Donchian Channel: The Turtle Traders' Breakout System

Most indicators are arithmetic performed on price. The Donchian Channel is not. Its upper line is the highest high of the last N bars and its lower line is the lowest low of the same window, which means you could draw the thing with a ruler and no calculator. That sounds like a downgrade until you notice what it buys: a breakout level that is completely unambiguous, identical on every screen, and impossible to argue with. It is the reason a group of novices in 1983 could be handed a rule sheet and trade tens of millions of dollars of someone else's money without ever forming an opinion about a chart. The channel is also the most misunderstood of the three band constructions in common use, and the misunderstanding is nearly always the same one: people assume it is a volatility band like the other two. It is not. There is no volatility term in it anywhere.

Quick Answer

What Is a Donchian Channel?

It is an overlay drawn on the price chart, and its two outer lines are not really calculations. The upper band is a lookup: of the last N bars, which one had the highest high, and what was it. The lower band is the same lookup in the other direction. The middle band is the halfway point between those two numbers, included for convenience rather than because anything is averaged.

That construction answers a different question from the one every other band on your chart answers. Bollinger Bands and Keltner Channels both measure distance: how far is price from its own average, expressed in units of volatility. A Donchian Channel measures rank: is this the highest price of the last N bars, yes or no. Distance and rank are not the same kind of fact, and three consequences fall out of the difference.

The bands are a step function. They sit perfectly flat for as long as no new extreme prints, then jump vertically the instant one does. There is no gradual drift, because there is nothing being averaged that could drift. Plot a Donchian Channel next to a Keltner Channel on the same chart and the visual difference is immediate: one is a staircase and the other is a curve.

A breakout is not a signal derived from the indicator. It is the indicator. When price touches the upper band it has, tautologically, just printed a new N-bar high. There is no lag between the event and the reading, no threshold to cross, no confirmation the formula is waiting on. That is a genuine advantage over any smoothed construction, and it is also why the level is so heavily watched: it is the visible high of the window, so every trader running the same look-back is staring at the identical number. The band behaves like a horizontal level you would have drawn anyway, just picked mechanically instead of by eye.

The band can move against price without price moving. This one catches everybody. Because the look-back is a rolling window, the bar holding the upper band up eventually falls off the back of it. When it does, the band drops to whatever the highest high of the remaining bars is, on a bar where price did nothing at all. The channel is a rolling statistic, not a ratchet, and the drop is a property of the formula rather than a glitch.

One precision point worth making, because it is a real tension in how this indicator gets described. The formula contains no volatility term, but the observed width of the channel still tracks how much the market has ranged: a market that has covered a lot of ground will have widely separated extremes. So the width is an informal volatility readout even though nothing volatility-related was computed. A Keltner Channel width is a measured volatility estimate. A Donchian Channel width is a side effect.

What Is the Donchian Channel Formula?

Run it on numbers. Over the last 20 bars, the single highest high your stock printed was 62.40 and the single lowest low was 57.80. The upper band is 62.40. The lower band is 57.80. The middle band is 60.10. The channel is 4.60 wide. That is the entire computation and it took one pass over twenty bars.

Now watch what happens over the next handful of bars. Price pushes to 62.55 intrabar. The upper band is 62.55 immediately, on that bar, with no smoothing lag whatsoever. Compare that to a 20-period average, which would absorb the new high as one twentieth of its value and shrug. Then suppose price stalls and chops for another twenty bars without exceeding 62.55. The 62.55 bar rolls out of the window, and if the highest high left inside it is 61.20, the upper band falls to 61.20. Price never sold off. The window just moved.

On look-back conventions: 20 periods is the classic default and what your platform will ship. Fifty-five periods is the standard slower trend-following variant, and 10 periods shows up constantly as an exit channel rather than an entry one. Those three numbers are not arbitrary and they are not modern. They come straight out of the rule set covered further down. The Wikipedia entry on the Donchian channel gives the definition in its shortest form, and the TradingView support page for the indicator documents what one major platform exposes as settings, which is usefully short compared to most overlays.

Richard Donchian Did Not Build the Donchian Channel

The indicator is named after Richard Donchian, the commodities trader widely credited as a father of systematic trend following. That attribution is correct and it is on every page written about the tool.

The part almost nobody carries is that the system Donchian himself actually published was a different mechanism. His signature method, distributed from 1960 onwards in his Hayden Stone newsletter Commodity Trend Timing and generally referred to in the literature as his weekly rule, was built on a 5-day and 20-day moving average relationship. Moving averages. Not highest highs and lowest lows. The channel that carries his name uses no moving average at all, which makes it a different construction from the thing he was known for.

This matters beyond trivia for one practical reason. When you read that "Donchian's system" produced some result, check which system is being described, because the two behave differently and are frequently conflated in the same sentence. A moving-average crossover and a breakout of an N-bar extreme respond to the same chart in visibly different ways and have different failure modes. The Wikipedia biography of Richard Donchian is the fastest place to see his published work laid out next to the indicator attribution.

How Did the Turtle Traders Use Donchian Channels?

This is the reason anyone still runs this indicator, and the story is worth the space.

In 1983 the commodities trader Richard Dennis got into an argument with his partner William Eckhardt about whether successful trading could be taught. Dennis said yes, Eckhardt said it was innate. They settled it the way people with money settle things: Dennis advertised, interviewed, selected a group of complete novices, trained them for two weeks in a fully mechanical trend-following system, and funded them with real capital. He called them the Turtles. The experiment is commonly cited as having produced somewhere around $175 million in profits across roughly four to five years, and Eckhardt publicly conceded the bet.

The entry rule that group was handed was a Donchian breakout, and there were two versions of it running side by side. System 1 entered on a break of the 20-day extreme, with a filter that skipped the signal if the previous breakout in that market would have been profitable, plus a 55-day failsafe so a skipped breakout that kept running was not missed entirely. System 2 entered on a break of the 55-day extreme and took every one without a filter. Exits used a second, shorter channel in the opposite direction: System 1 got out on a 10-day opposite extreme, System 2 on a 20-day opposite extreme. There were no profit targets anywhere in the rule set.

Look at how much of that table is not the entry. The breakout is two rows. Everything else is risk plumbing, and the plumbing is where the system actually lived.

The three rules everyone skips

Sizing was normalised to volatility. The Turtles worked in a unit they called N, which was essentially a 20-day average true range for the market in question. A unit was account risk divided by N, so a violent market got a small position and a quiet one got a large position, and the dollar risk per unit came out roughly the same either way. That single idea is doing enormous work: it means a fixed rule set can be applied to wildly different instruments without being retuned, because the position size absorbs the difference. The same logic underneath sizing a trade off the distance to your stop rather than off a fixed share count is this rule in modern clothes.

Every unit carried a hard stop at 2N from entry. Twice the volatility unit, no discretion, no moving it. Sized against the unit rule, that came out to roughly 1 to 2 percent of account equity at risk per unit. Note that the stop is not the opposite band. The exit channel takes you out of a trade that has stopped working; the 2N stop takes you out of one that went straight against you from the entry, which the exit channel would be far too slow to catch.

Winners were pyramided, losers never were. A working position added units every 0.5N of favourable movement, up to a maximum of four units in a single market, and each time a unit went on the stops for the earlier units were raised. That is the mechanism that turns a 35 percent win rate into a profitable system: the trades that are working get progressively more capital while the trades that are not stay at one unit and get closed. The original Turtle rule documentation lays the full sequence out, including the correlation and portfolio limits I have skipped here.

One honest framing point before anyone ports this to a 5-minute chart. The Turtle system traded daily bars on futures across a basket of markets, with enough capital to hold four units in several of them simultaneously. It was a position-trading system, never a day-trading one. Shortening the look-back and running it intraday is a legitimate thing to do, but it is a different strategy that borrows an entry rule, and it does not inherit the historical record.

How Is a Donchian Channel Different From Bollinger Bands and Keltner Channels?

There is one structural difference and it is larger than the difference between the other two. Bollinger Bands and Keltner Channels are both a moving average plus and minus a width, and they disagree only about how to measure the width: Bollinger uses the standard deviation of closes, Keltner uses average true range. Two definitions of volatility, same skeleton.

A Donchian Channel has no skeleton in common with either. No moving average, no width multiplier, no volatility estimate, no statistical quantity of any kind. Two raw prices lifted off the chart and a midpoint. That is why the three belong in the same section of a chart menu and in different sections of your head. If you have read how an ATR-set band behaves against a standard-deviation-set one, the mental model there is a family argument between siblings. Donchian is not in that family.

The row to internalise is the third one. Everything else in that table is a consequence of whether the construction smooths anything, and only one of the three does not. Practically it means the three tools answer three different questions, and you pick by the question rather than by which chart looks nicer. Is price at a new extreme? Donchian. How far has price stretched from its average in units of typical bar range? Keltner. Have the closes compressed unusually tightly around their mean? Bollinger, which is the reading behind the volatility-compression setup that nests two bands inside each other. A Donchian Channel cannot produce that setup even in principle, because it has no notion of compression to detect.

The regime split follows from the same place. A construction built around an extreme suits trend and breakout conditions. A construction built around a mean suits range-bound conditions. Putting all three on one chart is not confluence in any meaningful sense either, a distinction worth reading up on in the survey of which indicators genuinely measure different things, because three bands agreeing that price is high is one observation reported three times.

What Settings Should You Use for Day Trading and Swing Trading?

There is exactly one setting, which makes this the shortest tuning conversation of any overlay. Shorten the look-back and new extremes print more often, so you get more entries, earlier, with a larger share of them being noise. Lengthen it and you get the reverse. That is the whole trade-off and no value escapes it.

For intraday work the common adjustment is down to roughly 10 to 20 periods, which is what gives you signals at a frequency that matches a session. For swing work a medium 20 to 30 period look-back, or the classic 20 and 55 pairing, is the usual framing. Neither range is a tested recommendation, they are the conventions the literature repeats.

The thing the table cannot show is that shortening the look-back does more than add signals. It changes what the word breakout means. A 20-bar high on a daily chart is roughly a one-month high, and taking one out is a real event that other participants noticed. A 20-bar high on a 1-minute chart is a twenty-minute high, and taking one out happens several times an hour on a liquid name for no reason at all. Same rule, same indicator, two completely different claims about the market. This is why an intraday momentum playbook leans on relative volume and time of day rather than on the breakout level alone: the level stops carrying information once it is cheap enough to print.

Why Do So Many Donchian Breakouts Fail?

Because markets spend a large share of their time going nowhere, and a rule that fires on every new N-bar extreme will fire repeatedly inside that nowhere. In a sideways range, price grinding to the top of the box prints a new 20-bar high, triggers the entry, then rolls back through the middle of the channel and prints a new 20-bar low a few days later, triggering the other side. Nothing has gone wrong with the indicator. It correctly reported two new extremes in a market with no trend.

The numbers are not flattering and should not be hidden. Published tests of Donchian breakout systems commonly report win rates in the 30 to 40 percent range, and the backtest cited on the indicator's own Wikipedia entry puts it at 35 percent. That means most individual trades lose. What keeps such a system net profitable is the shape of the distribution rather than the hit rate: losers are cut at a fixed volatility-based stop and winners are held with no profit target until an opposite extreme prints, so the average winner is much larger than the average loser. If that arithmetic is unfamiliar, the breakeven win rate for a given reward-to-risk ratio is the calculation that makes a 35 percent system make sense on paper. Sitting through the losing streaks it implies is a separate problem and a harder one.

Four filters show up repeatedly in the tactical literature, and all four work by refusing some breakouts rather than by finding better ones.

A long-term trend filter. Only take upside breakouts while price sits above a slow moving average, commonly something like a 100-period one, and only take downside breakouts below it. This is the cheapest way to stop buying new highs inside a downtrend.

A momentum filter. Pairing the breakout with a momentum reading such as RSI is the standard suggestion, on the logic that a new extreme printed with no momentum behind it is a candidate to be given straight back.

Volume on the breakout bar. An expansion bar with no participation behind it is the classic false break. This is the same discipline that separates a break that holds from one that gets retested and rejected, and it applies unchanged here.

An ATR buffer on entry. Rather than entering the instant the line is touched, wait for price to clear it by an extra half to one ATR. You give up some of the move in exchange for skipping the pokes that never got anywhere.

Exits and stops, where the line is the wrong place to stand

Two conventions are worth adopting. The middle line makes a serviceable trailing-stop or partial-exit trigger: it sits at a mechanical halfway point between the extremes and a close back through it says the push has given up more than half of the range it was working in. And the protective stop belongs a buffer inside the opposite band, commonly around one ATR, rather than exactly on it. The reasoning is the same reasoning that makes the upper band useful in the first place. The line is the literal visible extreme of the window, so everyone can see it, which makes it an obvious place to hunt for stops. Sitting exactly on the most obvious price on the chart is not where you want your risk defined.

One further limitation worth naming rather than burying. Some published tests suggest breakout systems of this kind hold up better on trending markets such as commodities and currencies than on individual equities and equity indices, where they can underperform simply holding the index. That is a thin finding and worth treating as a hypothesis rather than a fact, but it is consistent with the mechanism: a system that needs sustained one-directional moves will struggle on an instrument that mean-reverts around an upward drift.

Where the Donchian Channel Fits on a Real Chart

It is a level generator, not a decision. What it contributes is a completely objective, reproducible price that nobody can argue with, which is exactly what a rule-based process needs and exactly what discretionary chart reading tends to lack. What it contributes nothing to is the question of whether the chart printing that level is one you should be trading at all.

The order that works is regime, then level, then trigger, then size. Establish whether the market is trending or rotating, which the channel cannot tell you, because a new extreme prints in both conditions. Let the channel hand you the level. Decide what has to be true on the breakout bar for you to act, which is where volume and momentum come in. Then size the position off the distance to your invalidation rather than off how confident you feel, which is the one Turtle rule that ports cleanly to any market and any timeframe. The broader case for treating an overlay as a layer on top of structure rather than a replacement for it is the spine of the technical analysis overview.

What a chart grader can and cannot see here

Worth being straight about, since this site sells a tool. SnapPChart has no Donchian Channel field. It does not scan back twenty bars and find the highest high, it does not compute a lowest low, it does not construct a channel, and it carries no dedicated state for one the way it does for the moving average stack, the VWAP relationship, the MACD cross and volume behaviour. It also has no notion of N, no unit sizing model and no concept of a 55-day failsafe. What it does is read a chart screenshot you upload, so if you plot a Donchian Channel on your own platform before taking that screenshot, those lines 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 your look-back is 20, it cannot tell a 20-period channel from a 55-period one, and it cannot know whether the band you are looking at just stepped down because an old high aged out of the window. Which indicator states a screenshot-based read genuinely 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 20-bar high is the reason you are taking the trade, verifying that 20-bar high stays your job.