Where Liquidity Rests
A stop a tenth of an average bar below the level has the best reward to risk on this page at 9.2 to one, needs a win rate of only 9.8 per cent to break even, and loses money on a method that wins 45 per cent of the time. Four trades in five never find out whether they were right. What removes them is a pile of other people’s stops resting where yours is, at a price nobody can see and everybody arrived at the same way.
Prerequisites: Lesson 21, which put the stop one ATR beyond the level and called that a starting point rather than a law, and lesson 17, for b, which is what every extra inch of buffer is paid for in.
Module 3 was about you: your edge, your fraction, your record, your worst hour. This module is about the market, and it starts here because every pattern in it is a consequence of one fact that is easy to state and rarely stated: orders that have not executed yet are not spread evenly across the price axis. They pile up. And they pile up in places you can name in advance.
A stop is a market order that has not happened yet
Start with what a stop order actually is, because the answer is less familiar than the name suggests. A stop is not an order sitting in the book waiting to be filled. It is an instruction with a trigger: when the market trades at or through your price, send a market order. Until that moment it is not an order at all, and until that moment nobody sees it — not other traders, not the exchange’s public feed, in most retail arrangements not even the exchange. It sits with your broker as a condition.
Two things follow, and both matter for the rest of the module. The first is that a stop, when it fires, is not a request to trade at a price. It is a request to trade at whatever price exists, which is why lesson 11’s slippage lands hardest exactly here. The second is that a hundred stops at the same price are a hundred market orders that will all arrive in the same direction within the same second, and that nothing in the visible market shows they are there beforehand. They are a quantity of certain future flow, at a known price, which nobody can observe.
Why they collect in the same places
Now ask where a stop goes. Lesson 21 answered it: beyond the price at which the reason for the trade stops being true. In practice that resolves into a small set of prices — below the swing low, above the swing high, beyond the round number, beyond yesterday’s extreme, beyond the trendline everybody has drawn through the same three points. Those are not arbitrary choices. They are the honest answers to the question, and there are not many honest answers.
So the clustering needs no coordination and no intent. It needs only a shared convention, which is the same reason queues form at one door of a building with four. Everyone who read lesson 21 and everyone who read the books that say the same thing will place a stop within a few cents of everyone else, and none of them will have communicated. The level does not attract stops because it is significant; it becomes significant because stops are there.
This has been measured rather than merely asserted. Osler, working with a currency dealer’s actual order records, found stop-loss orders clustering just beyond round numbers and take-profit orders clustering at them — and found that the two produced different price behaviour, the first accelerating moves through the level and the second reversing them at it. Kavajecz and Odders-White found the same thing from the other side, showing that the levels technical analysts mark coincide with where depth actually sits in the book. Neither paper needs anyone to be hunting anybody. They only need people to share a rule.
What the cluster does when price reaches it
Price arrives at the level. The stops trigger. For a few seconds there is a burst of market orders all on the same side, which pushes price further in that direction, which triggers the stops sitting slightly further out, which pushes it further again. That is a cascade, and it is the second thing Osler measured directly.
Then it stops, and the reason it stops is the interesting half. The flow that produced the acceleration was finite. Every stop in the cluster has now become a fill; there is nothing left behind the move. So the characteristic shape is not a break but a lurch followed by a stall — price travels quickly through a level on flow that has no opinion about value, runs out of that flow, and is left somewhere it has no particular reason to be. Whether it comes back depends on whether anyone wanted the other side; often somebody does, because the selling that produced the move was forced rather than chosen, and the price it reached was nobody’s estimate of value.
You can tell that story with intent in it, and it is usually told that way: somebody hunted your stop. The difficulty is that the version with intent and the version without it predict exactly the same picture, so the picture cannot decide between them. They do differ in one testable place. If the move is engineered by somebody with size, its depth should scale with what that participant is trying to fill. If it is a cascade through resting stops, the depth should scale with the ordinary volatility of the instrument, because that is what set the spacing of the stops in the first place. The second is measurable on a chart and the first is not, which is a reason to prefer the second that has nothing to do with charity.
How far beyond the level, priced
Lesson 21 said one ATR beyond the level and admitted it was a starting point. Here is what makes it a number. Take a long entered on the reclaim of a level, sized and stopped by lesson 20’s formula: entry 0.4 ATR above the level, target 5 ATR above it, and the system this course has carried throughout — it works 45 per cent of the time when it is left alone. Put the stop d ATR below the level and two things move at once.
The payoff ratio falls as d grows, because the risk is the denominator: at 0.10 ATR the trade is 9.2 to 1 and breaks even at a 9.8 per cent win rate, and at 2.00 ATR it is 1.92 to 1 and needs 34.3 per cent. On that column alone the tightest stop wins by a distance. The second thing that moves is the chance the stop is still there when the trade works, and that is where the cluster arrives.
Suppose a probe past the level travels, on average, half an ATR before turning — a number you will measure for yourself in problem 1, and which is the only number in this table that is not derived. Then:
| Stop, below the level | Payoff ratio | Breakeven win rate | Survives the probe | Expectancy |
|---|---|---|---|---|
| 0.10 ATR — “just below” | 9.20 | 9.8% | 18.1% | −0.17R |
| 0.25 ATR | 7.08 | 12.4% | 39.3% | +0.43R |
| 0.50 ATR | 5.11 | 16.4% | 63.2% | +0.74R |
| 0.75 ATR | 4.00 | 20.0% | 77.7% | +0.75R |
| 1.00 ATR — lesson 21’s default | 3.29 | 23.3% | 86.5% | +0.67R |
| 2.00 ATR | 1.92 | 34.3% | 98.2% | +0.29R |
The first row is the whole lesson. A stop a tenth of an ATR below the level has the best payoff ratio on the page and the lowest breakeven win rate on the page, and it loses money. It loses money on a system that wins 45 per cent of the time and needs 9.8 per cent to break even, which is the kind of margin that ought to be impossible to squander. It is squandered in the survival column: four trades in five never find out whether they were right, because the probe removed them first.
The last row is the other half. Buying certainty is expensive too. At 2 ATR almost nothing takes you out by accident, and the expectancy has fallen to little more than a third of its best value, because you are now paying 1.92 to 1 for a system that could have had 5 to 1. Somewhere between those two the two forces balance.
They balance around 0.6 ATR here, which is a little more than the probe averages — and the useful thing about that balance is how flat it is on top. Anything between 0.47 and 0.84 ATR keeps at least 95 per cent of the best expectancy available. Which means the honest instruction is not a number to three decimals but a zone: get inside it, then stop optimising. Inside the zone the curve barely knows one choice from another, and outside it nothing else you do makes the difference back.
And it puts lesson 21’s default in its place. One ATR is not the optimum here, but it is close enough to it to be a defensible thing to say to somebody who has not measured anything yet: it keeps 88 per cent of the best, and it errs on the side that costs you a little expectancy rather than the side that costs you the trade.
What this does not settle
That the shape of the probe distribution is known. The survival column assumes probe depth is exponentially distributed about its mean, and it is not obvious that it should be. Run the same arithmetic with a half-normal of the same mean and the optimum moves to 1.68 times the mean probe depth; with a lognormal of the same mean and a log standard deviation of 0.75 it moves to 1.34, against the exponential’s 1.26. So the finding that survives the assumption is a range and a shape — an interior optimum, at one to two times the mean, on a flat top — and the finding that does not survive it is any tidier rule than that. If you catch yourself repeating a specific multiple, you have kept the wrong half.
That the numbers are yours. Every row is computed at a 45 per cent win rate, a target 5 ATR out and an entry 0.4 ATR above the level. Your own three move every cell, and the mean probe depth is different for every instrument and probably for every session. The table is a method, and problem 2 is where you run it on your own numbers.
That every level has a cluster under it. A level price has touched once has almost nothing resting beyond it, because only one cohort of traders has had the chance to put anything there. This whole lesson is about a quantity that varies, and it varies most obviously with how many times a level has been visited and how visible it was.
That you can see any of this. Not one number in this lesson is observable in advance. The cluster is invisible by construction — stops live at brokers, not in the book — and everything above is inferred from what happens when price arrives rather than seen before it does. Which raises the obvious question about the one thing here that is visible, the displayed depth in the order book, and lesson 26 is the answer: it is visible, it is not reliable, and the reasons those two facts coexist are not the ones you would guess.
That the cascade is the same thing as manipulation. It is not, and the distinction is worth holding on to for the rest of this module. A cascade needs no author. Somebody deliberately pushing price to trigger a cluster is a real and separate thing, it is illegal in most jurisdictions where it can be proved, and it is lesson 27’s subject. Treat them as one thing and you give up the only testable difference between them, which is the one place in this lesson where a chart can settle anything.
The level is not where price turns. It is where a great many people have agreed to be wrong at the same price, and the turn — when it comes — is what that agreement costs them.
Problems
- Measure the zone on your own instrument. Find thirty occasions when price traded through an obvious level — an equal high or low, a round number, yesterday’s extreme — and turned back within the same session. For each one, record how far past the level it went before turning, divided by the ATR at the time. The mean of those thirty is the only input in the table above that you cannot derive, and it is the number the rest of this lesson is waiting on. Expect it to differ by instrument and by session; that is not noise in your measurement, it is the answer.
- Rebuild the table with your own four numbers. Your win rate and your typical target from lesson 17, your entry offset, and the mean you just measured. For each candidate stop distance, the payoff ratio is (target − entry) ÷ (entry + d), with target, entry and d all measured from the level, the survival term is the fraction of your thirty probes that fell short of d, and the expectancy is p × survival × b − (1 − p × survival). You are looking for the flat top, not the peak.
- Price the stop you are already using. Take the rule you actually follow, express its distance in ATR, and put it on your own curve from problem 2. If it sits short of your measured zone, you now know what that has been costing you and it is not a small number. If it sits well beyond the zone, you know what the safety has been costing you, and that is a smaller number but it is not nothing either.
Where the problems above ask you to go and count something, How to Collect a Base Rate is the appendix that says how: define the observation, fix the criterion before you look, take consecutive cases rather than the memorable ones, and count into four cells.
Sources. Carol Osler, “Currency Orders and Exchange Rate Dynamics: An Explanation for the Predictive Success of Technical Analysis” (Journal of Finance, 2003), which is the empirical basis for most of this lesson: working from a dealer’s actual order book, it finds stop-loss orders clustering just beyond round numbers and take-profit orders clustering at them, and shows the two produce opposite price behaviour. Carol Osler, “Stop-Loss Orders and Price Cascades in Currency Markets” (Journal of International Money and Finance, 2005), for the cascade itself — how a triggered cluster propagates into the stops beyond it, and why the resulting move is fast and then finished. Kenneth A. Kavajecz and Elizabeth R. Odders-White, “Technical Analysis and Liquidity Provision” (Review of Financial Studies, 2004), for the same clustering seen from the book rather than from the orders, and for the useful complication that the levels move as liquidity does rather than sitting still and being respected.
This lesson is about orders nobody can see. The next one is about the orders everybody can see — the displayed depth in the book, why a wall of size is an advertisement rather than a barrier, and the two tests that separate the size that means it from the size that does not.
The Order Book Is Theater
The orders you can see, and why they are worth less.
Read Lesson →Educational only. Trading involves substantial risk of loss. Not financial advice. Past performance does not guarantee future results.
💬 Discussion (0 comments)
Loading comments...
Ready to Trade with Signal Pilot?
Apply your trading education with professional indicators and real-time market analysis tools.
Back to Signal Pilot →