Turtle Trading Strategy: Rules, Breakouts, and What Tests Reveal

We completed 293,088 valid, related historical evaluations of a long-only Turtle channel adaptation across eight fixed samples. Waiting for a confirmed entry usually reduced executions, but its interaction with a volatility exit sometimes made the stricter rule busier. The rules below explain why entry, exit and position state must be read together.

What is the Turtle trading strategy?

The Turtle trading strategy is the systematic trend-following approach taught by Richard Dennis and William Eckhardt. Its best-known signal is a breakout: buy when price exceeds a previous high, then hold while a separate exit rule follows the move.

A channel strategy makes no forecast of where price should go. It asks whether price has done something it has not done within the chosen lookback. The difficult part is accepting repeated false starts while leaving enough room for a sustained move.

The familiar 20/10 and 55/20 pairs mean a longer entry lookback and a shorter exit lookback. They originally refer to daily bars. On a different chart, the same numbers describe a different amount of elapsed time; they do not preserve the original trading horizon.

Original Turtle trading rules

Curtis Faith, one of the original Turtles, published the complete rules. System 1 uses a 20-day breakout and a 10-day opposite-channel exit; System 2 uses 55 and 20. The first system also filters entries using what the previous breakout would have done, including a hypothetical trade not taken, and has a longer failsafe entry. The original system includes volatility-based sizing, a standard 2N protective stop, adding to positions and portfolio limits. Original Turtle Rules

That distinction matters when reading a backtest. A single channel entry is one component of Turtle trading. Our experiments below isolate its long-only channel behavior, using fixed allocation and candle-based signals. They do not reproduce the full original system.

Donchian channels: the current candle must stay outside the calculation

The upper boundary is the highest high of the preceding entry candles. The lower exit boundary is the lowest low of the preceding exit candles. Both exclude the candle being tested.

For example, if the prior entry high is 100, a current high of 101 breaks it. If that candle closes at 99.5, a rule that requires a closing price above 100 has no entry signal. Including the current high in the channel would move the boundary to at least 101 and make a strict breakout comparison impossible.

This is also why breaching with a wick and closing beyond a channel should be treated as separate choices. Confirmation can remove a brief excursion, but it can also delay a real trend. The same decision exists on the way out.

Select the two annotated moments. The first recognizes an upper-channel breach without a confirmed close. The second recognizes a lower-channel breach that recovers before the close. These are historical indicator observations, with no reconstructed trades.

In our research rules, each candle is evaluated after it closes. A wick signal recognizes that its high or low breached the prior boundary; a close signal requires the closing price to finish beyond it. Both submit market orders for the next candle’s open. The wick rule therefore does not claim an intraday fill at the boundary.

Can a stricter breakout rule trade more often?

We ran 293,088 valid, related historical evaluations of the channel rule across eight fixed samples. Entry and exit lookbacks, confirmation choices and an optional N-based exit formed the grid. Every paired comparison holds the other choices fixed and changes one rule.

Usually, requiring the close to finish beyond the entry channel reduced recorded executions: it did so in 146,059/146,544 matched entry comparisons. But 326 comparisons reversed that relationship, and 159 tied. A stricter signal produced more account activity in those reversals.

The interactive comparison shows one such neighborhood. Every line uses the same 32-bar closing-price channel exit and the same historical prices. Dashed lines omit the extra N exit; solid lines add the close-checked 2N rule, which sets a closing-price threshold two frozen volatility units below the actual entry fill. Only entry memory changes along the horizontal axis.

At 85 entry bars, requiring a close produces 47 executions versus 49 for the wick rule when the extra stop is off. With the 2N rule, the same comparison becomes 54 versus 49. Executions count buys and sells, not completed round trips.

The entry filter changes when the position begins, its actual fill and the volatility value frozen for its stop. Those references change when the account becomes flat and which later breakouts it can take. We repeated both 85/32 settings separately and checked all 54 and 49 recorded executions against official candles. In one observed retreat, put the closing price at 100: the close-entry holding’s frozen stop is 101.30, the wick-entry holding’s is 94.06, and their shared channel exit is 88.88. One stop triggers while the other holding stays open; neither channel exit triggers. This supports the mechanism in these two traces; it does not attribute every reversal in the grid to the same event.

All 326 reversals occurred with the 2N rule in one of the eight historical samples; none occurred with the extra stop off or at 4N. They are a small, concentrated exception, not the typical outcome. The figure was selected after observing that region of the frozen grid. Its useful lesson is that eligible entry candles and completed account paths are different things. A stricter signal cannot create more qualifying candles under otherwise identical channel inputs, but it can create more executions after position state and exits interact.

Does waiting for the close reduce drawdown?

Confirmation also does different work at entry and exit. Close-confirmed entry had lower reported maximum drawdown in 108,756/146,544 pairs. Close-confirmed exit had higher reported maximum drawdown in 96,677/146,544 pairs. Before entry, waiting filters exposure. After entry, waiting tolerates a retreat. Both usually reduced executions, but calling both choices “confirmation” hides their different roles.

These are descriptive associations within related historical settings, with exceptions. Return need not move with drawdown: close confirmation gave higher return in 100,875 entry pairs and 87,225 exit pairs, out of 146,544 each.

Turtle stops, false breakouts and the cost of waiting

N averages true range, including gaps from the preceding close. True range is the largest of the high–low range, the distance from high to prior close, and the distance from low to prior close. After its initial average, the familiar 20-bar calculation is N = (19 × prior N + current true range) / 20. A stop expressed in N adjusts its price distance to recent volatility. A wider distance gives the position more room, while a tighter one can remove a trade before the channel exit arrives.

Our optional protective rule freezes ATR over 20 bars at the entry signal, measures the distance below the actual fill, and checks it against subsequent closes. It then requests a next-open exit. This close-checked rule can lose more than its stated N distance; it is not a resting stop order or a guaranteed maximum loss.

A protective exit did not make the account’s drawdown uniformly smaller. Adding the close-checked 2N rule lowered reported maximum drawdown in 63,662 matched comparisons, raised it in 26,123, and left it unchanged in 7,911, out of 97,696. In a separate 20/10, close/close example, it raised reported drawdown from 34.06% to 38.65% while return fell from +7.16% to +4.56%. An exit can cut one trade short and still change later entries, missed recoveries and the path of account losses.

A false breakout can cause a loss even when the calculation is correct. Price may exceed the previous range and reverse, the next open may differ from the signal close, or costs may consume a small move. Adding confirmation changes the path through these events. Whether that tradeoff helped is an empirical question.

Which Turtle settings are best?

Treat 20/10 and 55/20 as familiar reference pairs. An entry lookback controls which highs qualify as new; an exit lookback controls how much retreat a holding tolerates. Shorter exits and stricter entries perform different jobs. A useful comparison changes one choice at a time and checks neighboring settings, activity, return and drawdown together.

The same grid was repeated unchanged. Around the familiar 55/20 pair, the table counts how many of the eight historical checks returned more than zero with close-confirmed entries and exits and no extra N stop. This neighborhood was chosen around the named rule, rather than around the best result. It measures repeatability of positive return; the download also retains worst reported drawdown and minimum activity for each cell.

Of the 36,636 fixed parameter tuples observed in every sample, 5,746 were profitable in all eight. 0 beat a passive control with the same startup in all eight. These counts carry the same settings through each check; they do not pick a new winner for every sample.

A profitable configuration alone does not establish an advantage over holding the same prices, and a larger sweep does not turn related historical evaluations into independent trials.

Compare entry and reentry timing in the Alligator strategy, or see how a trailing level changes execution timing in Parabolic SAR.

Copy the strategy script

Copy this tested long-only channel adaptation into your editor. The defaults use a 20-bar entry, a 10-bar exit, closing-price confirmation, and no extra N stop. Set the study’s combined trading cost to 0.1% per execution in your backtest settings.

//@version=6
strategy("Turtle Channel Mechanism Lab", overlay=true, initial_capital=10000, process_orders_on_close=false, default_qty_type=strategy.percent_of_equity, default_qty_value=99)

entry_length = input.int(20, "Entry channel bars", minval=2, maxval=200)
exit_length = input.int(10, "Exit channel bars", minval=1, maxval=100)
entry_mode = input.int(0, "Entry: 0 close, 1 wick", minval=0, maxval=1)
exit_mode = input.int(0, "Exit: 0 close, 1 wick", minval=0, maxval=1)
stop_n = input.float(0.0, "Close stop N: 0 off", minval=0.0, maxval=10.0)

entry_high = ta.highest(high, entry_length)[1]
exit_low = ta.lowest(low, exit_length)[1]
n = ta.atr(20)
valid = exit_length < entry_length
ready = bar_index >= 500
entry_signal = entry_mode == 0 ? close > entry_high : high > entry_high
exit_signal = exit_mode == 0 ? close < exit_low : low < exit_low
var entry_n = 0.0
stop_level = strategy.position_avg_price - stop_n * entry_n
stop_signal = stop_n > 0 and close < stop_level

plot(entry_high, "Prior entry channel", color=color.blue)
plot(exit_low, "Prior exit channel", color=color.orange)

if ready and valid and strategy.position_size > 0 and (exit_signal or stop_signal)
    strategy.close("Long")

if ready and valid and strategy.position_size == 0 and entry_signal
    entry_n := n
    strategy.entry("Long", strategy.long)