WhitmanTrading

MACD Divergence: You Choose the Peaks

MACD divergence is a mismatch between a new price extreme and the corresponding reading on the moving average convergence divergence indicator. Which two peaks get compared is chosen by the observer, and in a sustained trend the mismatch recurs repeatedly without a reversal following.

How it works

A candlestick chart of the site's shared price history. The headline on the chart reads: Price makes a high the indicator does not.
Price makes a high the indicator does not. Illustrative chart - not real market data.

Price makes a higher high. The MACD does not. That mismatch is a bearish divergence, read as the advance losing force even though price is still rising. The bullish version is the mirror image at lows.

A gently rising stretch of the long price series. The headline on the chart reads: It is a comparison between price and a lagging summary of price.
It is a comparison between price and a lagging summary of price. Illustrative chart - not real market data.

What is being compared is price against a lagging summary of price. The moving average convergence divergence indicator is the difference between two exponential averages, and those averages are weighted toward older data by construction.

So a mismatch between them is partly a mechanical consequence of the lag. If a second advance is steeper but shorter than the first, the smoothed measure can fall behind even though nothing about participation changed. That does not make every divergence meaningless — it means the tool produces some of them on its own.

The part that is entirely your decision

A flat but volatile stretch of the long price series. The headline on the chart reads: Which two peaks you compare is entirely your choice.
Which two peaks you compare is entirely your choice. Illustrative chart - not real market data.

A divergence requires two peaks, and nothing specifies which two. How far apart? How prominent must each be? Must they be consecutive swing highs, or can you skip one?

Change any of those answers and the divergence appears or disappears. On a chart with dozens of local peaks — this site’s shared history has 286 directional runs averaging 2.01 bars — the number of possible pairs is enormous, and some pair almost always diverges.

The discipline that fixes it is marking the peaks before checking the indicator. Define what counts as a swing high — a minimum number of bars either side, a minimum size — write it down, and then look. The number of divergences on your chart will fall sharply, and the ones that remain were not selected by knowing the answer.

A strongly rising stretch of the long price series. The headline on the chart reads: In any strong trend the divergence appears and keeps appearing.
In any strong trend the divergence appears and keeps appearing. Illustrative chart - not real market data.

In a sustained trend, divergence is close to permanent. The first leg is the sharpest, so every subsequent leg produces a lower reading on a momentum measure while price goes on making highs. The indicator is functioning correctly and the signal fires continuously.

A calmly advancing stretch of the long price series. The headline on the chart reads: Being early repeatedly is indistinguishable from being wrong.
Being early repeatedly is indistinguishable from being wrong. Illustrative chart - not real market data.

“It was early” is the defence, and at the level of an account there is no difference. Four losing trades and a fifth that works is the same equity path whether the first four were early or wrong. The only thing that separates them is the story told afterwards.

In practice

A flat, quiet stretch of the long price series. The headline on the chart reads: Hidden divergence is the same tool pointing the other way.
Hidden divergence is the same tool pointing the other way. Illustrative chart - not real market data.

Hidden divergence — a higher indicator low against a higher price low — is read as continuation rather than reversal. That gives the same tool two opposite readings depending on which pattern you match.

A framework with a reading for every configuration is worth being careful with, because it cannot be contradicted by anything the chart does.

A declining stretch of the long price series. The headline on the chart reads: And it correlates 0.71 with the relative strength index anyway.
And it correlates 0.71 with the relative strength index anyway. Illustrative chart - not real market data.

On this site’s shared history the bar-to-bar changes of the moving average convergence divergence histogram and a 14-period relative strength index correlate at 0.71. So confirming one divergence with the other is confirming a reading against a close relative of itself — the pattern the confluence page treats in full.

A candlestick chart with a volume histogram beneath it, with the volume histogram emphasised. The headline on the chart reads: Participation is the one check that is not derived from price.
Participation is the one check that is not derived from price. Illustrative chart - not real market data.

Volume is the genuinely independent check. A second high on materially lighter participation is a different observation from a second high with a lower oscillator reading, because it uses an input the oscillator does not have.

A long-horizon candlestick view of the same price series. The headline on the chart reads: On a higher timeframe the two peaks are one bar.
On a higher timeframe the two peaks are one bar. Illustrative chart - not real market data.

Aggregate the chart and the two peaks merge into one bar. The divergence exists at the resolution you chose to look at, which is worth knowing before treating it as a structural feature.

A candlestick series containing several opening gaps, with the largest opening gap marked. The headline on the chart reads: A gap creates a divergence with nobody diverging.
A gap creates a divergence with nobody diverging. Illustrative chart - not real market data.

A gap can manufacture one outright. Price jumps to a new high; the smoothed indicator, weighted toward older bars, does not follow proportionally. The divergence is arithmetic from a closed market.

A declining stretch of the long price series, with the entry price and the level at which a stop would trigger drawn as horizontal lines. The headline on the chart reads: The stop has to come from structure, not from the histogram.
The stop has to come from structure, not from the histogram. Illustrative chart - not real market data.

The signal produces no price. A stop has to come from the recent swing extreme or a volatility multiple, which means the divergence supplies the idea and structure supplies the risk — and the two can be far apart.

A candlestick chart of the site's shared price history, annotated with the round-trip cost. The headline on the chart reads: And each attempt costs a share of a bar.
And each attempt costs a share of a bar. Illustrative chart - not real market data.

Each attempt costs 2% of a typical bar’s range in round-trip costs on this history, and a trend that produces four divergences before turning charges four times for one eventual idea.

A 72-bar candlestick section of the shared price history. The headline on the chart reads: The divergence is between two of your own drawings.
The divergence is between two of your own drawings. Illustrative chart - not real market data.

And nothing diverged in the market. The order book does not contain a momentum reading. Two lines you drew moved differently.

What MACD divergence is not

It is not a reversal signal. It is a mismatch that sometimes precedes reversals and frequently precedes continuation.

It is not independent confirmation of anything computed from price. At 0.71 change correlation with the relative strength index, a second divergence is close to the same divergence.

It is not defined. Peak selection has no rules, which makes the pattern findable at will.

And it is not the same as a crossover. A crossover is an event on the indicator alone; a divergence is a comparison between the indicator and price.

When it fails

A sideways, range-bound candlestick series. The headline on the chart reads: In a range a divergence is available on every leg.
In a range a divergence is available on every leg. Illustrative chart - not real market data.

The expensive failure is the persistent trend. Divergence after divergence, each one a reason to fade a move that keeps going. Every large advance in history contains a sequence of these, and the last one gets remembered as the signal.

The second failure is peak selection after the fact. With the outcome known, a pair of peaks that diverges is always findable, which is why teaching examples are so convincing.

A third is the missing structural stop. A divergence trade without a defined invalidation point becomes a position held on a story, and the story survives adverse movement indefinitely.

A fourth is stacking divergences across oscillators. Two divergences on tools correlating at 0.71 is one divergence with a second opinion from a relative.

And a fifth is trading it against the higher timeframe. A divergence on a five-minute chart inside a daily uptrend is a small counter-trend bet, and calling it a reversal signal overstates what it can possibly be about.

The original data

Two figures from this site’s shared 576-bar history bear on this page: 286 directional runs with a mean length of 2.01 bars, and a bar-to-bar change correlation of 0.71 between the moving average convergence divergence histogram and a 14-period relative strength index. Both are in research/series-measurements.json, produced by site/measure_series.py.

A 72-bar window of the shared price history, cut short at the decision bar. The headline on the chart reads: Third divergence in a row. Is this the one?
Third divergence in a row. Is this the one? Illustrative chart - not real market data.

The run-length figure is the one that matters for divergence specifically. Direction changing every two bars on average means hundreds of local peaks on any chart of reasonable length, and hundreds of peaks means an enormous number of available pairs. A signal that requires you to select two points from hundreds, with no rule for selecting them, will be found on any data — including data generated at random. Write the swing-high rule down first. That single change does more for divergence trading than any refinement of the indicator settings.

MACD explains what the indicator is computing. RSI divergence is the same idea on the closest relative, at 0.71 change correlation. And MACD histogram covers the specific display most divergences are read from.

What I actually do

Divergence is the thing I have argued about most and measured least, until I made myself mark the peaks before the outcome instead of after. Doing it in that order turned a reliable-feeling signal into a list of times I was early, and the early ones cost the same as the wrong ones.

— Michael Whitman

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