Why stops get hit

The sequence, and what it is called 

The observation is specific and worth stating precisely before it is explained. Price does not merely reach the level. It reaches the level, passes a short distance beyond it, closes the position, and then reverses. The reversal is what makes the sequence memorable, because a level that is passed and never revisited produces no story at all. In trading vernacular the sequence is called a stop hunt, and the name has stuck because it describes how it looks from the position that was closed.

What follows is a structural account. It describes properties of an order book, of a two sided quote and of charting software, and it makes no claim about the conduct of any firm in either direction. The mechanisms below are sufficient to produce the sequence on their own, which is the useful thing about them: they are checkable against a chart and an order log, and they suggest questions that can actually be answered.

Why orders collect at the same levels 

A chart is a shared object. The prior swing high, the low of the previous session, the round number a few units above, the widely quoted moving average: all of them are derived from data every participant holds, by arithmetic every participant can perform. Independent traders working independently therefore arrive at very similar levels, and the conventions for placing a stop relative to a level are similar too, because a level is only useful as a boundary if the order sits on the far side of it. The result is not a coincidence and does not require coordination. A short distance beyond every visible level sits a band of resting orders, placed by people who have never met.

The second half of the mechanism is what those orders are. A stop that closes a long position is a sell instruction. A stop that closes a short position is a buy instruction. So the band sitting below a widely watched support level is a band of resting sell orders, and the band above a widely watched high is a band of resting buys. When the first of them is triggered it becomes a market order, which consumes the nearest resting bids, which moves price a little further down, which reaches the next orders in the band. The chain sustains itself for as long as the band lasts and stops when the band is exhausted, because the selling that produced the move was the stops themselves and there is none left.

That is the whole sequence, including the reversal. A cluster being consumed produces a burst of one sided volume that is unrelated to any change in what the instrument is worth, and when it ends, price is left at a level the rest of the book was not transacting at a minute earlier. The retrace that follows looks abrupt because the move that preceded it was mechanical rather than informational. Nobody has to want any of it.

One further observation is worth reporting with its limits attached. Seen from the other side of the book, a cluster of resting stops is a pool of liquidity, and a participant with a large order to work has to work it where liquidity exists. Some technical traditions treat that as the central fact of intraday structure. Others reply that the reading is unfalsifiable, since any move can be labelled a visit to a pool after the event, and that it therefore cannot be tested by the person using it. Both positions are held by serious practitioners, and this page does not resolve them.

The price that triggers a stop 

An instrument does not have a price. It has two at every moment: the bid, at which a position can be sold, and the ask, at which one can be bought. Which of the two triggers a stop follows from what the stop does. Closing a long position is a sale, so a long position's stop references the bid. Closing a short position is a purchase, so a short position's stop references the ask. The level typed into the order is compared against one specific series, and which series it is depends on the direction of the position rather than on any setting.

Charting software, meanwhile, draws one line. Conventionally it draws the bid, though the source is a setting and mid drawn charts are common. For a long position the drawn series and the triggering series are the same, so the candle that reaches the line is the candle that reached the order. For a short position they are different series separated by the spread, and the drawn chart can therefore stop short of the level at the exact moment the order was reached. Nothing anomalous has happened. The order referenced the ask, and the ask is not the line on the screen.

This matters most when the separation is widest, and the separation is not constant. Quoted spreads widen around scheduled announcements, at the daily rollover hour, at the thin ends of sessions and whenever the book empties, because a participant quoting continuously through an event carries an exposure they are not being paid to hold, and widening is how that exposure is priced. A separation that is ordinarily a fraction of a pip can be several pips for the seconds that matter, and the gap between the drawn line and the triggering series is exactly that wide.

The discrepancy is therefore asymmetric on any given chart. A chart drawn on the bid conceals the distance for short positions and shows it exactly for long ones, a chart drawn on the ask conceals it for long positions instead, and a chart drawn on the mid conceals half of it in both directions. Which of the three is on screen is a setting in the software rather than a property of the market.

How many orders are resting 

Price does not move because an order is large. It moves because an order is large relative to what is resting in front of it, and what is resting in front of it varies enormously by hour, by day and by instrument. The same order that crosses a fraction of a pip in a deep book crosses several pips in a thin one, and the book is thinnest at exactly the moments the previous section listed: the rollover hour, holiday sessions, the minutes bracketing a scheduled release, and the aftermath of an unscheduled one.

This is why a level can hold on one day and give way on the next with no change in how many participants care about it. Depth is a separate variable from interest, and a distance fixed beyond a level is measured against a book whose thickness was never fixed: a wide margin at one hour of the day, a negligible one at another.

The extreme case of a thin book is no book. Where trading is interrupted, over a weekend, a holiday, a halt or a disorderly moment after an announcement, prices between the last one printed and the next one printed do not exist. An order resting in that interval is reached at the first price that does exist. The level determined whether the order was triggered and played no part in where the position closed, and those two facts agree with each other only while trading is continuous. What the interruption does to the amount realised, and what a different class of order undertakes about it, is the subject of the next lesson.

A stop being reached is not a verdict on the stop 

Any level at a finite distance can be reached. A level that could not be reached is not a stop at all: it is a price the instrument is not expected to visit, and a position sized against that distance is a much smaller position than the same account would otherwise carry, because distance and size are two terms of one calculation. Being reached is therefore inside the design of the instruction rather than evidence against it, in the same way that an insurance policy paying out is not evidence the policy was mispriced.

It helps to be exact about what a placement encodes. A level records a condition: the point at which the reason the position was opened no longer holds, or the point at which the movement has left the range the instrument ordinarily covers. It is a statement about what would be concluded if price arrived there, not a forecast that price will not. Whether price returns afterwards is a fact about what happened next, and it does not alter what was knowable when the order was written.

There is also an asymmetry in what gets remembered, and it is worth naming because it distorts the impression this whole lesson exists to correct. A stop reached, followed by continuation in the same direction, leaves nothing to examine and no counterfactual to feel anything about. A stop reached, followed by a reversal, leaves a vivid one that remains visible on the chart for the rest of the session. The two are not equally memorable, so recollection over a month systematically over-represents the second. The imbalance is in the record keeping rather than in the market.

Separating a poorly placed level from an ordinary loss is possible, but only across many instances and only one attribute at a time. Whether triggers cluster within a small distance of the level, which would suggest the distance sits inside the instrument's ordinary movement rather than outside it. Whether they cluster at particular hours, which points at depth. Whether they cluster around scheduled releases, which points at the calendar. Each is an arithmetic question answerable from a record of trades and from nothing else, and the strength of feeling attached to any single instance is not information about the placement.

What practitioners do with clustering divides them, and the division is a trade-off with no free side. One convention places the level just beyond the visible boundary and accepts that the band is where it sits, on the reasoning that the boundary is the thing being tested. Another places it beyond the band, accepting a longer distance and therefore, on identical risk arithmetic, a smaller position. The placement question itself belongs to the two lessons before this one; what clustering adds to it is only that the cost is paid on both sides. Nearer levels are reached more often, and wider levels cost more when they are reached and permit less size.

Widening a stop after entry 

Moving a level further away while a position is open and losing is the response this lesson makes most tempting, so the arithmetic is worth following through. The amount at risk is the distance multiplied by the money value of one unit of movement at the size opened, and size was derived from distance at entry, so the two numbers were chosen together. Moving the level changes one of them after the other has been fixed, and the amount at risk moves in proportion. The result is not the position the sizing calculation authorised, and it is not a position that calculation would ever have produced.

The last two rows are the whole difficulty. The favourable case is highly visible: the position recovers, the wider level is never reached, and the outcome appears to vindicate the change. The adverse case records a loss larger than the one that was accepted, and it follows from precisely the same decision taken at precisely the same moment with precisely the same information. Judging the decision by which of the two occurred is judging a decision by its outcome, and the two are only the same thing when there is no uncertainty involved.

Two aggregates move as well, and neither is visible on the position itself. Reward to risk was computed at entry from the distance then in force, so a wider distance changes the ratio the trade was accepted on, downward, after the trade can no longer be declined. Portfolio heat was computed across the account from each position's stop risk, so widening one position raises the total the account carries without any new position being opened. Both figures were correct when they were written and are stale afterwards, which is what makes the change hard to notice from a list of open positions.

One distinction keeps the point from proving too much. A rule written into the plan before the position exists, such as a trailing arrangement or a stated point at which the level moves, is a different object from an adjustment made while a position is open and losing. The first was part of the arithmetic that produced the size. The second was not, and it is the only one this section is about.