Turtle Trading Rules: The Full System, Tested to 2026

Michael Whitman · Trading since 2017, full-time since 2020 · YouTube creator, 137K subscribers · Published 2026-09-25 · Updated 2026-11-03 · Reviewed 2026-11-03

Turtle trading is the breakout system Richard Dennis and William Eckhardt taught novices in 1983: buy a 20- or 55-day high, size each unit so 1N equals 1% of the account, and exit on a 2N stop or an opposite breakout. On seven Turtle markets, the rules gained heavily from 1984 to 2003 and lost from 2004 to October 2026.

Most retellings stop at the legend. The rules themselves were published in full in 2003, so this page works from that document line by line, sizes one real trade with them, and then asks the question the search results leave open: what the same rules did after everyone could read them.

How it works

Where the rules come from. The Original Turtle Trading Rules was released free in 2003 through OriginalTurtles.org, with a foreword signed by Curtis Faith, one of the trainees. It describes an experiment that began when Dennis and Eckhardt disagreed about whether trading could be taught. Every rule below is from that document, re-read on 1 Nov 2026.

The markets. Liquid US futures only: Treasury bonds, notes and bills, Eurodollars, the S&P 500, six currencies (Swiss franc, Deutschmark, British pound, French franc, Japanese yen, Canadian dollar), gold, silver, copper, crude oil, heating oil, unleaded gas, coffee, cocoa, sugar and cotton. A trainee could leave a market out, but then had to leave it out every time.

N, the volatility unit. True range is the largest of today’s high minus low, high minus the previous close, and the previous close minus low. N starts as a 20-day simple average of it and then updates as N = (19 × yesterday’s N + today’s true range) ÷ 20, which the document notes is the measure now called average true range (ATR).

The unit. Every position was built in units, each sized so a 1N move equals 1% of the account:

unit = (1% of account) ÷ (N × dollars per point)

The two systems, step by step

1. Entry. System 1 buys one tick above the 20-day high (sells one tick below the 20-day low). System 2 does the same at 55 days. Both act during the day, or at the open if price gaps through.
2. The System 1 skip rule. A 20-day breakout is ignored if the last 20-day breakout in that market, taken or not, would have been a winner. A losing one is a breakout where price moved 2N against it before a profitable 10-day exit. When a signal is skipped, the 55-day breakout becomes the entry, the “failsafe”. System 2 takes every signal.
3. Adding. One more unit every ½N beyond the previous fill, up to four.
4. Stops. Each unit’s stop sits 2N from its fill, so one unit risks 2% of the account. Each add raises the earlier stops by ½N, which usually puts the whole position 2N from the newest unit.
5. Exits. System 1 leaves on a 10-day break the other way, System 2 on a 20-day break. The document calls these the hardest rules to follow, because waiting for them means watching open gains shrink.
6. Limits. 4 units per market, 6 in closely correlated markets (crude and heating oil, for example), 10 in loosely correlated ones, 12 in one direction overall.
7. Drawdown. Each 10% fall from the year’s starting equity meant trading the account as if it were 20% smaller: a $1,000,000 account down $100,000 was traded as $800,000.

The entries and exits are Donchian channel breaks, which the document credits to Richard Donchian. Nothing in the rules forecasts direction; they only decide size and exit.

A worked example: the S&P 500, November 2023

This trade comes straight out of the test below, so every number is a real close of the S&P 500 index. The test uses closing prices only, so the breakout and the fills are closes rather than intraday ticks.

The breakout. The highest close of the previous 55 sessions was 4,515.77 (1 Sep 2023). On 20 Nov 2023 the index closed at 4,547.38, above it: System 2 goes long one unit. N from the session before was 29.92 points, so ½N = 14.96 and 2N = 59.84.

Sizing it. CME’s Micro E-mini S&P 500 is $5 × the index (contract specs read 1 Nov 2026), used here with the index level in place of the futures price. On a $100,000 account, 1% is $1,000, and N × dollars per point is 29.92 × $5 = $149.61. One unit = $1,000 ÷ $149.61 = 6.68, truncated to 6 contracts. The 2N stop then risks $1,795.30 a unit.

Adding. The next unit needs a close at least 14.96 above the last fill:

The exit. The index peaked at 5,254.35 on 28 Mar 2024. On 15 Apr 2024 it closed at 5,061.82, below the lowest close of the previous 20 sessions (5,117.09, set on 15 Mar), and all four units came out after 99 sessions. That is 17.19N, 16.51N, 15.61N and 14.68N, 64.00N in total. With 6 contracts a unit, the four units gained 1,915.03 index points between them, worth $57,450.89 before costs, and at the peak they controlled about $550,000 of index exposure on a $100,000 account.

What the Turtles’ own N would have changed. With true range taken from highs and lows, N on 17 Nov 2023 was 49.02 points, not 29.92, so a unit would have been 4 contracts instead of 6. From 1984 to October 2026, N from closes ran a median 58.33% of the true-range N on the S&P 500 (58.81% to 2003, 57.93% after). That matters for reading the test below: its N counts run larger and its 2N stops sit tighter than the Turtles’ own would have, by about the same amount in both periods.

The original data

The question the search results leave open is whether the rules still work. To answer it, both systems were run, with adds, stops, exits and the 12-unit limit, on seven markets from the Turtles’ list, from 3 January 1984 (the trainees started trading that month) to 30 October 2026:

• S&P 500 index closes (Yahoo Finance).
• 10-year Treasury yield (FRED), where a trade long the yield is read as short the note.
• Japanese yen, Swiss franc, British pound and Canadian dollar as the dollar price of each currency (FRED).
• WTI crude oil spot (FRED; the series starts in January 1986, so crude trades begin in March 1986, and it skips some days, up to 122 in 2003).

What was left out, plainly. Prices are daily closes: breakouts, stops and adds are judged and filled at the close, and true range is the close-to-close move. The series are spot prices and yields rather than the futures, so there are no contract rolls. No costs, slippage or interest, no drawdown rule and no correlated-market limit. Results are in N summed over units; under the rules, 1N is 1% of the account. A market opens no new trade on the close where its last one exited; allowing that would add 1 System 2 trade and 68 System 1 trades, and every conclusion below stays the same.

System 2: 1,735 trades, and 1,397 of them (80.52%) lost. The median trade lost 3.69N, and the net across all of them was +2,717.26N. The best tenth of trades, 173, added up to +8,730.35N, more than three times the net. That is the shape the document describes: most breakouts fail, and a few long trends pay for them.

Time is what separates them. Counting sessions from the first unit to the exit, the 338 winners lasted a median 49 sessions and the 1,397 losers a median 7. All 918 trades that ended within 10 sessions lost, -4,140.28N between them. At the other end, 115 of the 119 trades that ran 60 sessions or longer were winners, and together they made +5,923.11N, more than twice the net for all 1,735. In practice the system asks a trader to cut quickly and then sit still for months, and the 60-session trades were where its whole result came from.

The gap is unlikely to be chance, but the line was drawn after looking. Yearly results averaged +149.36N from 1984 to 2003 and -11.73N from 2004 on (Welch’s t-test on the yearly figures, p = 0.0033). That p-value assumes the year was fixed in advance; it was set at publication, but not before the results were seen, so read it as hindsight. Taking the widest gap from every dividing year from 1990 to 2018 and testing it against shuffled years still gives p = 0.033, and every one of those years splits the record with p below 0.05. The data look more like a long fade than a sudden break in 2003.

System 1, with its skip rule, shows the same pattern: 2,706 trades, 2,121 losers (78.38%), +2,656.93N to 2003 with 3 losing years out of 20, then -74.75N from 2004 with 12 losing years out of 23 (p = 0.0072, with the same hindsight caveat). Since 13 May 2022, System 2 has not regained its high, and by 22 Oct 2026 it sat 558.48N below it.

It was not one market. Before 2004, six of the seven markets were net positive under System 2; after 2003, five were net negative. The yen made the most and lost after: +1,049.63N from 1984 to 2003, then -179.00N. Crude oil ran the other way, -248.30N before and +134.32N after. Every trade, with its fills, N, exits and reason, is in the System 2 trade list, with the System 1 trade list beside it.

Checks. With one unit and no adds, System 2 averaged +52.65N a year to 2003 and +1.74N after (p = 0.0125): the gap held, but the later years were close to flat rather than a loss. FRED’s S&P 500 closes gave the identical 64 trades from 2017 that Yahoo’s did. For the yen, Yahoo’s JPY=X closes, taken at a different time of day than FRED’s noon New York rate, matched 114 of FRED’s 135 trades from 2005 within five days, and both lost (-147.68N on FRED, -97.31N on Yahoo). An earlier one-unit test of System 2 on the S&P 500 alone, with intraday highs and lows, made 239 trades from 1983 and lost on 153; its trade list and results by size are still published.

What the test cannot say is why. Publication in 2003 is one change; markets, interest rates, the number of trend-following funds and the cost of trading all changed too. This is one rule set, on seven markets, on past prices.

When it fails

The losing streaks are long by construction. System 2 once ran 26 losing trades in a row across the seven markets. A trader who stops after a run like that misses the trend that would have paid for it, which is the failure the document itself blames for most trainees who did not make money. Within 10 sessions SPY slipped back under the breakout level 148 times in 193 fresh 20-session breakouts since 1994 (76.68%), according to Donchian channels vs ATR.

Years can pass without a new high. System 2 went from 7 Jan 2015 to 5 May 2022 below its previous peak, 2,675 days, and has been under the 13 May 2022 high since.

Gaps beat stops. A 2N stop is a level, not a promise. The document describes October 1987, when some Turtles holding interest-rate futures lost 20% to 40% of equity in one day after an overnight rate cut.

Small accounts lose the precision. At the Turtles’ own N, a $100,000 account could hold only 4 Micro E-mini contracts a unit in the worked example, and many futures have no micro version, so one contract can already be more than a 1% unit.

A published edge invites a crowd. Anyone can run channel breakouts now. On these seven markets, the period after 2003 lost money before costs, and costs would only make that worse.

Questions people ask

Does turtle trading still work?

Not on this test. On seven Turtle markets, System 2 lost 269.86N from 2004 to October 2026 and System 1 lost 74.75N, both before costs, after very large gains from 1984 to 2003. One market list and one rule set cannot prove it fails everywhere, and a one-unit version without adds was close to flat rather than losing, but nothing here says the original rules kept working.

How long does a turtle trade last?

In the System 2 test above, a winning trade lasted a median 49 sessions, about ten weeks, and a losing one a median 7. The worked S&P 500 trade ran 99 sessions from first unit to exit.

How did Richard Dennis make his money?

By trading commodity futures. Wikipedia reports he started in the early 1970s with $1,600 borrowed from family, $400 of it left after buying an exchange seat, and became known for holding trends longer than most floor traders.

What happened to Richard Dennis?

Wikipedia, citing Time (17 Oct 1988), reports losses of about $50 million in 1987 and 1988, and says he stopped managing pooled client money in the spring of 1988. In 1990 his firm paid over $2.5 million to settle complaints that it had not followed its own trading rules, without admitting or denying wrongdoing (The New York Times, 2 Nov 1990).

What were the Turtles’ results?

The 2003 document says the group earned an average annual compound return of 80% over four years. That figure is reported, not audited, and the same document says many trainees did not make money because they did not follow the rules.

Related

Donchian channels draw the Turtle entries and exits on a chart. Average true range (ATR) is the N behind every unit and stop, and position sizing covers the 1% unit idea in more depth. Trend following is the wider approach, where a few large trends have to pay for a long list of small losses.

This page is educational, not financial advice. Test every idea on your own charts before risking money.