Skip to content

Delta divergence depends on which swings you compare

By Author of ORDER FLOW — Market Mechanics

A divergence is usually described as something you spot. It is more accurate to say it is something you define, because declaring one requires picking two points, and nothing in the data picks them for you.

The difference changes what a divergence can be used for, and it explains why the same chart produces confident and contradictory readings from different people looking at it at the same time.

The comparison needs two points

The standard formulation is that price makes a higher high while cumulative delta makes a lower high, so the advance is being carried by less aggression than the previous one.

The reasoning is sound. But read the sentence again and notice how much it assumes: it requires a previous high, and it requires agreement about which previous high. Change the reference and the conclusion can invert.

Reading A · highs 1 and 3 Price CVD no divergence Reading B · highs 2 and 3 Price CVD divergence The same data. The divergence appears or vanishes depending on which two highs are chosen as reference.
Schematic of one price and CVD series read against two different pairs of reference highs.

Between those two readings the data is identical. Only the earlier high used as the benchmark changed, and that was a decision made by whoever was looking.

Which highs exist is also a setting

It goes one level deeper than choosing between visible highs, because which highs count as highs is itself parameterised.

Divergence indicators generally identify swings with a zigzag, and a zigzag needs a deviation threshold, in ticks or as a percentage of session range, that decides how much retracement a wiggle needs before it registers as a swing at all. This is documented plainly in the platforms that offer it, usually alongside a note that the setting controls how many signals you get.

Coarse swing threshold 3 swing highs Fine swing threshold 7 swing highs Identical data. A deviation setting decides how many highs exist to compare.
Schematic of the same price series under a coarse and a fine swing threshold. Swing counts are illustrative.

Some tools go further and offer a choice between requiring both the price and the delta condition or accepting either one, described openly as a trade-off between signal frequency and conviction. The vendors present all of this as configurability. What follows from it is that the number of divergences available in a session is a number you chose.

Divergences stack through trends

There is a second problem, independent of parameter choice, and the literature on order flow is openly split on it.

Some sources describe delta divergence as among the most reliable leading signals available and treat a reversal as probable once one appears. Others note, correctly, that divergences can repeat several times while a trend continues, and conclude that a divergence is a condition rather than a trigger.

Price CVD divergence 1 divergence 2 divergence 3 Three divergences, one uninterrupted trend. None of them worked as a trigger.
Schematic of three successive divergence conditions forming during a single continuing trend.

The second description matches how trending sessions behave. An advance that keeps going on progressively less aggression is common, because once resistance has thinned, less aggression is required to move price at all. Weakening effort with a continued result can be exactly what a low-friction trend looks like.

The opposite reading has its own trap. On a break of the Initial Balance, the candle with the most delta can be the start of a trend or its climax, and which one it was only shows in the candles that follow.

Anyone who has taken the first divergence in a sequence knows the outcome. The ones who wait are treating it as a different kind of object: a condition to watch.

What makes the reading defensible

Divergence remains usable with a short, specific discipline.

Fix the reference before the comparison. The earlier high you measure against should be one that mattered structurally (a session high, the edge of a range you were already watching, the level your hypothesis rested on), and it should have been chosen before you went looking for a divergence. A reference selected afterwards, because it produces the divergence, is unfalsifiable by construction. The same applies to reference prices that look like facts: even yesterday's close comes in more than one version, and the choice between them has to be made before the numbers are read.

Know your threshold. If a tool is marking swings for you, the deviation setting is part of your analysis whether you have looked at it or not.

Check volume as well as the line. Price rising on falling CVD has at least two incompatible explanations: aggression being absorbed by a passive participant, or aggression simply thinning out in a market where nobody is offering. Total volume separates them, and the CVD line alone does not. That distinction is easier to see on a footprint, where the volume sits at the level it traded at.

Treat it as a reason to look. In a normal NQ session there are enough local highs to build whichever divergence you want in whichever direction you want. The ones worth anything are the ones you could have specified in advance.