WhitmanTrading

How to Spot a Divergence

To spot a divergence, first write down what counts: which indicator, how many bars, and how large a disagreement. Then look for price making a new high or low that the indicator does not confirm. Without the written rule the pattern is available on any chart at any time.

A divergence is price making a new extreme that an indicator does not confirm. It is the most persuasive pattern in technical analysis and the one most easily manufactured by looking, which makes the written definition the whole of the technique.

Before you start

A written definition of what counts, because any two series disagree somewhere. Which indicator, how many bars apart the two extremes must be, and how large the disagreement has to be.

One indicator, chosen in advance, not whichever one currently shows the pattern. Four oscillators means one of them almost always diverges.

A rule for how long you wait before the divergence is void. Without an expiry, a failed divergence never becomes a failure — it just stays open.

The steps

1. Write the three numbers down first

A range-bound stretch of price with two extremes marked.
Indicator, bar separation, minimum gap. Illustrative chart - not real market data.

Which indicator, minimum bars between the extremes, and the minimum difference in readings. Three numbers, decided when you have no position.

2. Find two comparable extremes in price

A slice of price data with two peaks compared.
Two highs, or two lows, separated by your minimum. Illustrative chart - not real market data.

A higher high and an earlier high, separated by at least your stated number of bars. Extremes closer together than that are noise being compared to noise.

3. Compare the indicator at those two points

A long-horizon price series with two diverging measures.
The indicator's readings at the same two bars. Illustrative chart - not real market data.

Not at nearby bars, at the same two. Sliding the comparison by a bar or two to make the pattern appear is where most divergences come from.

4. Check the gap is larger than your threshold

A slow-moving stretch of price with a measured difference.
A small disagreement is noise. Illustrative chart - not real market data.

A one-point difference on a bounded oscillator is noise. Your written minimum is what separates a divergence from two readings that happen not to be identical.

5. Require a structural reason as well

The first half of a price series at a tested level.
A divergence at a level is a different event. Illustrative chart - not real market data.

A divergence at a level price has already respected is worth more than one in open space. The pattern alone is a question; the level is what turns it into a setup.

6. Set an expiry

A section of a price series with a bounded window.
Without an expiry, nothing ever counts as failure. Illustrative chart - not real market data.

If price has not reacted within your stated number of bars, the divergence is void. On this site’s shared series direction runs average 2.01 bars and the longest ran 11, so a long wait is a long wait.

7. Take the stop from the chart

The first half of a price series with an invalidation level.
The pattern contains no price levels. Illustrative chart - not real market data.

Beyond the extreme that formed the divergence is the usual answer, and it has to be checked against ordinary movement rather than assumed.

How to tell it worked

All 3 numbers were written before the chart was opened.

The extremes were at least 10 bars apart, measured rather than eyeballed.

0 divergences were counted from a second indicator that was not the chosen one.

And any divergence still unresolved after your expiry was marked void, not left open.

Why it is so easy to find

A candlestick chart annotated with the round-trip cost of a switch.
Every divergence traded costs a round trip. Illustrative chart - not real market data.

Because two series that both move will disagree constantly. An indicator is a transformation of price, not a second opinion about it, so the disagreements are mostly artefacts of the smoothing rather than information.

A section of a price series drawn without volume context.
And in a thin market almost every extreme diverges from something. Illustrative chart - not real market data.

And because the pattern is identified after the fact by default. Scrolling back to find divergences always succeeds. The test is whether the rule you wrote would have found the same ones prospectively.

Regular against hidden divergence

Regular divergence is price making a new extreme that the indicator does not. It is described as a reversal signal, and it is the version almost everyone means.

Hidden divergence is the reverse: the indicator makes a new extreme that price does not. It is described as a continuation signal, which means the same underlying disagreement is read in opposite directions depending on which series made the new extreme.

Having both available is the problem. Every disagreement between two series qualifies as one or the other, so the pair covers every possible case — which means the framework can never be wrong, and a framework that can never be wrong is not telling you anything.

Keeping an honest record of them

Mark every divergence your rule finds, not only the ones you trade. The ones you skip are the control group, and without them there is no way to tell a working rule from a memorable one.

Record the date, the two bars, the readings, and what happened over your expiry window. Four fields, thirty seconds each.

After thirty entries the record answers the question the pattern cannot. How often did price react within the window, how often did it keep going, and how often did the indicator simply catch up instead.

That record is the only evidence available to you. Nobody publishes the divergences that failed, and the examples in every explanation were selected after the outcome was known — which is why a personal log of thirty is worth more than a hundred illustrations.

The original data

Of the 24,971 unique videos in research/search-study-corpus.jsonl, 99 mention divergence in the title, at a median of 9,671 views across 78 channels, and 66% of those titles are instruction-shaped. The relative strength index appears in 154 instruction-shaped titles at 4,398 and on balance volume in 26 at 15,517. The counts come from site/corpus_count.py and site/rank_howto.py.

A candlestick series with several gaps, the largest of them marked.
A gap creates an extreme with no bars behind it. Illustrative chart - not real market data.

99 videos at 9,671 across 78 channels. Substantial coverage for a pattern whose main practical difficulty — that it is trivially findable in hindsight and needs a written definition to be testable — is almost never the subject of the explanation.

A stretch of price bars cut short at a decision point.
Clear divergence, 14 bars old, price still rising. Wait longer? Illustrative chart - not real market data.

The answer to the question on that chart is that your expiry rule already answered it. Fourteen bars past a stated limit means the divergence is void — and waiting because the pattern still looks good is the specific move that turns a rule into a preference.

When it fails

The failure is the divergence found by scrolling, and it produces a perfect record in review and nothing in advance. Looking back at any chart, the divergences before major turns are obvious and striking. What is not visible in that review is the far larger number of identical-looking divergences that resolved into nothing, because nobody marks those. The pattern’s apparent reliability is entirely an artefact of which examples get remembered.

The second failure is no written definition. Every wiggle then qualifies.

A third is switching indicators. One of four always diverges.

A fourth is no expiry. A failed divergence never fails.

A fifth is trading it without structure. The pattern is a question.

And a sixth is using regular and hidden together. Between them they cover everything.

RSI divergence is the most common version. MACD divergence is the same pattern on a different indicator. And momentum indicator is the family these patterns are read from.

What I actually do

The discipline that made this usable was writing the rule before opening the chart: which indicator, how many bars between the two extremes, and how far apart the readings have to be. Without those three numbers I was finding the pattern after the turn, every time, and believing I had found it before.

— Michael Whitman

This page is educational, not financial advice. Test every idea on your own charts before risking money.