A Donchian channel summarises extremes in a chosen window. A breakout strategy tests whether price exceeds an earlier-known range. Keep the indicator, signal timing and execution rules separate: the published Turtle rules provide one historical system, while a close-based channel exercise is another specified variant.

What the Donchian Channel is

The upper line is the maximum High over N candles, the lower line the minimum Low, and the midpoint (upper + lower)/2. The midpoint is not an average of every close. N counts bars: twenty H1 bars and twenty daily bars cover different windows, and twenty daily bars are not automatically twenty calendar days. Our close-breakout variant excludes the evaluated candle, using High/Low from t−N through t−1. StockCharts’ Price Channels use that convention; check other implementations explicitly.

Including the evaluated candle makes Close above its current highest-High channel impossible: Close ≤ High ≤ that maximum. Using its eventual High to create an order supposedly at its open also introduces future information. Shift the input window in the signal calculation; moving the displayed line alone does not establish that the logic uses earlier data.

Which midpoint are you comparing with price?

Two levels from the same three candles can lie on opposite sides of the current price. The field named in your rule matters.

Scroll sideways to see the full diagram

Which midpoint are you comparing with price?Three fictional completed candles have highs110,112,120; lows90,95,100; and closes108,110,119. The Donchian midpoint of the highest high and lowest low is105. Their closing-price mean is112⅓. A separate fourth-candle Close106 is above the channel midpoint but below the closing-price mean. The price axis is linear.Channel midpoint ≠ closing-price meanOur window: three preceding candles.90105120112⅓1061234Ochre: (120 + 90) / 2 = 105Blue: (108 + 110 + 119) / 3 = 112⅓

Assumptions: our fictional data use arbitrary price units. The window includes exactly three preceding completed candles; both calculations exclude candle4. For candle4 we supply only Close106, not a complete intrabar path.

Channel midpoint=(highest High+lowest Low)/2=(120+90)/2=105. Closing-price SMA=337/3≈112.3333. Substituting one level for the other reverses the answer to Close above midpoint. That establishes no edge for either rule.

TradingView’s official description defines the midpoint as the average of the upper and lower channel boundaries, not the mean of closes. The shifted three-candle window is our example’s assumption.

Close106 is above the Donchian midpoint but below the closing-price mean.
CandleOpenHighLowClose
110011090108
210811295110
3110120100119

How the Turtles turned the channel into a system

The published Original Turtle Trading Rules specify twenty-day System 1 and fifty-five-day System 2 entries on intraday threshold crossings, without waiting for the daily close. System 1 also skips a breakout if the preceding breakout would have been a winner, with a fifty-five-day failsafe; System 2 does not use that skip filter. Opposite-channel exits use ten days for System 1 and twenty for System 2. These are parts of a fuller system with sizing, unit additions and portfolio limits, not the entire system or proof of current profitability.

Historical trend-following rules do not establish that most trades in every present test must lose or that one later trend will recover all losses. Define the market, period, entries, exits and costs, then report the complete sample. Do not infer a performance distribution from the strategy’s name.

How to trade the channel breakout

Our alternative close-based rule gives a long candidate when Close exceeds the earlier upper channel, or a short candidate below the earlier lower channel. This rejects intrabar crossings that reverse before the close, but does not by itself improve net returns. Here is a fictional EUR/USD Bid series with N = 3 and one pip = 0.0001.

Prior-window channel for candle 4 uses candles 1–3 only
CandleHighLowClose
11.102001.098001.10100
21.104001.099001.10200
31.103001.100001.10100
41.106001.101001.10500

Before candle 4, U = 1.10400, L = 1.09800 and midpoint = 1.10100. Its Close is 10 pips above that U. Wrongly using candles 2–4 instead gives U = 1.10600, so the same Close fails the comparison. The signal is known after the close, while a later order follows available prices and execution terms. TradingView’s default strategy emulator processes new orders no earlier than the next available tick; this is a simulation setting, not a guarantee of live execution at Close.

A Donchian breakout is one specified range breakout. It says price exceeded a chosen window, not that it made a record over all history. If the old maximum leaves that window, the next threshold can fall even while the older high remains unbroken.

The greatest weakness — whipsaw in a sideways market

Our separate fictional highs are 110, 100, 102, 103 and 104 in arbitrary price units. With N = 3, the prior U for candle 4 is 110; Close 102 does not break it. For candle 5, the window shifts to 100, 102, 103 and U becomes 103. Close 103.5 exceeds it although both that close and High 104 remain below the older 110. Record which bar set the threshold and when it expired. Requiring a break of the older high too is an extra filter, not a correction to the channel arithmetic.

Repeated failed breakouts can cause losses under a specified trade plan, but touching an edge is not the same as exceeding it or closing beyond it. Define “false breakout”, horizon and exit before counting. A higher interval or trend filter changes the test and does not automatically improve net results. Compare a mean-reversion strategy as a separate hypothesis; the trading-strategies guide provides context without establishing which future regime or method will profit.

A shorter channel as the exit, and sizing your position

A shorter opposite edge can define an exit condition, but its rolling minimum can rise or fall. In our fictional three-value windows, min(1.09800, 1.09900, 1.10000) = 1.09800, while the next min(1.09900, 1.10000, 1.09700) = 1.09700. If a long-side stop must not move away, define new stop = max(previous stop, new candidate). This separate trailing stop rule cannot cancel an earlier triggered exit by substituting a later lower channel. If a bar includes entry and exit prices, its OHLC alone may not establish their sequence.

Original Turtle sizing used N, a smoothed True Range measure related to average true range. The published formula is N = (19 × previous N + current TR)/20, seeded by a twenty-day mean. Units were sized so one N corresponded to 1% of the designated account, while a standard initial two-N stop implied about 2% price risk for one unit before costs and worse fills. The document also describes an alternate half-N stop and unit additions. Do not equate every modern ATR implementation, that historical sizing account, one unit and a whole portfolio’s maximum loss.

A separate risk-budget model starts from a stated entry-to-exit distance and additional costs, rather than assumes the original Turtle allocation is suitable for every reader. A 1% budget is an illustrative input, not a global recommendation or guaranteed loss ceiling. At fixed pip value, larger assumed stop distances require smaller sizes for the same planned amount.

In a fictional EUR/USD USD-account model, one lot is EUR 100,000 and one pip is USD 10 per lot. Equity is USD 10,000 and a chosen budget USD 100. If a defined ATR is 80 pips and the selected stop is two ATR away, the 160-pip distance costs USD 1,600 per lot. Ignoring fees, exact size is 0.0625 lot; rounding down to a 0.01-lot step gives 0.06 lot and USD 96 price risk. Add USD 10 fixed total costs and that size would cost USD 106, above the chosen budget. Reserving those costs first gives 90/1,600 = 0.05625 lot, rounded down to 0.05, with USD 90 total planned loss. Costs are model assumptions, not broker tariffs.

A second fill-based exercise assumes Ask entry 1.10520 and actual stop sale at Bid 1.10020, a 50-pip distance. With the same USD 100 budget and USD 10 fixed total costs, 90/500 = 0.18 lot, or EUR 18,000, gives USD 100 planned total loss. A worse stop fill at 1.09920 gives USD 108 price loss plus USD 10 costs, or USD 118. Spread is already in these assumed fill sides. Neither the channel, ATR nor a stated percentage guarantees execution or the maximum realised loss.

What to do tomorrow

  1. Specify lookback, interval, price basis and prior-window input logic. Check the actual indicator implementation rather than rely on a display offset.
  2. Freeze definitions of intraday crossing, close breakout and failed trade. Record all candidates, costs and outcomes instead of assume a low hit rate from selected charts.
  3. Test a shorter opposite-edge exit, a one-way ratchet and a fixed target as separate rules, retaining earlier triggered exits and intrabar-ordering assumptions.
  4. Reserve costs before size-step rounding and assess frozen rules on later data. A month of demo trading or a 1% input does not establish readiness, profit or a guaranteed loss limit.