Technical Analysis · Chapter 3 of 12
Trend and market structure
Higher highs and higher lows, and the uncomfortable fact that the whole labelling depends on a lookback parameter you chose arbitrarily.
Market structure is the most useful framework in discretionary technical analysis and the one most vulnerable to hindsight. Both of those are worth taking seriously, and most material on the subject takes only the first.
The definitions, made mechanical
Vague definitions cannot be tested and cannot be traded consistently, so here they are in a form you could code:
- Swing high — a bar whose high is higher than the highs of the n bars either side of it. n is your lookback and it is the whole ballgame. Common choices are 2, 3 or 5.
- Swing low — the mirror.
- Uptrend — the sequence of swing highs is rising and the sequence of swing lows is rising. Both conditions. A market making higher highs on lower lows is expanding, not trending, and it behaves very differently.
- Break-out (BO) — a close above the most recent swing high, on the timeframe you are working on. Not a wick through it, not a touch: a close. This is the primitive that the next two definitions are both built from, and it is the single most load-bearing event in structural analysis.
- Break-down (BD) — the mirror. A close below the most recent swing low, on the same timeframe.
- Break of structure (BoS) — a break-out in an uptrend, or a break-down in a downtrend. That is, a BO or BD in the direction of the existing trend. Confirmation that the trend continues.
- Change of character (CHoCH) — the first BO or BD against the trend: the first close beyond the most recent swing extreme in the opposite direction. The first evidence the pattern has failed, not yet proof that a new one has begun.
The lookback problem, stated honestly
There is no correct n. With n = 2 you get many swing points and a structure that changes character constantly. With n = 10 you get few, and you will be late to every genuine reversal.
The consequence is uncomfortable and worth internalising: two competent analysts looking at the same chart can disagree about whether the trend is up, and both be applying their rules correctly. Structure is not a fact about the market; it is a transformation of the price series with a parameter.
What to do about it. Fix n in advance and never change it mid-analysis. If you change it because the current labelling is inconvenient, you have stopped doing analysis. And when you test a structural rule, test it across a range of n. A rule that works only at n = 3 and fails at 2 and 5 has been fitted, not discovered. This single check kills a great many appealing structural systems.
Trendlines, and why they are weaker than they look
A trendline connects successive swing lows in an uptrend. It is the most drawn object in technical analysis and one of the least testable, for a specific reason: you choose which points to connect. Given five swing lows there are ten possible two-point lines, and the eye reliably selects the one that has held.
The honest treatments are: use a regression channel, which is computed rather than drawn and therefore reproducible; or use a moving average as a dynamic trend reference, which is also computed; or accept that a trendline is a visualisation aid rather than a signal, and do not build a rule on it. All three are defensible. Drawing a line, watching price touch it, and calling that confirmation is not.
Trend strength: the part people skip
Knowing the direction is less than half the problem, because trend-following rules lose money in ranges and the same rule is profitable or ruinous depending on which regime it meets. Two reasonable measures of whether a trend exists at all:
- ADX — a normalised measure of directional movement. Conventionally, above 25 suggests a trend and below 20 suggests a range. Treat those thresholds as conventions rather than as findings; the useful part is the relative reading against the instrument's own history, not the absolute number.
- Efficiency ratio — net movement over the period divided by the sum of absolute bar-to-bar movement. A value near 1 means price travelled in a straight line; near 0 means it went nowhere by a long route. It is simple, has no magic constants, and is arguably more interpretable than ADX.
- Volatility-adjusted comparison — a 3% move means something completely different in a stock that typically moves 0.5% a day than in one that moves 4%. Always normalise by ATR before comparing anything across instruments. This is the most commonly skipped step in cross-sectional work and it invalidates the results when skipped.
A scenario worth working through
Situation. A stock has made a clean uptrend for four months. It has just printed a lower low for the first time. Your framework says CHoCH.
The trap. Treating CHoCH as a reversal signal and shorting. Most changes of character do not become reversals — they become pullbacks that resume, or ranges that go sideways for weeks. The base rate here is genuinely unfavourable, and this is the most common way a structure-based trader loses money in a bull market.
What the framework actually says. CHoCH removes permission to buy the dip automatically. It does not grant permission to sell. Those are different, and conflating them is the single most expensive error in structural trading.
What would change your mind. A subsequent lower high followed by a break of that low — the BoS. That is the confirmation, and it costs you the first part of the move, which is the price of not being wrong repeatedly.
The measurable version. Test it. Of all CHoCHs on your universe over ten years, what fraction were followed by a BoS within k bars? What was the expectancy of entering at CHoCH versus at BoS, net of costs, against a matched random control? That number exists, you can compute it, and it will settle an argument that people otherwise have forever.
Check yourself