use the channel when you want an edge with no shape assumption, and stop treating any deviation multiple as a probability.
And when you do use deviation bands, count your instrument’s own tail before trusting what a multiple implies.
How often SPY moved past each multiple
The test. Each session’s close-to-close move was divided by the standard deviation of the previous 20 sessions’ moves, so it reads in the units a volatility band uses. From 3 January 1994 to 2 October 2026 each session counts once. A bell curve sets the expected share beyond each multiple.
| Move, in prior 20-session deviations | SPY sessions | Share | Bell curve share | Bell curve count |
|---|---|---|---|---|
| 2 or more either way | 614 | 7.45% | 4.55% | about 375 |
| 3 or more | 142 | 1.72% | 0.27% | about 22 |
| 4 or more | 39 | 0.47% | 0.0063% | about 0.5 |
| 5 or more | 20 | 0.24% | under 0.0001% | under 0.01 |
The tails were far fatter than the curve. Three-deviation moves came more than six times as often as expected, and four-deviation moves, which a bell curve puts at about one in sixteen thousand, came 39 times. Of those 39, 32 were down moves.
The middle was close to textbook. 5,587 sessions (67.78%) moved less than one deviation, against 68.27% for the curve. The statistic describes ordinary days well and the rare ones badly.
The extreme case. On 27 February 2007 SPY fell 3.91%, from $145.17 to $139.50, after a quiet stretch had shrunk the 20-session deviation; the move measured 10.10 deviations.
What the channel’s width says in deviation units
The 20-session high-low channel was a median 4.25 deviations of its own closes wide, from 3.58 to 5.21 for the middle 80% of sessions. For twenty independent draws from a bell curve, measured the same way, the range averages about 3.9 deviations; prices trend, so their channel runs wider than that.
Why it matters for reading bands. With the whole channel a median 4.25 deviations wide, its edges and a two-deviation band often sit close together, and a close beyond one need not be beyond the other. They are measuring different things, and neither is a ceiling.
The original data
The sample. SPY daily bars from Yahoo Finance at traded prices, starting 29 January 1993, with the count running from 3 January 1994 to 2 October 2026. Moves are close to close, price only. Standard deviation is the population form of the previous 20 moves, not including the session measured. The channel width uses the 20 sessions up to and including the current one, divided by the population deviation of the same 20 closes.
The headline figure. Moves of three deviations or more made up 1.72% of sessions, where a bell curve allows 0.27%.
The file. Every session’s move, prior deviation, move in deviation units, channel and width ratio are in the SPY deviation and channel file.
What it does not show. Any trading result, or that the tails come in any particular season. The deviation estimate itself is noisy over 20 sessions, which accounts for part of the excess; that is the same estimate a 20-session band uses, so the excess is what a band reader faces. The ATR vs standard deviation page compares the deviation of closes with average true range.
When it fails
The failure is fading a three-deviation move as if it were a once-a-year event. The band says the move is extreme, the trader takes the other side, and the size is set for a move that should not extend. On SPY moves of that size came 142 times in about 33 years, several a year, which is not what the word extreme suggests. The arithmetic was right about the past twenty sessions and wrong about the market.
The second failure is treating an old Donchian edge as current. It can be one bar from weeks ago.
A third is reading a narrow band as a forecast. It says recent moves were small, nothing more.
A fourth is stacking both and calling it confirmation. Both come from the same prices.
And a fifth is trusting deviation multiples without counting your own tail. The count takes minutes.
Related
Donchian channels covers the observed high-low range. Standard deviation covers the dispersion statistic. And Bollinger Bands covers the most common way it is drawn on a chart.
Treat a standard-deviation band as a description of the last few weeks, not as a boundary. On SPY the moves past three deviations came more than six times as often as the textbook says, and they are the moves that size a loss.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.