Chart tools · Lines
Trendlines: What Two Points Actually Claim
Two points fix a line, and the analyst picks both. Then the chart's scale decides what a straight line even means, so the same two points give two different lines, and a break of one is not a break of the other.
Two choices, one line
A trendline is the most discretionary tool on a chart, and the discretion arrives in two separate places. The first is well known and endlessly argued about: which two points. The second is rarely mentioned and is entirely objective: which scale.
The second is the more interesting one, because it can be settled by arithmetic. A straight line on a linear chart rises by a constant number of points per session. A straight line on a logarithmic chart rises by a constant percentage. Over a period in which price changes materially, those are different paths through the same two anchors.
That is the case for stating the scale on any line you intend to rely on, and for preferring the logarithmic version whenever the price level has moved much. A constant percentage is the more plausible description of price behaviour, and it is the only one that makes two different periods comparable.
The two-point problem
A line through two points passes through those points by construction. It has demonstrated nothing. Everything a trendline can be said to have shown lies in what happened after the second anchor, which means a chart with a beautifully fitted line through two old lows contains no evidence at all until price returns to it.
The convention of requiring a third touch is a reasonable discipline and it is not a statistical criterion. On any chart of a few hundred sessions there are a great many candidate pairs, and finding a line with three touches among them is not difficult, the same multiplicity problem the method page demonstrates with parameter sweeps.
Four decisions to fix before drawing
| Decision | The options | Why it matters |
|---|---|---|
| Scale | Linear or logarithmic | Decides what "straight" means. Over a large price change the two lines diverge substantially, as the figure shows. |
| Anchor type | Highs and lows, or closes | Extremes include single unrepresentative prints; closes ignore the range. Switching between them mid-analysis rescues lines that should have broken. |
| Which points | A judgement, or a mechanical rule over a stated lookback | The only decision that can turn the tool into something testable. "The obvious swing" is chosen after seeing the outcome. |
| Break rule | A close beyond, a close beyond by a fraction of ATR, or a break held for n sessions | Each gives a different date and a different count of false breaks. Without one fixed in advance, the break is decided retrospectively. |
The fourth row is where average true range earns its place: "a close more than half an ATR beyond the line" is a rule that means the same thing on a quiet instrument and a volatile one, while "a close beyond the line" means something different on each. That is the general pattern in this reference, a distance expressed in the instrument’s own units travels, and a distance in points does not.
Why horizontal levels are the stronger tool
Both a trendline and a horizontal level can become self-fulfilling: participants watch them, orders accumulate, and the level becomes a genuine feature of the book. The difference is how likely it is that everyone is watching the same one.
A round number or a previous high is identified identically by everybody. A sloping line requires agreement on two anchor points and a scale, which is a much weaker coincidence. So the mechanism that gives drawn levels whatever validity they have operates strongly for horizontal lines and weakly for sloping ones, which is a reason to treat the two as different tools rather than as variants of one, and to prefer the horizontal version where either would do.
The mechanical alternative
If the value of a trendline is the record of a decision, the honest improvement is to remove the decision. Fit a regression to a stated window and the line is reproducible by anyone with the same data; connect the extreme points of a fixed lookback and the same is true. A regression channel is precisely that tool, and it is the reason the chart-tools hub places it further from the discretionary end than a hand-drawn line.
What that costs is the thing people value about drawing: the ability to incorporate a judgement about which part of the history matters. That judgement may well be worth having. It is simply not evidence, and a line that came from it should be described as a decision rather than as a finding.
Frequently asked questions
What does a trendline claim?
That price has been changing at a roughly constant rate, and that the rate is worth extending forward. Both halves are assumptions rather than findings. The line is fixed by two points the analyst selects, so the claim is partly the analyst’s, and it is the most discretionary of the common chart tools for exactly that reason.
Why does the chart scale change the line?
Because a straight line means different things on the two scales. On a linear chart it is a constant number of points per session; on a log chart it is a constant percentage per session. Over a period in which price doubled, those are entirely different paths, the figure on this page draws both through the same two points and they end far apart. A trendline is therefore not reproducible unless the scale is stated.
Which scale should be used?
Log, for anything spanning a substantial price change, because a constant percentage is the more plausible description of how prices move and because it makes different periods comparable. Linear is defensible over short windows where the price level barely changed, and there the two are nearly identical anyway. The one indefensible option is not saying which was used.
How many touches make a line valid?
Two fix the line and the third is the only one that is a test of it, which is worth stating plainly: a line drawn through two points passes through those points by construction, so any claim about it rests entirely on what happened afterwards. The convention of requiring three touches is a reasonable discipline. It is not a statistical criterion, and with enough candidate pairs on a chart, three touches are not hard to find.
Should the line touch highs, lows, closes or bodies?
State which and be consistent, that is the whole answer. Extremes include single unrepresentative prints, especially on thin instruments; closes are more robust and ignore the range entirely. Neither is right, both are used, and switching between them mid-analysis is how a line survives evidence that should have broken it.
What counts as a break?
Whatever you decided in advance, which almost nobody does. A close beyond the line, a close beyond it by some fraction of average true range, a break held for two sessions. Each gives a different date and a different count of false breaks. Without a rule fixed beforehand, "the line broke" is a judgement made after seeing what followed, and the observation carries nothing.
Do trendlines have a mechanism behind them?
A weaker version of the one behind horizontal levels. A widely watched line attracts orders, so it can become a genuine feature of the book, but a sloping line requires everyone to have chosen the same two points and the same scale, which is far less likely than agreement on a round number or a previous high. That is the substantive reason horizontal levels are the more defensible of the two, and this page treats them as different tools rather than variants.
How would a trendline be tested properly?
By making the drawing mechanical: fit a regression to a stated window, or connect the extreme points of a fixed lookback. Then the line is reproducible, the break rule can be specified, and the whole thing can be measured across a long sample with the failures counted. A regression channel is essentially that — the tool that removes the analyst from the drawing — and its output can be argued about on the data rather than on the chart.