The Second Exit

The claim, stated the way its believers state it

Always use a stop. It is the least controversial sentence in retail trading and there is a real argument under it. A position without a stop has no defined worst case, and a trader without a defined worst case cannot size, because lesson 20’s fraction is a fraction of something. A stop turns an open-ended exposure into a bounded one, which is the precondition for every piece of arithmetic in module 3. Nobody sensible argues against that.

What the advice never says is what happens when the thing you are stopping already knows how to get out. Lesson 63’s rule has an exit: it goes flat when the fast average crosses back below the slow one. Adding a stop does not add protection to that exit. It adds a competitor.

The column this lesson adds to the sheet

Module 12’s running object is the seven trades lesson 86 laid out, and this lesson adds the column that decides whether a stop is ever touched: the worst close inside each trade, measured from the entry in R. Every price below is one of the sixty closes, and every distance is that close less the entry, divided by 1.5443.

Read the fifth column before anything else. Four of the seven trades never close below their own entry at any point in their life; their worst moment is a gain. Whatever a stop is doing on those four, it is not being tested by them, and no distance you could choose would change what they made. Of the three that do go under water, the deepest goes 0.453 of an R, on trade 2. That single number decides the whole of the next table, because a stop wider than 0.453 R is a stop that never fires on this record at all.

The dial, at every ordinary setting

Here is the rule, in one sentence: leave the trade at the first close at or below the entry price less some distance, and if that never happens, leave it where lesson 63’s rule would have. The trade count never changes, so the costs never change: seven round trips at 0.1230, which is 0.861 in every row.

The bottom four rows are the same row. A stop at half an R, at one R, at two R and no stop at all produce a record identical in every figure, because the deepest water any trade goes into is 0.453 R and none of those distances reaches it. A trader running the half-R stop and a trader running none have the same year, the same statement and the same story about how disciplined they are, and only one of them is telling it.

Then look at the top two rows, which are also the same row while firing on a different number of trades. At a quarter of an R the stop fires twice; at a tenth it fires three times; the record is 3.94 both ways. The third firing is trade 5, the only loser on the sheet, and its stop price and its rule exit are the same price, 106.6 at bar 53. A stop that fires and costs nothing exists, and on this record it is the one case out of five where a stop had anything to do with a losing trade.

What separates 9.84 from 3.94 is the interval between 0.35 and 0.46 of an R. Outside it a stop is decoration; inside it a stop is the strategy. Nothing on the page or in the reasoning behind the rule tells you where that interval is before you measure it, and it is a property of this rule on this window rather than a number to carry anywhere. Almost nobody you trade against has located their own, and locating it costs an evening: the entries are already in the statement, and the only new column is how far under water each trade went before it worked.

Trade 2, end to end, with a quarter-R stop under it

The rule goes long at the close of bar 9, at 103.7. The stop sits at 103.7 less a quarter of 1.5443, which is 103.3139.

Bar 10 closes at 106.7. That is a gain of 3.00 a share, the largest open profit any of the seven trades ever shows, and 1.943 R. At this moment the trade is the best thing on the sheet.

Bar 11 closes at 104.6. Still a gain, 0.90.

Bar 12 closes at 103.0. That is 0.70 below the entry and 0.4533 of an R, the deepest water any of the seven trades enters, and it is 0.3139 below the stop. The stop fires. The trade closes at 103.0 for a loss of 0.70, which after the round trip is 0.823.

Bar 13 closes at 105.7, and this is where the rule was going to leave anyway, for a gain of 2.00.

The stop did not protect the trade from a loss. It converted the trade that had held the largest gain on the sheet into the largest loss on the sheet, one bar before the exit it already had. And it did the same thing to trade 1, which went 0.324 R under water on bar 7 and would have made 2.70. Both trades the stop fired on were winners. Neither of the two trades it never touched was.

A stop does not cut your losses. It cuts whichever trade reaches it first, and on a rule that works, most of those are winners.

This is not an argument for trading without one, and the bounds below say why at some length. It is an argument for knowing which of your two exits is doing the work, because one of them is, and the record will not tell you which unless you ask it twice.

So take your own last twenty closed trades and add one column: how far under water each one went, in your own R, before it finished.

What this does not settle

That stops are bad. This page measured one thing: a stop added to a rule that already exits. It says nothing about a stop as the only exit, which is a different instrument doing a different job, and nothing about a position sized off a stop, which is what lesson 20 needs a stop to exist for and remains true. The finding is about competition between two exits, not about protection.

That sixty closes can see the case a stop is actually for. They cannot, and this is the concession that costs the most. A stop earns its keep in the tail: the gap through the level, the halt, the morning the instrument opens twelve per cent lower on news that broke overnight. Nothing like that happens in this window, and a stop measured on data with no tail in it will always look like a tax, because in that data it is one. The number this page cannot give you is the one that matters most, and no window of sixty closes can give it to anybody.

That a close is where a stop fires. It is not. A resting stop order triggers intrabar, at a price this series does not print, and every figure above is therefore the friendliest possible version of the story: a stop that only ever checks itself once a bar fires less often and later than a real one. On the same seven trades a real stop would fire at least as often as this table says and never less, so the direction of the error is known even though its size is not.

That the interval between 0.35 and 0.46 R means anything beyond this rule and this window. It is a property of how far these particular seven trades happened to dip, on a series with a first-order autocorrelation lesson 85 measured at minus 0.6594, which is not what an ordinary instrument does. What generalises is the shape of the finding rather than its location: there is such an interval for whatever you trade, it is narrow, and outside it your stop is doing nothing at all.

And the objection a careful reader is already forming, which is right: the two trades the stop killed would not have existed in the form shown. Stopping out at bar 12 frees the capital, and the next signal arrives against a flat book rather than a full one, so the trades after a stop are not the trades on this sheet. Everything above holds the entries still while moving the exits, and a live account does not do that. Which raises the question this module has been walking toward since lesson 86: if the exits change the entries, then the one thing you actually choose every time, and the only column not yet on the sheet, is how much to put on. Lesson 88 puts it there, and finds that flat sizing returns the same 9.84 in every one of the 5,040 orders these seven trades could have arrived in, while doubling after a win and halving after a loss gives up 43.7 per cent of what its own average position would have earned.