The head and shoulders pattern

Most retail chart patterns do not deserve the attention they get. They are shapes with names, and the name does most of the work: once something is called a rising wedge, the label starts to feel like a forecast. The head and shoulders belongs to that family, and it is the loudest member of it. Search for it and you will find the same three-hump cartoon reproduced thousands of times, usually drawn with a ruler, usually with no volume, usually accompanied by a number claiming how often it works.

And yet the structure is not empty, which is why it has outlived so much bad teaching. Strip the diagram away and what is left is a description of a specific, observable failure. A market in an uptrend is a market where each attempt to push price higher finds enough buyers to succeed. The pattern describes three consecutive attempts in which that stops being true: the second attempt succeeds and makes the highest price of the move, the third attempt is made and does not reach the second, and then the level that had been absorbing every pullback gives way. That is not a shape. That is a sequence of events with a meaning attached to each one.

The reason the shape gets taught instead of the sequence is that the shape is easy and the sequence is not. Recognising three humps takes a second. Establishing that buying conviction actually drained across three advances takes a volume pane, a comparison, and a willingness to conclude that the pattern in front of you does not qualify. That last part is the expensive one, because a trader who has already spotted the shape has already formed an opinion, and the volume pane is now in the position of having to talk them out of it.

The shape is the part that photographs well. The volume is the part that is evidence. Almost all the teaching keeps the first and drops the second.

So the honest framing of this pattern is narrower than the one you were probably given. It is not a sell signal. It is not a forecast. It is a name for a particular loss of conviction in an uptrend, and the name is only worth anything when the participation data underneath it agrees, and when the market has actually done the one objective thing the structure requires.

The usual teaching gives you five landmarks: two shoulders, a head, and the two lows between them. That list quietly leaves out the two elements that decide whether any of it means anything, which are the trend that came before and the break that comes after. Read the structure instead as a sequence of seven events, in order, because the order is the entire content of the pattern. Reshuffle them and the meaning disappears, even though the picture might still look much the same.

Read down that table and notice what is actually being claimed. Not that three peaks predict a fall. The claim is that a specific group of buyers, the ones who had been reliably showing up on every dip through the whole uptrend, made their biggest effort at the head and then were not there for the third attempt. The right shoulder is the interesting landmark precisely because it is a failure to repeat, and a failure to repeat is a change in behaviour, which is the only kind of thing a chart can honestly tell you about.

The neckline deserves its own note, because it is the only element that is not a peak or a trough. It is a line you draw, connecting the low after the left shoulder to the low after the head. That makes it the most subjective element in the structure and simultaneously the one everything else depends on, which is an uncomfortable combination. It may be horizontal, it may slope. A neckline that slopes down means the second low already undercut the first, so weakness had appeared before the right shoulder even formed. A neckline that slopes up means the opposite and makes the whole reading weaker, because the market was still making higher lows while you were preparing to call a top.

A definition worth being strict about. A head and shoulders top is a reversal pattern, and a reversal requires something to reverse. Three peaks with a higher middle peak sitting inside a sideways range is not this pattern, however much it resembles the diagram. The prior uptrend is not context or background; it is part of the definition, and it is the first thing to check and the most commonly skipped.

Here is the part that most teaching leaves out, and it is not a refinement. It is the reason the pattern has any claim on your attention at all.

The classical description, set down by Edwards and Magee in the late 1940s and repeated in every textbook since, is that the left shoulder advance carries the heaviest volume, the head advance carries less, and the right shoulder advance carries the least of all. In practice the ordering of the first two is not stable, and arguing about whether the left shoulder or the head was busier is a distraction. The part that carries the argument is the third one: the right shoulder should be the quietest advance of the three, and the break should be louder than any of them.

Think about why that ordering is the claim rather than an arbitrary rule. Volume is the count of how many hands were involved. An advance on heavy volume means a lot of people were willing to pay up. An advance to a similar price on much lighter volume means the same price level was reached with far less buying behind it, which tells you the pool of willing buyers has thinned. The right shoulder is therefore the diagnostic bar of the whole pattern: it is the market attempting the same thing it succeeded at twice before, with visibly fewer participants. If you want the underlying mechanics of what a volume bar is and is not measuring, our guide to volume in trading covers it.

The break bar completes the argument in the opposite direction. Sellers pushing price through a level that had held twice, on turnover clearly above the recent average, is participation arriving rather than participation leaving. A break on thin volume says the level gave way because nobody was defending it that day, which is a much weaker statement and a common way for a break to be undone in the following sessions.

A head and shoulders without the volume story is not a weak head and shoulders. It is a shape somebody drew.

Three honest cautions about applying this in Indian markets. First, index volume is not a clean measure of conviction in the index itself, because it aggregates thousands of unrelated decisions across constituents; the volume signature reads more cleanly on a single liquid stock than on Nifty or Bank Nifty. Second, in the derivatives segment, volume around expiry and around rollover weeks is distorted by position management that has nothing to do with directional conviction, so a volume spike in those windows is often noise wearing the costume of evidence. Third, on illiquid instruments a handful of large orders can produce any volume profile you like, which means the pattern is least trustworthy exactly where it is easiest to find.

Everything discussed so far is interpretation. The shoulders are drawn by a reader. The neckline is drawn by a reader. The volume comparison requires a reader to decide which bars belong to which advance. There is exactly one element of this pattern that the market produces without a human supplying it, and that is a close on the far side of the neckline.

This is why the pattern does not exist until the break. It is not a cautious way of speaking. Before the close, what you have is a hypothesis about a structure that may complete, and the base rate for hypotheses that never complete is uncomfortably high. Anticipating the break, entering at the right shoulder because the shape looks obvious, is the single most expensive habit associated with this pattern, and it is expensive precisely because it trades the one objective element away in exchange for a better price. Our guide to double tops and double bottoms works through the anticipation trap in detail, and every word of it applies here with one more peak attached.

The word doing the work in that table is decisive, and it is worth being blunt about the fact that nobody can define it for you in a way that removes the judgement. A wick through the neckline is not a break; the market went there and did not stay. A close a few paise beyond it is technically a close and practically nothing. Traders adopt filters, a close beyond by some fraction of the recent average range, or two consecutive closes, or a close beyond on expanded volume, and every one of those filters is a trade. Each removes some false breaks and removes some real ones with them. There is no setting that only removes the bad ones, and any teaching that implies otherwise is selling something.

For Indian instruments there is a specific mechanical reason to judge the break on a closing basis rather than intraday. Overnight news moves NSE and BSE instruments in a gap at the open rather than through continuous trading, which means price can appear on the far side of your neckline without a single transaction having occurred at the level. A gap through a neckline is not the market grinding through supply; it is the market reopening somewhere else. The mechanics of what does and does not count as a genuine break are covered in our guide to breakouts and fakeouts.

Now the uncomfortable part, and the reason this page refuses to quote you a success rate.

The three panels above contain the identical candles. Not similar candles: the series is generated once and drawn three times without modification. Only the annotation layer changes. In the first panel a neckline is drawn under two of the lows and a head and shoulders appears. In the second, two horizontal lines are drawn instead and the same bars become a tired range poking at one ceiling. In the third, the lows are joined and the same bars become an intact rising channel that has never been broken. Each drawing uses real pivots. Each would be defended by whoever drew it.

This is not a trick played on a contrived series. It is the ordinary condition of price. Any market that has made three pushes has made three pushes, and the middle one was either the highest, the lowest, or in between; if it was the highest, you can call the outer two shoulders. The pattern is extremely easy to find because the criteria for finding it are extremely loose, and the criteria are loose because they were written to describe charts rather than to be executed.

That has a direct consequence for every performance statistic you have ever seen attached to this pattern. To count how often something works, you must first be able to say when it occurred, and occurrence here depends on a human judgement about which pivots count. Any such number therefore measures the person doing the identifying at least as much as it measures the market. This is not a hypothetical objection. When Lo, Mamaysky and Wang tested technical patterns properly in 2000, the first and hardest part of the work was building an automated definition, using kernel regression to smooth prices and locate extrema, precisely because the pattern as normally taught cannot be evaluated at all. The definition had to be invented before the question could even be asked.

Why there are no percentages on this page. Any figure of the form this pattern works N percent of the time encodes a specific person's rules for what counts as a pattern, on a specific sample, over a specific period, with a specific definition of working. Change any one of those and the number changes. Quoting it to you would imply a precision that the underlying identification cannot support, so this page does not quote one, and you should treat sources that do with more suspicion than they usually receive.

The defence against all of this is not cleverness. It is order of operations. Draw the neckline before the break, not after it. Decide what would falsify the read before you have a position. Commit to the volume comparison while it can still tell you no. A pattern identified after the outcome is known has told you nothing, however convincing the drawing looks, and the drawing will always look convincing, because you drew it to fit.

Suppose the break happens: a decisive close through the neckline, on volume clearly above the recent average, after three advances that show conviction draining. The structure has done everything it can do. The question is now a narrower and more practical one, which is where the entry goes.

The broken neckline frequently gets retested. Price closes through it, then returns to it from the other side, and either fails there or reclaims it. The level that had been absorbing selling all the way up now has to absorb buying from below, and whether it does is genuinely informative. This return trip is called a throwback or a retest, and it is where the more disciplined entry lives, for two reasons that are worth keeping separate because only one of them is verifiable.

The first reason is filtering. A break that is going to be undone tends to be undone quickly. Waiting for the retest lets a certain population of false breaks eliminate themselves before any money is committed, at no cost other than patience. The second reason is arithmetic, and it is the one you can check. The level that invalidates a head and shoulders read sits above the right shoulder, because a move back through there says the third attempt did not fail after all. That level is fixed. A retest entry is nearer to it than a break entry is, so the distance you are risking to hold the same idea is smaller, as the two point figures in the middle panels show.

What the retest does not do is make the pattern more likely to work. Nothing done after the break changes what the market does next. This distinction matters because it is routinely blurred in teaching, where waiting is presented as though it improves the odds rather than the price. It improves the price. That is enough of a reason on its own, and it is an honest one.

And it has a cost, shown in the third panel, which most teaching omits: sometimes there is no retest. Price leaves and does not come back, and the rule that protected you from false breaks has also kept you out of a real one. Any entry rule that filters is a rule that also misses. Someone who tells you their rule filters without missing has either not run it for long or is not telling you about the misses.

On the invalidation level. A level above the right shoulder is where the reversal thesis is falsified, which is a different question from where any individual should place a stop or how much to risk. Those depend on the size of the account, the instrument, and the individual's own rules, and nothing on this page is a recommendation about any of them. Note also that a stop is an instruction to transact, not a guaranteed exit price: a gap through it fills where the market reopens.

The conventional projection is simple. Measure from the top of the head down to the neckline, then extend that same distance from the point where the break occurred. That gives a level below the break which is routinely called the target, and calling it that is where the trouble starts.

Consider where the number comes from. The height of the pattern is a measure of how far this market swung while it was building this structure. Projecting it forward is a statement of the form: a market that has just been moving this much per swing might reasonably move about this much more. That is a statement about the instrument's recent volatility. It is not a commitment by anyone to transact at any price, and there is no mechanism by which the market owes it to you.

Which is why drawing it as a single line is a category error. The measurement is built out of swings and bars, and the finest distinction it can honestly resolve is roughly the size of one ordinary bar in the pattern. The figure above therefore draws the projection as a band whose half-width is the mean bar range of the pattern's own candles, computed from the bars drawn there. That is not a stylistic preference. A single line claims a precision the arithmetic never had, and a band roughly the width of the market's own noise is the honest rendering of the same calculation.

In practice, moves fall short of the projection, land in it, and run well past it, which is why the figure shows all three. Treating the zone as a place to start paying attention, rather than a destination to hold out for, is the difference between using the measurement and being used by it. The trader who refuses to take anything until the projected level prints has quietly converted a rough estimate into a rule, and the market was never consulted about that rule.

The inverse head and shoulders is the same structure upside down at the end of a downtrend: a lower central trough between two higher troughs, with the neckline drawn across the intervening highs, and completion on a decisive close above it. Geometrically it is a clean reflection, and everything said so far about the break being the only objective event, about hindsight, about the projection being a zone, transfers without modification.

The volume, however, does not reflect, and this is the detail most often mishandled because mirroring the picture makes it feel as though the evidence should mirror too. It should not, and the reason is that tops and bottoms are made by different processes.

A top is exhaustion. It is made by buying running out, and buying running out looks like participation falling away across successive advances. A bottom is accumulation. It is made by selling drying up and then by buyers arriving, and arriving buyers look like participation rising. So in an inverse pattern the expectation is that selling pressure contracts into the right shoulder, and then the advance out of it, through the neckline, is louder than the advances before it. Fading volume on the push out of an inverse right shoulder is not confirmation. It is a warning that the base is being built by an absence of sellers rather than a presence of buyers, which is a much thinner foundation.

There is a second asymmetry worth knowing, though it is a tendency and not a rule. Tops tend to form faster and more sharply than bottoms, because the emotions driving them differ: a market can run out of buyers quickly, while a market that has been sold heavily usually needs time before size will commit. An inverse pattern that takes noticeably longer to build than its mirror image is behaving normally, not weakly.

Failure here is not an exception to be explained away. It is ordinary, and a reader who has not thought carefully about the failure modes is not reading the pattern, only recognising it. The failures come in a small number of recognisable shapes.

Two of those deserve expanding because they are the ones that feel like skill while they are happening. The first is drawing the neckline to fit. There are usually several candidate lows, and the one that produces the cleanest line is not always the one that reflects where the market actually turned. Once the line has been nudged, the break through it means nothing, because the level was chosen by the person who wanted the break. The discipline is to draw the line off obvious pivots, accept it if it is untidy, and abandon the read entirely if no believable line exists.

The second is timeframe shopping. A structure that has not completed on the daily chart can usually be found somewhere on the hourly, and if not there then on the fifteen minute. Every step down that ladder buys a cleaner-looking pattern with noisier data behind it, and the volume evidence degrades fastest of all, because intraday volume is dominated by session effects that have nothing to do with conviction. If the pattern is only visible after you have changed the timeframe to find it, what you have found is the timeframe, not the pattern.

The general point underneath all of them is that this pattern is unusually vulnerable to motivated reasoning, because it has many degrees of freedom and no fixed definition. Where the two-test structures at least constrain you to a specific number of touches, three pushes plus a hand-drawn line can be assembled from almost any stretch of chart by someone who wants to assemble it.

Strip away the diagram, the arrow and the confident percentage, and what remains is worth keeping. The head and shoulders names a loss of conviction in an uptrend. It says: the strongest push these buyers had was the second one, the third one fell short, and the level that had absorbed every pullback finally gave way. As a description of a market changing behaviour, that is precise and useful.

What it is not is a forecast, and the distance between those two things is exactly the distance between the shape and the evidence. The shape is available to anyone with a mouse and hindsight. The evidence is narrower and harder: participation actually falling across three advances, a neckline drawn off pivots that were obvious before the break, a decisive close through it on expanding volume, and a projection held as a zone rather than a promise. When those agree, the name has content. When they do not, the name is decoration on a chart.

The pattern names a loss of conviction. Only the volume and the break turn that name into evidence.

That order, structure first, then evidence, then event, is the method we teach, and it generalises well beyond this one pattern: a piece of chart geometry is a question, not an answer, and the work is deciding what would count as an answer before the market supplies one. Learn that order and the patterns stop being a vocabulary to memorise and start being a set of questions you already know how to test.

• Robert D. Edwards and John Magee, Technical Analysis of Stock Trends (first edition 1948). The work that codified the head and shoulders as a named structure and set out the volume expectation across the three advances. The volume description used on this page, and the insistence that a reversal pattern requires a prior trend to reverse, both trace to this text and its many later editions.
• Andrew W. Lo, Harry Mamaysky and Jiang Wang (2000), Foundations of Technical Analysis: Computational Algorithms, Statistical Inference, and Empirical Implementation, Journal of Finance 55(4), 1705 to 1765. Built an automated pattern-recognition method using nonparametric kernel regression to smooth prices and locate extrema, then applied it to United States stocks from 1962 to 1996. Cited here for the methodological point that the pattern had to be given an algorithmic definition before it could be evaluated at all. onlinelibrary.wiley.com
• P. H. Kevin Chang and Carol L. Osler (1999), Methodical Madness: Technical Analysis and the Irrationality of Exchange-rate Forecasts, The Economic Journal 109(458), 636 to 661. Evaluated head and shoulders forecasts on daily dollar exchange rates over 1973 to 1994 against two criteria, profitability and efficiency, with significance assessed by bootstrap. The authors report that the rule is profitable but not efficient, since it is dominated by simpler trading rules. Cited here because a pattern that is outperformed by simpler alternatives is not carrying its own weight, which is a more useful finding than any success rate. academic.oup.com
• Exchange session and settlement conventions. Indian exchanges publish an official open, high, low and close for each instrument for each session, and the continuous equity session runs from 09:15 to 15:30. Every statement on this page about judging a break on the close, and about gaps at the open, follows from those conventions rather than from any vendor's charting defaults.