Divergence — when price and momentum disagree
The answer
Divergence is when price makes a new extreme and a momentum indicator does not. It means the new high was reached more slowly than the previous one, or the new low with less force than the previous one.
It is a statement about the speed of a move, not about its direction. A trend can slow down for months and keep going.
Why this costs you money
Divergence is the most seductive pattern in technical analysis, because when it works it looks like foresight. You spotted the top before it happened.
Here is the loss, and it is a large one.
A stock is in a strong uptrend. It makes a new high, and RSI makes a lower high. You see bearish divergence. You sell your position, or you short. The stock rises another 9% over 6 weeks, making 3 more new highs, with 3 more lower RSI highs each time. Every one of them was a valid divergence. Every one of them was correct about the deceleration and wrong about the outcome.
You have exited a good position early and then paid to re-enter higher, or you have shorted a rising market and been removed.
The second version of the loss is subtler and it affects your judgement permanently. Divergence is only identifiable after both pivots exist. You need a high, then a pullback, then a second high, before you can draw the line. On the right edge of the chart, in real time, you do not know whether the current high is the second pivot of a divergence or the first pivot of something else. In hindsight, every top has a divergence before it, and you will find them effortlessly. That produces a false memory of a reliable pattern, and false memories are what people trade on.
If you have never counted the divergences that failed, you have no idea what the success rate is, and you are certain it is high.
How it works
Take RSI as the example, because it is the most common. RSI compares the average size of up moves to the average size of down moves over the last 14 bars, and expresses the result on a scale from 0 to 100. It was introduced by J. Welles Wilder Jr. in 1978.
Now suppose a stock rises from 100 to 120 over 10 days, in daily jumps of about 2%. RSI goes high. It pulls back, then rises from 112 to 124 over 20 days, in daily jumps of about 0.6%.
The second move made a higher price high — 124 beats 120. It also produced a lower RSI, because the average daily gain was smaller.
That is the whole mechanism. There is nothing mysterious in it. Divergence is arithmetic detecting that the recent gains are smaller than the earlier gains.
The 4 kinds
| Kind | Price does | Indicator does | Suggested reading |
|---|---|---|---|
| Regular bearish | Higher high | Lower high | Uptrend is slowing |
| Regular bullish | Lower low | Higher low | Downtrend is slowing |
| Hidden bearish | Lower high | Higher high | Downtrend likely to continue |
| Hidden bullish | Higher low | Lower low | Uptrend likely to continue |
The 2 "regular" kinds are the ones everybody means when they say divergence. The 2 "hidden" kinds are continuation patterns, and they are worth knowing mainly because they explain why divergence is so easy to find: with 4 named variants, almost any pair of pivots on any chart can be labelled as one of them.
That last point is not a joke. Be suspicious of any pattern framework that has a name for every possible outcome.
Which indicators
Divergence can be drawn using RSI, MACD, Stochastic, CCI or OBV. Three of those 5 are computed from closing prices in almost the same way, so choosing between RSI, Stochastic and CCI changes very little. OBV is the one genuinely different choice, because it uses volume, which is an input the others do not have. Article 18 explains why that distinction matters more than which oscillator you prefer.
What it tells you, and what it does not
It tells you the current leg is being achieved with less force than the previous one. That is true, useful and computable.
It tells you the market is drawing in fewer or less aggressive buyers at these levels than it did before. That is a reasonable interpretation of the arithmetic.
It does not tell you the trend will reverse. Deceleration is not reversal. A car slowing from 100 to 60 is still moving forward. Trends routinely decelerate, consolidate sideways, and then accelerate again, producing a divergence that resolves upward.
It does not tell you when. This is the practical killer. Even in the cases where divergence correctly precedes a reversal, the gap between the signal and the turn can be 3 days or 4 months. A signal without a timeframe cannot be traded, because the stop that survives 4 months is far too wide to be worth taking.
It does not survive new data. If price makes another new high and the indicator goes with it, the divergence you drew is gone. Traders call this a "disappearing divergence". It is not a special case. It is the normal fate of most of them.
And it is not a trigger. This is the one professional consensus in an area with very little consensus: divergence is a warning to prepare, and price must confirm before you act.
The decision rule
Divergence changes what you look for. It never places an order by itself.
When you see regular bearish divergence: do not sell. Do 3 things. Tighten your stop to a level based on ATR rather than on hope. Stop adding to the position. Write down the price level that would confirm the warning — usually the low of the most recent pullback.
Act only when price confirms. For bearish divergence, confirmation is price closing below that recent pullback low. Until that happens, the trend is intact and the divergence is a description of speed.
Discard the divergence if price makes a new extreme with the indicator. The warning has been cancelled by the market. Do not keep it in your head.
Ignore divergence entirely in the strongest trends, when ADX is above 40. That is the condition in which divergence fires most often and works least often, because a very strong move cannot keep accelerating and must therefore produce lower indicator highs almost by definition.
That last line is worth reading twice. In a powerful trend, divergence is close to automatic. An indicator that is bounded at 100 physically cannot keep making higher highs forever while price does. The pattern appears because of the shape of the scale, not because of anything the market decided.
Try this now
Five minutes, on a chart you already care about.
- Open a daily chart of 1 stock or index you follow, set to 1 year. Add RSI with the default period of 14.
- Working from left to right, find the first 3 clear examples of regular bearish divergence: 2 price highs where the second is higher, with 2 RSI highs where the second is lower. Mark the date of the second high in each case.
- For each of the 3, look at what happened over the next 20 trading days. Write down whether price fell by more than 3%, or whether it rose.
- Now do the same for the first 3 examples of regular bullish divergence — lower price low with higher RSI low.
- Finally, and this is the step that teaches the most: find the 2 largest falls on the chart. Look at what RSI was doing in the 10 days before each one started. Was there a divergence?
What you should see. In step 3, expect a mixed result. On most charts, 1 of the 3 resolves the way divergence is supposed to, 1 goes nowhere, and 1 continues straight up. That is the honest base rate, and it is nothing like the impression you get from articles that show only the examples that worked.
In step 5, expect something that will change how you use this pattern. On many charts, the largest falls of the year were not preceded by a clean divergence. The sharpest declines often come from a market that was still accelerating upward, and they are caused by information, not by exhaustion.
Put those 2 findings together and you have the correct weight to give divergence: a modest warning that is right some of the time, and absent before many of the events you would most want to be warned about.
Three real cases
1. NIFTY 50, March 2020 to February 2021 — divergence firing repeatedly inside a large advance The Indian index rose steeply from its low in late March 2020 through to February
- That advance was not a straight
line — it contained several pullbacks and consolidations. Each time the index made a new high after a pause, the new leg was typically slower than the explosive recovery leg that preceded it, which is the arithmetic that produces bearish divergence on RSI. A trader selling on each divergence would have exited a historic advance repeatedly. The signals were correctly computed. The interpretation "the trend will now reverse" was wrong every time until eventually it was not.
2. S&P 500, January and February 2020 — the top with no warning The US index made its high on 19 February 2020 and then fell by roughly a third over the following 5 weeks. The advance into that high was strong and broadly steady. There was no long, obvious deterioration in momentum of the kind divergence is supposed to detect. The cause of the fall was external information arriving, not an internal exhaustion of buying. This is the case that defines divergence's limit. A pattern that measures the slowing of a move cannot warn about a move that stops for reasons outside the price series.
3. Nasdaq Composite, 1999 to March 2000 — the case everyone quotes, and the problem with it The Nasdaq Composite peaked in March 2000 after an enormous advance, and the period leading into that peak is the standard example in every divergence tutorial. Momentum measures did deteriorate during the final phase, and market breadth had been narrowing through 1999. The reason this case should make you more careful rather than less is selection. It is one of the most examined market tops in history, and it is examined because it was a top. Nobody publishes the 40 other occasions in the same decade when the same pattern appeared and the market carried on. Learning a success rate from famous examples is how every unreliable pattern acquires its reputation.
The question that resolves it
A novice sees a lower indicator high and asks: is this a reversal signal?
An expert sees the same lower high and asks: has price broken the level that would confirm it?
The second question is answerable from the chart, has a specific price attached, and produces a stop level. The first has no answer until months later.
There is a question underneath both, and it is the one that separates people who use divergence well from people it costs money. Would I be able to see this divergence in real time, or only because I am looking at a completed chart? If you drew the line after seeing what happened next, you have not found a signal. You have found an explanation.
What would make this wrong
If divergence carried a real edge, then regular bearish divergences would be followed by declines more often than the base rate of decline for that instrument over the same horizon. That comparison to a base rate is essential. On a stock that falls over any given 20 day window 45% of the time anyway, a divergence that is "right 50% of the time" has added almost nothing.
The exercise above is a start. A fair version requires 3 things most people skip: counting every divergence including the ones that failed, defining a fixed horizon before you look, and comparing against the instrument's own base rate.
The honest limits are 4.
First, the definition of a "pivot" is not fixed. Two traders looking at the same chart will identify different swing highs, and therefore different divergences. Any test depends on that choice, and it is not objective.
Second, this article uses RSI with a period of 14 on daily bars. Divergence found on MACD, on a weekly chart, or with a 7 period RSI is a different pattern with a different frequency. Claims about divergence in general are usually claims about one specific configuration.
Third, multi-timeframe divergence — the same pattern appearing on the daily and weekly chart at once — is more selective and is regarded by many traders as stronger. That is plausible and it is not established here. Test it before you believe it.
Fourth, and in fairness to the pattern: divergence used only as a risk-reduction trigger, meaning tighten stops and stop adding, has a much better case than divergence used as a reversal trade. It is a cheap way to become more careful. Being more careful in a decelerating trend is rarely a mistake.
In India
Divergence is widely taught in Indian trading education and is applied most often to NIFTY 50, BANK NIFTY and liquid single stocks, usually with RSI or MACD.
Three Indian features matter.
Price bands distort the indicator side. RSI, MACD and CCI are all computed from closing prices. On a stock locked at a circuit limit, the close is the band, not the market's opinion. A sequence of band-limited closes produces indicator values that describe the band rather than the demand. Divergences drawn across such a period are drawn on censored data.
Index concentration affects index divergence. NIFTY 50 is dominated by a small number of very heavily weighted companies. A divergence on the index can be produced entirely by 3 or 4 large stocks slowing down while the other 46 do something else. If you are trading the index, that is fine. If you are using index divergence to make decisions about a mid-cap holding, you are using a signal about different companies.
Fewer liquid instruments for testing. Anyone wanting to measure divergence's base rate needs a large sample of instruments with clean, continuous price histories. In India, the set of stocks with deep liquidity and long unbroken histories is smaller than in the US, and many mid and small caps have gaps, suspensions and band days in their records. Be careful about how much a backtest on 30 Indian stocks can tell you.
In the United States
Divergence is equally standard in US trading education and is applied to indices, single stocks and futures.
Three US features matter.
Overnight gaps create artificial pivots. A US stock that gaps 9% on results creates a price extreme that never traded during a session. Both the price pivot and the indicator value at that pivot are shaped by a single jump. Divergences drawn to or from an earnings gap are unreliable in a specific way: the price high is real, and the momentum reading behind it reflects one event rather than a process.
A much larger sample for testing. With several thousand liquid US equities and decades of clean daily data, divergence is one of the more heavily studied patterns. The published results are, to put it politely, not strong.
Index breadth is measured properly. US markets publish advance-decline lines, new highs versus new lows and other breadth series with long histories. That gives US traders a genuinely independent check on a divergence: if the index is making new highs on a lower RSI and fewer stocks are making new highs, those are 2 different observations from 2 different data sources. Article 19 covers this.
Where they differ, and what that tells you
Both markets produce divergences from the same formulas. What differs is what a second opinion is available from.
In the United States, a trader who sees a momentum divergence on the S&P 500 can immediately check the advance-decline line and the count of stocks making new 52 week highs. Those series are published, have long histories, and are computed from a different input — the number of individual stocks doing something, rather than the index price. So they can genuinely confirm or contradict.
In India, breadth data exists — NSE publishes advances and declines — but the histories available to a retail trader are shorter and less commonly charted, and the index's own concentration means breadth and index price can diverge for structural reasons rather than informative ones. When 10 companies carry a very large share of NIFTY's weight, the index can rise while most stocks fall without that being a warning about anything. It is a description of the index's construction.
What that tells you is where to spend your effort. In the US, when you find a divergence, your next move is to check breadth, because it is a real second source. In India, your next move should be to check the individual stock rather than the index, and to check whether the index move was carried by a handful of names. Copying the American habit of treating index breadth as confirmation does not translate, because the underlying index is built differently.
Carry this
- Divergence measures speed, not direction. Slowing is not reversing.
- It is a warning to tighten and stop adding. Price confirmation is the trigger.
- If a new extreme comes with the indicator, the divergence is cancelled. Let it go.
- Count your failures. Without them you have no success rate, only a memory.