The rule in one paragraph
The upper line is the highest high of the last N bars. The lower line is the lowest low of the last N bars. A middle line, if drawn, is the average of the two. The basic rule is to go with a close or a trade beyond the upper line in the direction of that break, and to exit or reverse when the opposite line is breached. Nothing is averaged, nothing is weighted, and no volume or volatility term appears anywhere.
That gives the method one virtue that is rare and genuinely valuable: there is nothing to tune except N, so there is very little room to fool yourself with parameters. It also gives it a specific character. A rule based on extremes acts only when a new extreme occurs, which means it does nothing at all during a range and then acts on the first move out of one. Whether that is desirable depends entirely on what the market is doing, which is why this rule is usually paired with some judgement about whether a breakout environment exists before the signal is accepted.
Why the level moves in steps on its own
This is the property most people never think about. Because the line is a maximum rather than an average, it is set by a single candle. That candle holds the level for as long as it remains inside the lookback window, and then it drops out. The moment it does, the line jumps down to whatever the next highest high in the window is.
So the channel can move significantly without price doing anything at all. A quiet stretch can see the upper line step down several times, purely as a function of old candles ageing out, which means the distance to a breakout shrinks while nothing happens. Traders who watch the line without understanding this see the level approaching price and read it as compression. Sometimes it is. Often it is bookkeeping. The lower line behaves the same way in reverse. If you want to use the rule, you need to know whether today's level is close because the market has tightened or because last week's spike has simply expired out of the window.
The off by one that flatters every test
Here is the error that quietly improves a lot of published breakout results. If the channel is computed over the last N bars including the current one, then the current bar can never exceed its own channel, because its high is part of the maximum. Implementations that handle this carelessly end up comparing the current price against a window that already contains it, and depending on how the comparison is written you either get no signals at all or signals that use information from the bar you are trading.
The correct construction computes the channel from the N bars before the current one, so the level is fixed before the bar opens. The difference sounds trivial and is not. A test built the wrong way can enter at a price it could not have known in advance, which inflates results in a way that is invisible unless you check the code. This belongs on any honest testing checklist, alongside the related habit of evaluating a signal on a bar that has not closed yet. If you cannot state exactly which bars went into the level you traded against, you do not yet know what you tested.
Entry channel and exit channel
The classical refinement uses two lookbacks: a longer one for entry and a shorter one for exit. Enter on a break of the long channel, leave on a break of the short channel in the opposite direction. The logic is asymmetric on purpose. You want a demanding condition to commit and a less demanding condition to stop committing, because the cost of staying in a finished move is larger than the cost of leaving a continuing one slightly early.
That asymmetry has a consequence for expectations. A rule designed to exit quickly and enter rarely will produce many small adverse outcomes and occasionally a large favourable one, because it keeps paying small amounts for the option to be in the few moves that extend. That is the intended shape of the result distribution, not a defect, and it is the reason this family of rules is psychologically difficult. The relationship between how far you are willing to be wrong and how far you need to be right is the whole question, and it is worth working through the distance arithmetic before deciding whether you can sit with it.
On gold the channel edge is where stops rest
There is a practical collision between this rule and how gold actually trades. The highest high of the last several bars is a visible, obvious level. Other traders see it too, and a lot of protective orders end up placed just beyond it. A move that runs through the line therefore has a reason to run through the line that has nothing to do with trend continuation: the orders sitting there are being filled.
The result is that a Donchian entry is frequently placed into the middle of a stop run, and then price returns inside the range. Nothing about the rule can distinguish that from a genuine expansion, because the rule only looks at whether the extreme was exceeded. Any filter you add is a departure from the rule, which is a real trade off: you gain selectivity and lose the mechanical simplicity that was the main attraction. If you want to see how often it happens, mark the recent extremes on the live chart and watch what price does in the hour after each is taken.
No volatility normalisation at all
The channel has no volatility term, which means its width varies enormously between quiet and active periods for a fixed N. During a compressed stretch the distance between the two lines may be small, and during an active stretch it may be several times that. If your exit references the opposite line, the implied risk of each trade changes with conditions even though your rule has not changed.
Two reasonable responses exist. You can express the channel width in units of average range so that you know, before entering, whether this particular signal involves an unusually wide or narrow structure. Or you can size each position against the actual distance to the exit line rather than using a fixed size. What you should not do is leave the size fixed and the exit floating, because then the rule is quietly taking much larger risk in active markets, which is precisely when gold is most likely to move against a fresh breakout entry. This is the least discussed weakness of pure channel rules and the easiest one to fix.
What it gives you and what it costs
The honest accounting is short. You get a rule with one parameter, no discretion, a clearly defined exit and no capacity to deceive yourself about why you entered. In a market that trends, that package is difficult to improve on, and the absence of tuning knobs is a real advantage when you consider how much damage parameter fitting does elsewhere.
What you pay for it is everything the rule cannot see. It cannot see that the level it is trading is where stops sit. It cannot see that the market has been ranging for weeks. It cannot see that today's break is happening on a release schedule. And because the level steps on its own, it cannot distinguish genuine compression from expiry of an old spike. Used as a complete system it will take every one of those trades. Used as one input, with a judgement about conditions in front of it, the mechanical nature becomes useful rather than naive. Knowing which of those two you are doing is the part that actually determines the outcome.
FAQ
What lookback should I use for a Donchian channel on gold?
Any number you choose is a convention, and the honest answer is that the correct value depends on how long you intend to hold and how often you are willing to be wrong. Longer lookbacks act less often on larger moves, shorter ones act more often on smaller ones. Pick one, write it down, and do not change it after seeing a chart.
Why does my channel level fall when price has not moved?
Because the line is set by a single candle rather than an average. When that candle ages out of the lookback window, the line drops to the next highest high in the window. Nothing has happened in the market. If you are reading a narrowing gap as compression, check whether an old spike has just expired instead.
Should the current bar be included in the channel?
No, not if you want a level you could have known in advance. Compute the channel from the bars before the current one, so it is fixed before the bar opens. Including the current bar lets a test enter at a price derived from information that bar is still producing, which quietly improves results in review.
Why do so many of my breakouts reverse immediately?
Because the recent extreme is a visible level where protective orders accumulate, so price has a reason to trade through it that is unrelated to continuation. The rule cannot distinguish the two cases, since it only checks whether the extreme was exceeded. Adding a condition about what happens after the break costs simplicity and buys selectivity.
Is one channel for entry and a shorter one for exit better?
It is a different trade off rather than an improvement. A shorter exit channel leaves finished moves sooner and also leaves continuing ones sooner, which produces more small adverse outcomes in exchange for keeping less of a reversal. Whether that suits you depends on how you size and how long you can sit with a run of small losses.
ⓘ See these ideas on real price: open the free XAUUSD live chart.