Turtle Traders — The Donchian Breakout and N-Unit Sizing System
Turtle Traders (Richard Dennis · William Eckhardt) · The original 1983–1988 experiment (U.S. futures, currency, and bond markets) / the methodology was released free of charge in 2003 by Curtis Faith with the founders' permission and has circulated ever since.
Published 2026.07.08
In 1983, Richard Dennis ran an experiment to see whether dozens of novices, handed nothing but rules with all discretion and intuition removed, could beat the market. The participants called themselves "Turtles," and what they were left with was not a prediction that nailed the timing of a buy but a single complete system: ride the breakout, size by volatility, and cut mechanically when wrong. This page breaks down that system's structure — breakout entries, volatility-based sizing, mechanical stops — for educational purposes. It is not a buy or sell instruction, and the parameters that appear are examples of values fitted to that era and those markets.
- It is a mechanical system that enters after the fact on the 'event' of a 20- or 55-day Donchian channel breakout, not on prediction.
- Volatility N (the 20-day ATR) fixes every position's loss risk at roughly 1% of the account.
- With a 2N hard stop and an exit on an opposite-direction channel breakout, no discretion enters the stop.
- Low-correlation basket diversification is a premise — going all-in on a single coin or using high leverage shatters the system's skeleton.
① Identification — Regime & Conditions
The Turtle system is not a way to dig deep into a single instrument. The starting point is tracking a basket of several low-correlation futures markets at once, and diversification is not optional but a premise for the system to work at all. Since you cannot know in advance which market will produce a big trend, you cast lines in many places and ride only the trend that bites, all the way to the end.
N is the exponential moving average (ATR) of the last 20 days' True Range. From there, unit size, stop distance, and pyramiding intervals are all converted into this N unit. The more volatile the asset, the automatically smaller the same one unit becomes, so that a single loss risk is matched to a similar level even across different markets.
- Build the basket — pick low-correlation markets (a few low-correlation pairs if trading crypto) and observe them simultaneously.
- Measure N — compute each market's 20-day ATR to set that market's own volatility unit.
- Fix the channel — draw the upper and lower boundaries from the 20-day and 55-day highs and lows.
- Mark the regime — first judge the 'location': whether price is ranging inside the boundaries or breaking across them. Location comes before judgment.
② Entry Frame (Observation & Lag)
Entry comes not from prediction but from the lagging event of a breakout. System 1 goes long when the close crosses above the 20-day channel's upper band and short when it breaks below the lower band. The slower System 2 takes the 55-day channel breakout as its signal, serving as a safety net that catches, from behind, the big trends the shorter system missed.
System 1 has a peculiar condition. If the previous 20-day signal ended in a profit, skip the next signal. It is a device to avoid being caught repeatedly by the false breakouts that cluster right after a big trend. If skipping causes you to miss a genuinely big trend, the 55-day System 2 catches that move instead as a safety net.
Do not predict. Ride a trend that has already begun, and cut immediately when wrong — that is all the signal is.
③ Risk & Sizing
The real engine of this system is not entry but sizing. One unit is sized so that 'a 1N adverse move loses roughly 1% of the account.' The number of units is (account × 1%) divided by (N × the per-contract unit value). Whichever market you enter, one unit's loss risk is matched to the same 1%, so the key is not the hit rate but making the loss size of every trade a constant.
- Compute the unit — back-calculate the number of contracts so that a 1N adverse move equals roughly 1% of the account.
- Pyramid — add one unit for every 0.5N the price moves in your favor, up to a maximum of 4 units, raising the stop along with each add.
- Exposure caps — hard-cap at 4 units per single market, 6 units per closely correlated group, 10 units per loosely correlated group, and 12 units total in one direction, structurally blocking all-in bets.
- Drawdown reduction — cut unit size by 20% for every 10% the account falls from its peak, restoring full size only after a new high is recovered.
Exit without exception on a 2N adverse move from the entry price. The system allows no discretion to widen or defer the stop. Trail the stop only in the favorable direction, never widening it against you.
④ Invalidation & Limits
The invalidation coordinate is already set before entry. The stop is fixed 2N on the opposite side of the entry price, and the trend-exit signal is a 10-day breakout in the opposite direction for System 1 and a 20-day breakout in the opposite direction for System 2. When this point is reached, you exit without attaching a reason — averaging down is not a strategy but a refusal to make the judgment.
① Lag — because you enter after the breakout and exit after the opposite breakout, you always give back the beginning and the tail of a trend. ② Whipsaw — in a sideways market, consecutive false breakouts pile up small losses. ③ Low hit rate — the structure is many trades with small losses and a few with large gains, which is psychologically hard to endure. ④ The founder's downfall — even Richard Dennis was widely reported to have halted money management after large losses in 1987–88. ⑤ Edge dilution and parameters — after the rules were made public, the effectiveness of a simple breakout weakened, and numbers like 20, 55, and 2N are merely examples of that era's and that market's values, not a guarantee of validity everywhere.
In short, this system is not a tool for 'being right' but a reward-to-risk and money-management frame that makes losses small when wrong and gains large when right. Many trades ending in losses is not a flaw but the design, and if you cannot endure that stagnation with discipline, the system itself collapses.
⑤ Porting to Crypto (Perpetuals)
Crypto perpetuals trade continuously, 24 hours with no weekends, so the 'daily close' node blurs. You typically redefine the breakout against the daily or 4-hour candle close. The bigger problem is diversification — the original system's power came from some twenty low-correlation markets, but in a bull run most coins move together, so even a basket has weak diversification.
The original 2N stop presumes ample margin. At high leverage, the liquidation price can come before the 2N stop price, ending the account before the rule can even execute. You must first back-calculate the leverage and units so that the 2N stop sits comfortably inside the liquidation price.
The principle of 'riding a trend to the very end' means paying long-term funding fees in perpetuals, so you must reflect funding fees and open-interest costs in your expectancy. And because crypto's frequent false breakouts make a simple 20-day breakout produce more false signals, variations that add a regime filter — such as the 'skip the last winning signal' filter or a long-term moving-average direction — are common. That said, no variation is a profit guarantee, only an attempt to reduce small losses.
Take the distance from the entry price to the 2N stop as 1R. If one unit was sized so that "1N adverse = roughly 1% of the account," then the amount lost when the 2N stop is hit is about 2% of the account — that is, 1R is roughly 2% of the account. The Turtle structure absorbs this 1R loss many times over while riding a few trends in a big way. For example, if ten trades end as −1R, −1R, −1R, −1R, −1R, −1R, +3R, +3R, +3R, +12R, the losses total −6R, the gains total +21R, and the net is +15R. It is a reward-to-risk structure where a few large gains cover the many losses. These numbers are an example of computing position and size, not a profit guarantee, and the key is that the position was back-calculated from the stop distance (2N), not from the hit rate.
- Did I spread risk across a small low-correlation basket rather than going all-in on a single coin?
- Before entering, did I measure N (volatility) and write down the 2N stop coordinate and unit size first?
- Is my entry rationale a lagging event — a 20- or 55-day channel breakout — rather than a 'prediction'?
- If using high leverage, did I calculate whether the 2N stop sits inside the liquidation price?
- Can I endure most trades ending in small losses as 'normal'?
FAQ
What is the win rate of the Turtle method?
It is not a method described by win rate. It is a reward-to-risk structure where many trades end in small losses and a few end in large gains, and the hit rate is low by design. Expectancy comes not from the proportion of correct calls but from the reward-to-risk profile of 'cutting small many times and riding a few big.'
Can I use numbers like 20, 55, and 2N as they are?
Those numbers are example parameters fitted to the 1980s U.S. futures market. If the market, timeframe, or volatility environment differs, recalibration is needed, and there is no guarantee that a specific number is valid everywhere. What matters is not the numbers but the structure: lagging breakout entry, N sizing, and mechanical stops.
Can I apply it to crypto perpetuals as is?
You must adjust two things together. First, the low-correlation diversification that gives the original system its power weakens as coins move in lockstep, so avoid going all-in. Second, at high leverage the liquidation price can be hit before the 2N stop, so back-calculate the leverage so the stop sits inside the liquidation price, and reflect long-held funding fees in your expectancy.
Why enter only after a breakout has occurred? Isn't that late?
Lagging entry is not a flaw but the design. In exchange for giving back the start and the tail of a trend, you avoid losing the account to failed predictions and ride only trends that are already confirmed. The trade-off is that in a sideways market false-breakout losses accumulate — which is likewise a cost the system is designed to accept.