Donchian Channel Strategy: The Rule, the Origin, and What the Tests Found

A Donchian channel strategy buys when price closes above the highest high of a fixed number of prior bars and sells when it closes below the lowest low, holding the position until the opposite edge is breached. The channel itself is arithmetic — two running extremes and their midpoint — and the rule attached to it has an unusually long paper trail: the 2004 AgMAS review of technical trading dates the empirical literature from a 1960 study of copper futures by R. D. Donchian.

That record is what makes the question answerable with figures. Two research reports from the University of Illinois AgMAS project trace the family from its origin through a 26-year out-of-sample test on twelve futures markets.

What is a Donchian channel?

A Donchian channel is three lines computed from recent price extremes: an upper band at the highest high of a lookback window, a lower band at the lowest low, and a middle band halfway between them. Nothing is smoothed and nothing is weighted, and Sierra Chart's implementation lists two inputs of its own.

Sierra Chart's documentation for its Donchian Channel study, last revised 24 January 2025, describes exactly that: a high band, a low band and a middle band drawn over a bar count set by a Length input. Its formula takes the upper band at bar t from the highest high of bars t−n through t−1, so the bar being drawn is excluded from its own band, and the study ships with a default displacement of +1 that shifts the whole thing forward a bar. One toggle swaps the highs and lows for closing prices. The middle band is the average of the other two.

Two things follow from how plain that is. A band changes value only when a new extreme enters the window or the old extreme drops out of it, so between those events it holds a constant level. And the study page documents a calculation and those two inputs — the entry rule is something a trader adds on top, and in the research below it comes from the report, not from the software.

What is the Donchian channel breakout rule?

The rule is a closing-price comparison. At the end of each day you compare the close with the highest high and the lowest low of the channel; a close above the top triggers a long, a close below the bottom triggers a short, and in the version tested by Park and Irwin the system is always holding one or the other.

The 2005 AgMAS report calls this the Outside Price Channel system and specifies it tightly. The channel is n days long including today. The highest high is measured over the bars from yesterday back to t−n+1, so today's own high cannot trigger today's signal. If a daily price limit makes trading on the close impossible, the order moves to the next day's open. The report's stated reason for modelling limits at all is that a trend-following rule gives its buy signals in rising markets, so on a limit day the buy actually fills at a worse price than the one that triggered it; crediting the trade with a fill at the locked price would seriously overstate returns.

The 2004 review records two other names in circulation for the price channel: trading range breakout, and support and resistance. The premise it states for the family is the same in each case: a move to a new extreme implies the direction already established carries on. That premise is what the figures below put a number on.

How do the three price-channel systems in the research differ?

The 2005 report tests three price-channel systems, and they are not three settings of one rule. They differ in whether the system is ever allowed to be flat, what the breakout level is measured from, and how a position is closed.

The middle column needs unpacking. M-II does not use the extreme of the whole window; it uses the first day of the channel, taking the larger of that day's high and the preceding day's close as the upper level and the smaller of that day's low and the preceding close as the lower one. The right-hand column is the only one of the three that can stand aside: L-S-O sits a daily-recalculated stop at the midpoint of a cluster of days at the far end of the channel, and that stop can put the system flat before any opposite signal arrives.

What channel length should you use?

Neither AgMAS report recommends one. The 2005 report, which is the one that trades the rule, arrives at a length by re-optimising every year over a grid of candidates. A search for the word "recommend" across that report returns nothing at all, and across the 2004 review it returns four hits, every one of them the phrase "recommended positions" describing the positions a trading rule took in a study the review summarises — not a channel length for a reader.

What the 2005 report does instead is a three-year rolling re-optimisation, the method inherited from the Lukac, Brorsen and Irwin study it set out to confirm. For each market and each system it simulates the previous three years across the whole parameter grid, picks whichever setting earned the highest mean net return, and trades that setting for the next year only. Then it does it again. Nineteen ninety-three's parameters come from 1990 to 1992; at the end of 1993, 1994's come from 1991 to 1993. The first three years of data, 1975 to 1977, are consumed by the first optimisation, which is why the out-of-sample results start in 1978.

The report publishes the chosen lengths in an appendix, one per market per year, and they move around a great deal. Corn's optimal channel runs 30, 20, 35, 30, 30 for 1978 to 1982 and then sits at 65 for three straight years. Cocoa's goes from 60 days in 1979 to 2 days in 1980. If your instinct is that a parameter chosen that way is being fitted to the last three years, that instinct is the subject of our post on curve fitting in trading.

Do channel breakout systems make money in futures?

In the 2005 report's 26-year out-of-sample test the channel system was the weakest of the twelve. Over 1978 to 2003 the Outside Price Channel returned −5.78% a year on the twelve-market portfolio with a Sharpe ratio of −0.22, against −3.14% for all twelve systems weighted equally.

The two halves of that period look nothing alike. For 1978 to 1984 the channel portfolio returned 0.46% a year, Sharpe 0.05, while M-II returned 6.42% and L-S-O 3.09%, and the all-systems aggregate of 4.13% was statistically significant at the 10% level. Park and Irwin write that "All of the 12 trading systems, except the CHL system, show significant returns in more than one market." None of the channel system's twelve individual market results carried a significance star; its best was world sugar at 20.80% and its worst pork bellies at −19.73%.

For 1985 to 2003 the channel portfolio returned −8.08% a year — again the lowest of the twelve — and no system in the study was positive on the twelve-market portfolio. A time-trend regression over the full period puts the channel system's decline at 0.35 percentage points a year (t = −2.00), and the trend is negative and significant at the 10% level for all twelve systems.

Every return above carries a flat $100 per contract per round trip for the whole period, commission plus bid-ask spread. The report also re-simulates all twelve systems under a second cost scenario, in which lower brokerage commissions after the original study's sample period bring the round-trip total to $75 for 1985 to 1994 and $50 for 1995 to 2003; the channel system's full-period figure improves to −3.56%, still the lowest of the twelve and still negative. The 2004 review's summary of futures as a whole is that technical trading appeared profitable from the mid-1970s to the mid-1980s, and that as of its August 2004 survey cut-off no study had comprehensively documented profitability after that. Our post on the moving average crossover strategy works through the same source pair for the other trend rule those reports treat as widely used.

Why do two simulations of the channel system report different returns?

Because the trading model is as much of a choice as the parameter. The 2005 report rebuilt the earlier Lukac, Brorsen and Irwin test and put the two side by side for 1978 to 1984, and the twelve-market portfolio net return reported for the channel system went from 33.4% in the original to 3.6% in the re-simulation, over the same markets and the same years.

Those are returns on what the original called the total investment method: a base of which 30% is posted as margin and 70% held back against margin calls, so each figure is about a third of what the same profit would read as a return on margin alone. On the gross line the gap is 65.4% against 18.5%. The difference between them is transaction costs of 32.0 percentage points against 14.9, plus a different simulation of the rule — the report's own footnote records that the original ran a different version of the channel system and mis-specified the companion system's parameters, which is why its channel figures are restated using the 2005 study's parameters.

The honest reading of that table is not that one study was wrong. Park and Irwin treat the original's conclusions as confirmed and say they find more evidence of profit than the original did. What moves by a factor of nine is a single system's headline number, under modelling choices nobody would describe as exotic: which variant of the rule got coded, and which prices were used to turn a dollar cost per round trip into a percentage of contract value. Any backtested figure you are shown for a channel system is carrying those same two assumptions whether or not they are stated.

Is there any evidence on stock index futures?

Yes, through the 2004 review rather than the twelve-market test, and it does not amount to much once the size of the search behind it is accounted for. The review summarises a data-snooping study that applied roughly 8,000 trading rules, drawn from five systems including channel breakouts, to the Dow Jones Industrial Average over 1897-1996 and to S&P 500 index futures over 1984-1996.

On the S&P 500 futures the best of those 8,000 rules returned 9.4% a year at a nominal p-value of 0.04, which on its own looks significant. Adjusted for the fact that it was the best of 8,000, the p-value was 0.90. The review does not say which of the five systems produced that rule, so this is evidence about the rule universe rather than about channel breakouts specifically — and the study did not build transaction costs in directly.

The twelve markets in the 2005 test contain no stock index contract at all, which matters if index futures are what you trade. They are grains, softs, lumber, meats, metals, two currencies and T-bills, on daily data. A channel result from world sugar in 1981 is evidence about world sugar in 1981.

What does a daily channel breakout system mean on a funded account?

It means the system as tested cannot be run. All three price-channel systems in the 2005 report enter at the day's closing price, so a position cannot be closed before the following day at the earliest, and a Lumen account — a simulated account with a real payout on the profits — has to be flat by 4:45 PM ET, with no overnight or weekend positions at all. Every trading day starts flat — see trading hours for the exact times and the shortened-session rule.

What survives the translation is the shape of the rule: a breakout entry and an exit at the opposite edge of the channel. And on a funded account that opposite edge is where the arithmetic bites. A losing trade in a reversing channel system spans at least the full width of the channel, because the entry is a close beyond one edge and the exit a close beyond the other. On a $50K account the drawdown amount is $2,000, per the drawdown rules. At $50 a point on the E-mini S&P 500 — tick values are listed on our instruments page — $2,000 is 40 index points of adverse movement on one contract, or 400 points if you are trading the micro at $5 a point.

So the question for anyone adapting this is whether the channel's two edges sit closer together than 40 ES points, because beyond that a single reverse at mini size exhausts the line on its own. The drawdown line also only rises on a new highest closing balance, so an intraday breakout that gives the profit back before 4:45 PM ET buys no extra room. Sizing the entry so the far edge of the channel is survivable is the whole problem, and futures position sizing works through how few stops actually fit inside a drawdown.

FAQ

Is the Donchian channel the same thing as the price channel systems in the research?

Not quite, and the 2005 AgMAS report is careful about it: the Outside Price Channel system it tests is described as analogous to a system Donchian introduced, not as a reproduction of it. The phrase "Donchian channel" does not appear in either AgMAS report. The name belongs to the convention and to platform implementations like Sierra Chart's study; the exact rule that was tested belongs to the report that specified it.

What channel length did Donchian himself use?

According to the 2005 report, Donchian's own system used the ranges of the two preceding calendar weeks as its channel. The 2004 review's table of early studies records what he published: a channel rule on daily copper futures over 1959-60, costed at $51.50 per round trip, with net gains of $3,488 and $1,390 against $1,000 of margin on single contracts of the December 1959 and December 1960 deliveries. The same row records no benchmark strategy and no optimisation.

Does a Donchian channel predict where price is going?

Sierra Chart's page for the study does not say so. It sets out the calculation and the study's two inputs, and carries no entry rule and no forecasting claim. The prediction is in the rule a trader attaches to it, and the 2004 review states that premise plainly: a move to a new extreme is taken to imply the established direction continues. Whether that held in futures is what the figures above address.

Can you trade a channel breakout system on a prop firm evaluation?

You can trade breakout entries, but not the daily, position-holding version the research tested, because an account that must be flat by 4:45 PM ET cannot hold a signal overnight. Adapting it means moving the channel to an intraday window and sizing the position so the opposite edge sits inside the drawdown line, which is the constraint that decides whether it is possible at all.