Sweeps, Beyond the First

A sweep is price trading past a level and coming back. Three settings decide whether one happened at all: which levels count, whether a wick past is enough or the close has to stay inside, and how soon back inside still counts. On the sixty bars below, ordinary values of those three find four sweeps, or fifteen, or forty-six — and neither of the two levels that reached a second sweep stopped there.

Prerequisites: Lesson 25, for what is actually resting beyond a level and why price runs through it with nobody arranging anything, and lesson 32, whose swing rule decides which levels exist and whose bars this lesson is still using.

Sixty bars, three dials

The series is the one from the last three lessons. Lesson 34 published its sixty closes and lesson 33 published the highs and lows of the first twenty bars; a sweep needs highs and lows for all sixty, so here are the missing forty. They were not chosen. Each bar from 21 onward spans from the previous close to its own close, plus an overshoot cycling through 0.3, 0.1, 0.5, 0.2 and 0.4 above and 0.2, 0.4, 0.1, 0.5 and 0.3 below, so the ranges are fixed by a rule written before anything was counted.

The highs of bars 21 to 60 are 101.9, 100.6, 100.5, 99.3, 99.5, 99.8, 99.6, 99.7, 99.8, 100.4, 100.3, 100.0, 100.8, 101.7, 102.5, 103.3, 103.5, 103.9, 103.3, 102.8, 102.0, 102.2, 103.6, 104.6, 105.2, 105.9, 106.7, 107.1, 106.9, 107.1, 107.0, 106.8, 107.4, 107.1, 107.0, 107.4, 107.2, 107.3, 107.2 and 107.4. The lows are 100.3, 99.6, 99.0, 97.7, 97.9, 98.9, 98.4, 98.7, 98.7, 99.3, 99.3, 99.1, 99.8, 99.8, 101.2, 101.9, 102.6, 103.0, 101.9, 101.0, 101.1, 101.3, 102.0, 102.6, 104.1, 104.6, 105.2, 106.0, 105.6, 105.6, 105.7, 105.5, 105.8, 106.1, 105.5, 105.6, 106.4, 105.9, 105.5 and 106.2. The whole series runs from 97.7 at bar 24 to 107.4 at bar 53, and the mean bar range is 1.35.

Now mark the swing levels, and from the bar each is confirmed, watch for excursions past it. An excursion is any bar whose high goes above a swing high or whose low goes below a swing low. It is a sweep if a close comes back inside within two bars, and a failure if none does — and a failure retires the level, because a level price has closed beyond and stayed beyond is not a level any more. Three details decide the counts and are worth stating, because a reader who resolves them differently will not reproduce the table. Beyond and inside are both strict, so a close sitting exactly on the level has not come back. An excursion runs until it is reclaimed and takes those bars with it, so the next excursion of the same level begins after the reclaiming bar rather than on it. And the clearance recorded for an excursion is its deepest point, not its first bar’s.

Swing rule	Levels	Excursions	Sweeps	Never closed beyond	Levels swept twice or more
1 bar either side	32	62	43	18	10
2 bars either side	15	22	15	6	2
3 bars either side	5	10	8	5	1

The fourth column is the sweep count under the ordinary reading and the fifth is the same count under the strict close-not-a-wick reading, which keeps only the excursions that never closed beyond at all. At two bars either side that is six sweeps where the loose reading found fifteen. Tighten the reclaim window to the same bar and the loose count falls to five; open it to three bars and it rises to eighteen. Run the tightest combination available here — three bars either side, reclaim on the same bar, closes never beyond — and sixty bars contain four sweeps. Run the loosest and they contain forty-six.

The sixth column is the one this lesson is named after, and it moves as hard as the rest. Ten of the thirty-two levels were swept twice or more under the loose swing rule; two of fifteen under the middle one; one of five under the tight one. Whether the double sweep is everywhere or nearly absent is settled before any price is looked at.

Beyond the first

The folklore has a mechanism, and half of it is sound. A first sweep takes the stops that were resting under an obvious level. Everyone who traded that first sweep then places new stops under the low it made, so a second sweep takes a second, freshly built cluster — and lesson 25’s account of where stops go supports that much. The conclusion drawn from it is that the second sweep is the real one and the first is bait.

The problem is that the rule contains no instruction to stop at two. At two bars either side this series has exactly two levels that reached a second sweep, and neither of them stopped there: the swing high at bar 6, at 104.2, was swept four times, and the swing high at bar 11, at 106.8, was swept eight. Take the second one. Price stayed below 106.8 for thirty-six bars — the highest bar in between is bar 47 at 106.7, a tenth of a point short — and then swept it at bars 48, 49, 50, 51, 53, 55, 56 and 59.

Their clearances past the level, in order, are 0.30, 0.10, 0.30, 0.20, 0.60, 0.20, 0.60 and 0.60. The largest is 0.60 on a series whose mean bar range is 1.35, so the deepest of the eight went less than half a bar past the level. This is not eight liquidity events stacked on one price. It is a flat top and a bar size, and the definition has no way to say so.

Which is what makes “wait for the second” unusable as stated. It is a rule you can only apply after you know how many there were, and you know that only at the end. At bar 49 you are being told to act; at bar 50 the same rule tells you the thing you acted on was itself the bait for the next one. Every sweep in a sequence is somebody’s second sweep.

The invalidation the market hands you

There is one genuinely useful thing in the wreckage, and it survives all of the above. Once an excursion has happened, its extreme is a price the market produced rather than a number you picked. If your reason for being in the trade is that the level held, then the level not holding is the extreme being exceeded, and the stop writes itself.

What that stop costs is the part nobody prices. Take the fifteen sweeps at two bars either side and sort their clearances: 0.1, 0.2, 0.2, 0.3, 0.3, 0.4, 0.6, 0.6, 0.6, 0.6, 0.9, 1.2, 1.3, 1.7 and 2.6. A stop placed a given distance beyond the level survives every sweep whose clearance was shorter than that distance. Hold the target fixed at four points — a little under half the range of the whole series — so that the only thing moving is the stop.

Stop beyond the level	Sweeps it survives	Payoff to a four-point target	Break-even hit rate
0.25	3 of 15	16.0 to 1	5.9%
0.50	6 of 15	8.0 to 1	11.1%
1.00	11 of 15	4.0 to 1	20.0%
1.50	13 of 15	2.7 to 1	27.3%
2.00	14 of 15	2.0 to 1	33.3%
2.75	15 of 15	1.5 to 1	40.7%

Read it as an exchange rate. Surviving five more of the fifteen excursions, from the half-point stop to the one-point stop, halves the payoff and nearly doubles the hit rate the trade has to deliver. Surviving all fifteen leaves 1.5 to one against the 16 to one the tightest stop showed, which is nine tenths of the payoff given away. Nothing here says which row is right, and the table cannot say, because the second column is a description of fifteen excursions that happened and the fourth is a requirement about the future.

Two things it does settle. Waiting for a second sweep and then placing the stop beyond that one is not a better version of the same trade; it is a trade with a different payoff, and the difference is the width of the second excursion. And the seven excursions in the same sample that never came back are losses at every row of that table, so the second column is not a win rate and must not be read as one.

Problems

1. Count the levels before you count the sweeps. Take sixty consecutive bars from your own instrument and mark the swing highs and lows at one bar either side, then again at three. Write down only the two level counts. That ratio is the size of the universe each setting is granting itself, and every sweep either setting can ever find is a sweep of one of those levels. Most people have never seen their own two numbers side by side.
2. Run the close rule in both directions. On those same bars, find every excursion past a live level. Sort them into three piles: never closed beyond, closed beyond and came back within two bars, and closed beyond and stayed. The first pile is what the strict close-not-a-wick rule calls a sweep; the first two together are what the loose reading calls a sweep; the third is the level being gone. Note the size of the middle pile, because that is the entire disagreement between two people who both think they are using the same word.
3. Measure your own clearances, then price the stop. Take thirty excursions that came back and record how far past the level each one went, in points and in units of the mean bar range. Sort them. Read off the distance that covers three quarters of them, and divide your usual target distance by it: that is the payoff a stop beyond three quarters of your own sweeps actually leaves you, and one divided by one plus that payoff is the hit rate it would have to deliver. How to Collect a Base Rate is how the count is kept honest.

Sources. Carol Osler, “Stop-Loss Orders and Price Cascades in Currency Markets” (Journal of International Money and Finance, 2005), for the mechanism underneath the word: a triggered cluster propagates into the orders beyond it, which is why the excursion is fast and why it ends, and none of it requires an author. Andrew W. Lo, Harry Mamaysky and Jiang Wang, “Foundations of Technical Analysis” (Journal of Finance, 2000), for the first dial stated as a technical problem rather than a preference: before any pattern can be counted, a local extremum has to be defined, and the definition carries a parameter that is chosen and not observed. Lawrence Harris, “Stock Price Clustering and Discreteness” (Review of Financial Studies, 1991), for why the levels sit where they do: prices themselves cluster on round increments, which is a documented property of transacted prices and needs no story about who wanted them there.

That closes the module. Eleven lessons on reading the auction, and nearly every one of them ended in the same place: the reading depends on a setting nobody displays. The next module changes the question. Instead of asking what a pattern means, it asks when a pattern means anything at all — because the same rule that pays in one market condition loses in another, and telling the two conditions apart is a measurement rather than an impression. Lesson 36 begins with the crudest version of that measurement and the reason it is not as crude as it looks.

Educational only. Trading involves substantial risk of loss. Not financial advice. Past performance does not guarantee future results.