Market Structure

A break of structure and a change of character are worth more than every chart pattern with a name, for one reason: they predict something you can count. What nobody quotes alongside them is the definition of a swing they were computed from, and that definition is a dial. On the twenty bars below, the same prices contain eight structure events, or two, or none, and for four of those bars two entirely ordinary settings give opposite answers to who is in control.

Prerequisites: Lesson 25, because the swing points this lesson is about are the same prices stops rest beyond, and lesson 30, which is the second time a summary turned out to belong to a setting rather than to the market.

Twenty bars, three settings

Here is a short series, stated in full so it can be checked. The highs, bar by bar, are 100.8, 101.6, 101.1, 102.9, 102.4, 104.2, 103.5, 103.0, 105.4, 104.7, 106.8, 106.1, 104.3, 105.9, 103.6, 105.1, 102.2, 103.4, 100.9 and 101.8. The lows are 100.0, 100.7, 100.2, 101.5, 101.3, 102.8, 102.4, 101.9, 103.8, 103.5, 105.0, 104.4, 102.9, 103.9, 102.0, 103.0, 100.6, 101.4, 99.2 and 100.1. It rises for eleven bars, tops at 106.8, and falls for the rest. Any reader would call that shape without help.

Two conventions have to be said out loud first, because the definitions above leave two cases open and a reader who resolves them differently will count differently. A swing is confirmed k bars after it forms, and once a swing has been taken out it stops being a reference: the next break has to clear a swing confirmed since. And the first break in a series, when no trend has been established yet, counts as a change of character rather than a break of structure, because the character it changed was having none. Every figure below follows from those two together with k, and from nothing else.

Now run the standard rules on it three times, changing only k.

Swing rule	Swing highs	Swing lows	Structure events	Bars with a trend	Trend at bar 20
1 bar either side	8	8	8	17 of 20	Down
2 bars either side	2	1	2	12 of 20	Down
3 bars either side	1	0	0	0 of 20	No trend

Start with the bottom row, because it is the one that ought to be impossible. On a series that plainly rises and then plainly falls, the three-bar rule finds one swing high, no swing lows at all, and therefore not a single break of structure or change of character in twenty bars. It never even assigns a trend. That is not a bug in the rule; it is the rule doing exactly what it says on a series this short. A reader using it would have looked at an obvious rally and an obvious decline and had nothing to say about either.

The top row looks like the opposite failure and turns out not to be. One bar either side finds eight events in twenty bars, one every two and a half. But sort them: two are changes of character and six are breaks of structure. The sensitive rule did not invent a single spurious reversal here. What it did was mark the rally and the decline six more times, at a granularity where a two-bar pause is a swing.

The middle row is what most people actually use, and its two events are both changes of character, with no breaks of structure at all. So the two working rules agree on how many turns this series contains — two, one each way — and disagree about everything else: which bars the turns were on, and whether there were six continuations between them or none. And the honest summary of the table is that the same twenty bars hold eight structure events, or two, or zero, decided entirely by a value that appears nowhere on the chart.

Where the two settings disagree

Counting events is the mild version. The sharper question is whether two settings ever contradict each other about the state of the market, and on these twenty bars they do, for four consecutive bars.

Bar	High	Low	1 bar either side says	2 bars either side says
11	106.8	105.0	Uptrend	Uptrend
12	106.1	104.4	Uptrend	Uptrend
13	104.3	102.9	Downtrend	Uptrend
14	105.9	103.9	Downtrend	Uptrend
15	103.6	102.0	Downtrend	Uptrend
16	105.1	103.0	Downtrend	Uptrend
17	102.2	100.6	Downtrend	Downtrend

Bar 11 is the high of the whole series and both settings call it an uptrend, which is correct and useless: the top always looks like an uptrend from inside it. Then they part. The one-bar rule registers its change of character at bar 13 and the two-bar rule not until bar 17. For bars 13 to 16 the two readers are looking at the same screen and would give opposite answers to a question neither of them thinks is ambiguous.

Twelve of the twenty bars carry a trend under both rules. They agree on eight of those and disagree on four, so two ordinary settings of one number agree about who is in control two thirds of the time.

And notice what each one paid for its answer. The one-bar reader was told the trend had turned at bar 13, with the low at 102.9 — four bars and 2.3 points earlier than the other, off a high of 106.8. The two-bar reader was still being told uptrend at bar 16, and only turned at bar 17, by which point price had fallen 6.2 points from the high. That is the case for the sensitive setting, and on this series it is a good one. The case against it is not that it invented turns, because it did not. It is the six breaks of structure the other rule never reported: six further moments at which the rule said something, on twenty bars, each of them a decision that costs a spread and a commission if you act on it. Lesson 10 priced what a decision costs before you are right or wrong about it.

What k actually is

Put those two paragraphs side by side and the swing lookback stops being a preference. A small k turns early and speaks often; a large k is quiet and late. Those are the two error rates lesson 26 wrote as a pair — catching the turns that happen, and not calling turns that do not — and the swing lookback is the dial between them. Which of the two a given setting is spending is exactly what twenty bars cannot tell you, because twenty bars contain two turns and no false-positive rate can be measured from that. How to Collect a Base Rate is the appendix that does the measuring, and the third problem below is where it gets done.

Which is where the falsifiability comes back, with a condition attached. A break of structure predicts continuation, but continuation is not a claim until you say how far and by when. Fix three things in advance — the lookback, a horizon in bars, and a threshold in ATR — and the claim becomes countable: of the breaks of structure on your instrument, how many were followed by that much movement in that direction within that many bars? Leave any of the three loose and you have the triangle back, in better clothing.

One thing this lesson does supply, and it settles a debt. Lesson 21 said a stop belongs where the reason for the trade is gone, and left the level to be named later. The swing point is that level: if the trade was taken on a break of structure, the reason for it is gone when the swing that defined the break is taken out the other way. That is where the stop goes, and lesson 21 already priced how far beyond it to sit.

What this does not settle

Which lookback is right. Nothing here says one, two or three, and the exercise was not designed to produce a winner. The three-bar rule found nothing on twenty bars, which says something about twenty bars as much as about the rule; on two hundred it would find plenty. What the table settles is that the number is load-bearing and unstated, not that any particular value is correct.

That the fractal is the only definition. It is the common one and it is the one this lesson tested. Zigzag rules use a percentage retracement instead, and the turning-point algorithms economists use to date recessions impose minimum phase and cycle lengths on top. Every one of them has at least one number in it that has to be chosen, which is the finding — not that the fractal in particular is fragile.

That the disagreement is always four bars. It is four bars on this series, at these two settings. Another series gives another answer, and nothing here says which way. The point that carries is that the disagreement is not zero and is not visible: neither reader has any way of knowing, from their own chart, that the other exists.

That a break of structure predicts continuation. This lesson has established that the claim can be tested, which is genuinely more than the rest of the chart-pattern literature offers, and it has not tested it. The number that would settle it is a base rate over your own instrument with the lookback, the horizon and the threshold fixed in advance, and the third problem is that experiment. Until it exists, structure is a description of what price has already done, stated in a vocabulary that at least admits it could be wrong.

That structure is about institutions. Nothing in the definitions mentions who is trading, and nothing needs to. A higher high followed by a higher low is a fact about a price series; it happens in markets with no institutions in them at all. The story about smart money reversing direction from a zone is added afterwards, is not measured by anything above, and would not change a single row of either table if it were true.

A break of structure is a claim, which puts it ahead of almost everything else drawn on a chart. It is a claim about a swing, and a swing is a claim about a number nobody says out loud.

Problems

1. Find out what your own lookback is. You already have one, because you have been marking swings. Take thirty bars of your own instrument, mark the swing highs and lows the way you normally do, then run the fractal rule at one, two, three and five bars and see which set of marks matches yours. Most people come out between two and four and are surprised by the spread within their own chart. If none of them matches, your rule involves something you have not written down, which is problem 1 of the appendix in a different costume.
2. Re-mark the same chart at two settings and count the contradictions. Take sixty consecutive bars. Run the structure rules twice, at your own lookback and at one bar tighter, confirming each swing the right number of bars after it forms. For each bar, write down what each setting says the trend is. Count the bars where they disagree. On the twenty above the answer was four in twelve; yours will differ, and the number is how much of your conviction about the current trend is really a conviction about your setting.
3. Test the claim, having fixed all three numbers first. Write down your lookback, a horizon in bars and a threshold in ATR, before looking at anything. Then over thirty consecutive breaks of structure, count how many were followed by that much movement in that direction within that many bars. That is the base rate for continuation on your instrument, it is the only thing that makes a break of structure a reason rather than a label. How to Collect a Base Rate is how the count is kept honest.

Sources. Andrew W. Lo, Harry Mamaysky and Jiang Wang, “Foundations of Technical Analysis: Computational Algorithms, Statistical Inference, and Empirical Implementation” (Journal of Finance, 2000), for the step this lesson is about: before any chart pattern can be counted, a local maximum has to be defined, and their definition carries a smoothing parameter that has to be chosen rather than observed. Gerhard Bry and Charlotte Boschan, Cyclical Analysis of Time Series: Selected Procedures and Computer Programs (National Bureau of Economic Research, 1971), for the same problem solved carefully in another field: dating the turns of a cycle needs explicit minimum lengths for a phase and for a full cycle, written into the algorithm because they cannot be read off the data. Adrian Pagan and Kirill Sossounov, “A Simple Framework for Analysing Bull and Bear Markets” (Journal of Applied Econometrics, 2003), for what happens when that machinery is pointed at stock prices, including how the dating of the phases moves when the minimum lengths are changed.

The next lesson takes the other half of the folklore this one left alone. If a change of character says control has changed hands, the obvious question is where the hands were, and the answer on offer is a zone: the last candle before the move. Lesson 33 is about what that zone is, what displacement adds to it, and how many of them a chart actually produces once the definition is written down.

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