Turtle Soup Trading Rules: Which Extreme You Are Fading

Turtle Soup is a setup that buys when a market has just made a new low, and sells when it has just made a new high. The trade is taken against the break, on the argument that a large share of fresh extremes fail and the market snaps back inside the range it just left. A snap back that runs out and gives way to a new extreme is a failed recovery inside a downtrend, which carries a naming problem of its own.

That much every published description agrees on. What none of them settles is the part a reader has to know before placing a single order: which extreme, measured over what window, and how recently the last one occurred.

What Turtle Soup Fades, and What It Does Not

The setup is named after the group it was aimed at. The Turtles were traders trained in the early 1980s to follow trends by buying breakouts of a channel measured over a fixed number of past sessions. Turtle Soup takes the opposite side of that same signal. Where the breakout system buys a new 20-day high, Turtle Soup sells it.

That dependency is the whole point, and it is why the setup cannot be described without naming a window. A fade needs something to fade. If you cannot say which level the crowd was buying, there is no failed signal to trade against, only a market that moved.

The system being faded is a separate subject with its own position sizing, unit limits and correlation caps, and it is set out in full on our page on the original Turtle breakout rules. This page stays on the fade side of that signal and does not repeat the entry system itself.

What Turtle Soup is not is a general reversal method. It says nothing about a market drifting inside its range, nothing about a trend that has been running for months, and nothing about a break of a level you drew by hand. It applies to one event: a fresh extreme of a defined lookback, immediately reclaimed.

Which Extreme the Rule Names, and the Window Nobody States

Four descriptions of the setup on the first page of results could be read in full. Searching all four for the word lookback returns nothing. Not one occurrence in any of them.

Two of the four never state any window at all. They describe the setup entirely in terms of a swing high or a session high being taken out, which is a different object: a swing high is whatever the reader judges to be one, and a session high depends on which session boundary the reader uses. Both are movable. A 20-day extreme is not.

The other two do name a 20-day window, and both attach it to the original rule. So the four pages, read together, describe two rule sets under one name, and a reader moving between them has no way to notice the substitution.

This is not a quibble about vocabulary. The three rows above produce different trades from one chart on one day. A 20-day low is a single number that a script can compute and a record can be kept of. A swing low is a judgement, and a page that skips the window has quietly handed the reader the hardest part of the rule while appearing to have given them a rule.

If you want the counting rules for a fixed-window channel, including whether the current bar is included in its own calculation, our page on how a 20-day channel is drawn covers that and the feed differences behind it.

The Age Condition, and a One-Session Disagreement

The original rule set does not fire on every new extreme. It requires the previous extreme of the same window to be some minimum number of sessions old before the new one counts.

The purpose of that condition is easy to state once you see it. A market grinding to new lows day after day is producing a new 20-day low every session, and each one is a trend doing what trends do. The age condition removes those. What survives is the case where the market went quiet, the old low sat untouched for a while, and then price reached down through it in one move. Only that second case is a break with something to fail.

Across the four readable explanations, the strings four days, 4 days and three days each return zero occurrences. None of them mentions any age condition in any form.

Two references outside that group do state it, and they put the threshold in different places. An article hosted on the MetaTrader developer site sets it at three, counting forward from the earlier low. A charting vendor’s concept library sets it at four, expressed instead as how old that earlier low has to be.

Those two framings admit a different set of bars. Read as one rule described from opposite ends, the gap between them is inclusive counting and nothing more. Read as two rules, one of them is simply wrong. Neither reference shows the working, and the 1995 book both are summarising is not a document this page has read, so no single figure is put forward here as the condition.

What you can do is decide the count yourself and write it down before you start, because a setup whose qualifying condition you have not pinned is a setup whose results cannot be compared with anyone else’s, including your own from last month.

What the Rule Depends On That Your Broker Decides

Both halves of the qualifying test are counted on bars, and on a retail forex or contract-for-difference account those bars are your broker’s, not the market’s.

The forex week has no closing auction and no single tape. Where a broker puts the daily boundary decides which prices land in which bar, and the number of bars between two dates depends on how that broker treats the Sunday open and any holiday. Change the boundary and both the 20-day low and the count of sessions since the previous one can move.

None of the four explanations mentions a price feed, a session boundary or the five-day week even once. They are written as though the chart were a fact about the market rather than a rendering of one provider’s data.

The consequence for this setup specifically is narrow and checkable. The same chart can qualify at one broker and fail to qualify at another, not because the market did anything different but because the two feeds disagree about where a day starts. That is a reason to run the test on the feed you will actually trade on, and to record which one it was.

The wider question – why one provider’s channel sits at a different level from another’s, and why a band built from bid prices is not the price your order fills at – is handled on the Donchian channel page named above rather than repeated here.

Entry Is a Resting Order Back Inside the Range

The mechanic that gives this setup its shape is the order type. The trade is not entered when the new low prints. It is entered by an order left sitting a small distance back above the level that was just broken, so it only triggers if price returns through it.

That single choice does most of the work. It means the market must already be reversing before any position exists, and a break that keeps going never fills the order at all. The cost is the mirror of that benefit: the fill is worse than the extreme, and a violent reversal can jump past the resting level.

Order behaviour on a leveraged account is worth understanding on its own terms, and our guide to how a stop order rests above the market sets out how these orders trigger and what happens when the market gaps through them.

The protective stop in the original construction goes just beyond the new extreme. That puts it beyond the level the trade is betting against, which is coherent: if the market makes a further new low after you have faded the first one, the argument for the trade has gone.

Turtle Soup Plus One, and the Trade-Off in Waiting

A second version of the setup exists in the same source material. Instead of buying the reclaim during the session that made the new low, it waits: the session closes at or below the broken level, and the reclaim is allowed to happen on the following session instead.

The trade-off is direct and does not need a statistic to see. Waiting a session filters out the reclaims that happen inside one bar and then fail by the close, which is the noisiest version of the pattern. It also gives up the trades that reversed hard and never came back to offer a second chance, and it moves the entry further from the extreme, which widens the distance to a stop placed beyond that extreme.

Neither version is a refinement of the other. They are two different bets about how quickly a failed break reveals itself, and they will produce different trade lists on the same data. Testing them as one setup, or switching between them when a trade goes wrong, produces a record that cannot answer anything.

The Win Rates Being Published, and What Supports Them

One of the four explanations states a win rate for the setup, given as a band, alongside a reward-to-risk range. It carries no source, no sample size, no instrument, no date range and no description of how the setup was defined for the purpose of counting. The other three state no win rate at all.

A figure like that cannot be checked and cannot be reproduced, and the earlier sections explain why it could not be even in principle. If four published descriptions cannot agree on what window defines the extreme, and none of them states the age condition, then two people counting trades from the same chart are not counting the same setup. A single percentage over an undefined rule set is a number without a referent.

This page states no expected win rate, no average return and no success rate, because no source it could verify publishes one. If you want a figure for this setup, the only one worth anything is the one you generate yourself, on your own feed, with the window and the age condition written down before the first trade is counted.

Before You Take One: A Five-Point Check

Everything above reduces to five things to settle before a Turtle Soup trade is worth taking, in order.

1. Write the window down. Twenty sessions is the figure the name descends from. If you use something else, that is a decision, not a detail, and it belongs in your notes rather than in your head.
2. Write the age condition down. Pick the count of sessions the previous extreme must have gone untouched, apply it identically every time, and treat a market printing consecutive new extremes as disqualified rather than as an opportunity.
3. Check it on the feed you trade. Run the test on your own broker’s daily bars, not on a chart from a different provider, and note which feed the record was built on.
4. Place the entry as a resting order, not a click. The setup is defined by requiring the market to come back through the level. Entering at the extreme is a different trade with a different risk.
5. Decide in advance which version you are trading. Same-session reclaim or next-session reclaim, chosen before the trade rather than after seeing what the bar did.

If a break does not clear those five, the honest answer is that it is not this setup. A market can break a level and come back for reasons unconnected to a failed breakout, and separating the two is the same problem covered from the other direction in our page on avoiding false breakouts.