Donchian Channel: Formula, Turtle Rules, and a Breakout Test on Five Markets

The Donchian channel draws the highest high and the lowest low of the last 20 bars around price, with a middle line halfway between. When price trades above the upper line it has made a new 20-day high; below the lower line, a new 20-day low. That simple rule was the entry signal of the Turtle traders, and it is still the core of many trend-following systems. Below: the formula, the Turtle rules as published by former Turtles, every breakout on a recent Bitcoin chart with what followed, and a test of three channel lengths on five markets over 9 to 33 years.

How the Donchian channel is calculated

StockCharts publishes the indicator as Price Channels and notes that it was developed by Richard Donchian. The upper line is the 20-day high, the lower line the 20-day low and the centerline their average. StockCharts builds the channel from the bars before the current one: a 20-day channel for October 21 uses the 20 days that end on October 20. That is what makes a break possible. If the current bar were included, its own high would set the upper line, and price could touch the line but never close above it.

The lines only move when an old extreme drops out of the window or a new one is set, which is why they look like steps. The channel width, upper minus lower, is a plain measure of how far price has ranged over the period: it widens in volatile markets and narrows in quiet ones. Unlike Bollinger Bands or the Keltner channel, there is no average or volatility formula behind it, only the extremes.

How traders read the channel

StockCharts adds that the channel can also be used to spot overbought and oversold levels inside a larger trend, for example a pullback to the lower line in an uptrend. That is a different use from the breakout, and the two give opposite instructions at the same line, so pick one before the trade.

The same channel on the S&P 500 ETF, over its last 120 sessions from May 8 to October 28, 2026, shows a market that rose (from 733.90 to 772.31 on our adjusted closes): it closed above the upper line on 14 days and below the lower line on only 3, on July 29, September 10 and September 16.

The Turtle rules

The best-known use of the channel is the system taught to the Turtle traders by Richard Dennis and William Eckhardt. The rules published in 2003 by former Turtles describe two entries, each based on Donchian's channel breakout:

• System 1: enter on a 20-day breakout, a price above the high or below the low of the preceding 20 days. Exit longs on a 10-day low and shorts on a 10-day high. A System 1 breakout was skipped if the previous breakout would have been a winner, but a 55-day breakout was always taken so that a big trend was not missed.
• System 2: enter on a 55-day breakout, exit on a 20-day breakout against the position.
• Entries during the day: the Turtles entered when the level was exceeded during the day, not at the close or the next open; when the market opened through the level, they entered at the open.
• Position size and stops: N, a 20-day average of the true range, set the size of each unit so that a 1 N move was 1% of account equity, and stops were placed 2 N from the entry, so a stopped-out unit lost about 2%. N is an average true range by another name; see also position sizing.

The channel is the entry; most of the system is in the exits, the stops and the sizing. A plain channel breakout without them is not the Turtle system, which is why the test below calls itself a breakout test.

The channel on a real chart

The 120-day Bitcoin chart above shows the daily Donchian(20) from July 1 to October 28, 2026, using the 20 bars before each day. The marks show the first close outside the channel; a close that stays outside on the following days does not get a new mark. There were five, and this is everything that followed each one, with the classic exit, the first close below the low of the previous 10 days after an upside break, or above the high of the previous 10 days after a downside break.

Read as a single long-or-short position on the closes, starting flat on July 1 (the last short, opened on the downside close of May 22, had been closed on June 14), the window held three trades. The long from July 14 went nowhere for a month, then rode the August rally and was closed on September 15 for +16.30%. The same close was a new 20-day low; the short taken there lost 6.93% in three days as price jumped, and the upside break on September 21 lost 5.62% before its exit on October 8. No signal was open on October 28. Two of the three trades lost, and the one winner paid for both: that is the usual shape of a breakout system, a minority of trades carrying the result. Bitcoin itself rose from 60,024 to 81,343.50 over the window, +35.52%, against about +2.16% for the three trades compounded.

A breakout test on five markets

We ran the bare channel breakout on five daily series, long only and then short only, with three entry and exit pairs: 10/5, 20/10 (Turtle System 1 without the skip rule) and 55/20 (System 2). Entries and exits are taken during the day as in the Turtle rules: at the channel level, or at the open when the market gaps through it. No stops, no filters, no position sizing (the whole account in each trade), no costs and no interest on cash. Every system and the buy-and-hold line start on the same day, the 56th bar of each series, so that the 55-day channel exists; equity is marked to market each day, and a position still open on October 28, 2026 is valued at that close. A trade cannot be closed on the bar it was opened, and after an exit a new entry is possible only from the next bar. A short's return is (entry − exit) / entry, and returns are compounded trade by trade; an open short is valued at 1 + (entry − close) / entry.

Trades and winners count closed trades. Positions open on October 28, 2026, valued in the totals at that close: the 20/10 long on the Treasury ETF since October 22 (entry 81.00, −0.01%), the 20/10 long on the commodity ETF since October 16 (33.52, +0.81%) and the 55/20 long on it since August 17 (30.57, +10.53%); among the 20/10 shorts, the S&P 500 ETF since October 23 (758.79, −1.78%), gold since September 10 (396.45, +5.79%) and Bitcoin since October 24 (79,800, −1.93%). ETF prices are adjusted for dividends, which favors buy and hold, since the system holds the ETF only part of the time. Three things stand out:

• Smaller drawdowns everywhere. The long breakout was in the market 38% to 51% of the days with the 20/10 pair, and its worst drawdown was smaller than buy and hold's on all five markets, at both lengths.
• Beating buy and hold depends on the market. Both lengths beat buy and hold on the commodity ETF and on Bitcoin, the two markets with the deepest buy-and-hold drawdowns. On the S&P 500 ETF and gold they made far less, and on long Treasuries somewhat less (+100.0% and +52.4% against +114.5%). On the S&P 500 ETF the 20/10 system spent 12.1 years under its previous equity high, from January 8, 1999 to February 7, 2011.
• Short breakouts mostly lost. With the 20/10 pair, shorting new lows lost 78.3% on the S&P 500 ETF, 56.2% on gold and 29.7% on Treasuries, and gained 16.1% on the commodity ETF and 41.0% on Bitcoin. The shorter 10/5 pair did worse on the stock ETF (−87.3%). Markets that rise over time punish a system that sells every new low.

The faster 10/5 pair traded about twice as often as 20/10 and its long side ended at +146.1% on the S&P 500 ETF, +463.9% on gold, +36.7% on Treasuries, +97.6% on the commodity ETF and +1,650.2% on Bitcoin, from the same start dates. With costs, the extra trades would cut into that.

Does it work?

The test above is one family of rules, on five markets, before costs. It shows what a bare channel breakout does: it cuts drawdowns by sitting out long declines, and it wins or loses against buy and hold depending on whether the market trends hard in both directions. It is not a test of the full Turtle system, which traded a diversified futures portfolio with sizing, stops and the skip rule. Our bench measures chart patterns detected by rule against random entries, not indicators (see our methodology), and the channel pattern page has those statistics for a drawn channel, a different object from this indicator.

For other channel tools, see the Keltner channel, Bollinger Bands and the STARC Bands, or the full list of volatility indicators.