Reversal chart patterns
The answer
A reversal pattern is a picture drawn around 2 events that you can already measure without the picture: the trend stopped making new extremes, and then a price level where real volume traded gave way. The shape adds a name. It does not add information.
Why this costs you money
The names are the problem.
When you call a shape a head and shoulders, you have not described it. You have classified it, and the classification carries a prediction: the trend is about to reverse and here is the target. That prediction arrives with the name, before you have checked anything.
Here is the specific loss. You see the shape forming. You short, or you sell your holding, at the neckline break. Price goes 3% against you. Now you cannot easily exit, because the pattern told you a target far below, and you are only 3% away from your entry. So you widen the stop. The shape has replaced your judgement.
The second cost is larger and harder to see. Nobody counts the failures. A shape that leads to a reversal gets named, screenshotted and taught. A shape that looks identical and then goes the other way is never named at all, because a pattern is only labelled once it has completed. Your whole sense of how reliable these patterns are comes from a collection that was assembled after the outcomes were known.
So you are estimating a probability from a sample where every failure has been quietly removed. That is not a small bias. It is the entire reason the patterns look reliable.
How it works
The shapes, stated plainly
Head and shoulders (a top). Three peaks. The middle one is the highest. The lows between the peaks define a neckline, which is a support level. The pattern is considered complete when price closes below the neckline.
Inverse head and shoulders (a bottom). The same shape turned upside down, at the end of a downtrend.
Double top. Two peaks at roughly the same price, with a low between them. It completes when price closes below that low. Triple top is the same with 3 peaks. Double and triple bottoms are the mirror images.
Rounding top or bottom. A long, slow curve with no sharp turn. It has no precise completion point, which makes it the least testable shape here.
The measured move
Every one of these comes with a target rule.
For a head and shoulders, measure the vertical distance from the top of the head to the neckline, and project that same distance down from the neckline break. For a double top, measure from the peaks to the low between them and project down from the break.
Understand what this rule is. It is a claim that the next move will be the same size as the previous move. There is no mechanism behind it. No order sits at the target. It is a convention, and it has been repeated so often that it is quoted as though it were derived from something.
What is actually in the pattern
Take the head and shoulders apart and only 2 things are left.
Fact 1: a break of structure. The right shoulder is a lower high than the head. The low before it may be a lower low than the low before the head. That is the definition of a downtrend beginning, and you can measure it by counting swing points, exactly as in the first article of this cluster.
Fact 2: a support level giving way. The neckline is a horizontal price where the market turned up twice. If real volume traded there, it is a real level for the reasons set out in the second article: trapped holders, resting limit orders, possibly a round number.
That is all. Both facts are measurable without drawing a single line, and both are available to anybody who has read the 2 previous articles.
Everything else in the pattern is decoration. The symmetry of the shoulders. The slope of the neckline. Whether the right shoulder is slightly higher or lower than the left. The number of weeks the shape took. None of these has been shown to add anything, and each one gives you another degree of freedom with which to convince yourself the shape is present.
Volume, honestly
The traditional claim is that volume should decline through the pattern and expand on the break.
The second half of that is real, and for a plain reason. A large increase in volume on a breakdown means many more people transacted at the new price. That is genuine evidence that opinion changed, not just that a few orders were absent. A break on low volume can be nothing more than a thin afternoon.
The first half — a specific declining pattern of volume across the 3 peaks — is folklore. It is quoted everywhere, it has never been shown to add predictive value, and in practice it is judged by eye after the fact.
So use volume as a filter on the break, and ignore it inside the shape.
The reproducibility problem
Here is a test you can run on any trading group. Show 5 people the same chart and ask each to mark the head and shoulders. You will get 5 different necklines, and often disagreement about whether the shape is there at all.
An instrument that gives different readings to different users is not measuring anything. Compare this with a break of structure, where 2 people using the same swing setting will always agree, and will agree on the date.
This is why the serious academic work on chart patterns starts by defining them algorithmically. Lo, Mamaysky and Wang did this in 2000, smoothing US price series mathematically and detecting patterns by rule rather than by eye. They found that some patterns did carry statistical information — the distribution of returns after a detected pattern differed from the unconditional distribution. That is a real result and it should be reported honestly. It is also a long way from a trading edge: the differences were modest, the study did not establish profitability after trading costs, and the patterns were defined by a computer, not by a person with a mouse. the precise findings and the sample period before citing figures.
Work on the head and shoulders specifically has been less kind. Osler's 1998 study of the pattern in US equities found that trading it was not profitable. this reference and its exact conclusion.
What it tells you, and what it does not
A completed reversal pattern tells you the trend has stopped making new extremes and a support level has broken. Both facts are worth knowing.
It does not tell you that a reversal will occur. Most breaks of a level are followed by more of the same, some are followed by an immediate return, and there is nothing in the shape that separates the 2.
It does not give you a valid target. The measured move is a convention.
It does not tell you anything about the size of the eventual move. A top that took 8 months to form can be followed by a 6% decline. A top that took 3 weeks can be followed by a 40% one.
And it tells you nothing about the company or the economy. A shape is a description of price. If a business is compounding earnings, a head and shoulders on its chart is a description of a pause.
The decision rule
Trade the 2 facts, not the shape.
If the sequence of swing points has turned down, and a level where large volume genuinely traded has broken on clearly increased volume, then treat the trend on that timeframe as over, regardless of whether the shape has a name.
If only 1 of the 2 is present, wait. A pretty shape with no volume on the break is a drawing. A volume break with the structure still making higher lows is usually a shakeout.
Unless the break happened on a scheduled event — a results announcement, a policy decision, an index rebalancing — in which case the volume tells you nothing about conviction, because volume is always high on those days. Wait for the next session.
Two practical instructions that follow from this.
Do not set your stop just above the right shoulder. That is the most obvious price on the chart and it is where everybody else's stop is.
Do not use the measured move as a target. If you need a target, use a level where price has actually traded, or a multiple of the amount you are risking.
Try this now
Five minutes. This is the base-rate test, and it is the one thing that separates people who use patterns well from people the patterns use.
- Open a daily chart, 3 years, of an index or stock you follow.
- Find one completed reversal — a double top, a double bottom, or a head and shoulders — where the shape is obvious in hindsight. It will not take long, because obvious shapes are obvious in hindsight.
- Now cover the chart from the neckline break onwards with your hand or a piece of paper. Look only at what a person had on that day. Write down 2 things: what you would have called this shape at the time, and where you would have placed your stop.
- Uncover. Compare what you wrote with what happened.
- Now the important step. Scroll back through the same 3 years and find 3 places where price made 2 highs at a similar level and then went up instead of down. Mark them. Count them.
What you should see. Step 5 is the one that lands. Those 3 places had the same shape as the beginning of a double top, and none of them is called anything, because the label is only applied when the outcome is a reversal. Once you have seen 3 unnamed twin peaks on your own chart, you cannot go back to believing the named ones are rare.
If you find more than 3 in 3 years, that is the normal result, and it is the answer to the question of how often the shape appears without the outcome.
A second step, worth 60 seconds. Open your own trade history and pick a trade you took on a pattern. Look at the chart as it was on your entry date and ask whether the level you traded was visible then, or whether you can only see it now. Do this for 3 trades and write down the score.
Three real cases
1. The NIFTY 50, January 2008 and November 2010 (India) — a double top that was real The index reached roughly 6,357 intraday on 8 January 2008, fell by more than half through that year, recovered over 2009 and 2010, and reached roughly 6,338 intraday on 5 November 2010. Those 2 peaks are within about 0.3% of each other, nearly 3 years apart. The index then fell to roughly 4,531 by December 2011. This is as clean a double top as any market produces. all 4 figures against NSE historical data. Note what made it informative: enormous volume had traded in that zone in 2007 and 2008, and a very large number of investors were holding losses from it. The mechanism was trapped holders, which is the same mechanism as ordinary resistance. The shape was the symptom.
2. The S&P 500, June and July 2010 (United States) — the most famous failed pattern of its decade Through June 2010, a head and shoulders top in the S&P 500 was described in almost every US market commentary, with a neckline in the region of 1,040. The neckline broke. The index fell further, reached a low in early July around 1,010, and then rallied for the rest of the year to finish 2010 higher. the exact neckline level, the exact July low and the closing level for 2010. The pattern was correctly identified in advance by a large number of people, it completed exactly as the textbooks describe, and it did not work. This case is worth more than 10 successful examples, because successful examples are the ones you are always shown.
3. The NIFTY 50, October 2021 to June 2022 (India) — right for 8 months, wrong for 2 years The index peaked around 18,604 intraday on 19 October 2021, and after a pullback made a second, slightly lower peak in January 2022. The structure turned down and the index fell to roughly 15,183 intraday by mid-June 2022, a decline of about 18%. A trader acting on the shape was correct. The index then recovered, exceeded the 2021 peak within about 18 months, and went far higher. all figures and dates. The lesson is about timeframe: the same pattern was a correct signal on a 6-month view and a wrong signal on a 3-year view, and the shape itself contains nothing that tells you which view you are in.
The question that resolves it
A novice looks at a shape and asks: what pattern is this?
An expert asks: what would this look like on the day it completed, before I knew the outcome?
The first question can always be answered, which is exactly the problem. Every chart can be described as some pattern. The second question requires you to delete information you already have, and that is uncomfortable, which is why almost nobody does it.
What would make this wrong
If reversal patterns carried no information, then breaks of algorithmically defined necklines would be followed by returns no different from the returns following randomly chosen days with similar volatility. The evidence here is genuinely mixed. Lo, Mamaysky and Wang found some statistical content in algorithmically defined patterns in US stocks. Osler found the head and shoulders unprofitable in US equities. Both results can be true at once: a pattern can carry a small amount of information and still lose money after costs.
What would overturn the practical conclusion in this article is a study showing that patterns identified by eye, by ordinary traders, in advance, produced better outcomes than the 2 underlying facts alone. whether such a study exists. The design is difficult, because it requires recording people's identifications before the outcome, which is precisely the record nobody keeps.
Three honest limits.
First, this article does not say patterns never work. It says the working part is the structure break and the volume-confirmed level break, both of which you can measure directly.
Second, self-fulfilment applies here as it does everywhere. A neckline that thousands of traders are watching has real stop orders below it, and those orders produce a real move when they trigger. That is a reason the break is often sharp. It is not a reason the move continues.
Third, on very long timeframes — a top that takes 2 years to form on a weekly chart — the shape is describing a genuine, slow change in who owns the asset. The criticisms here apply most strongly to shapes found on daily and intraday charts, where the sample of shapes is enormous and the selection problem is worst.
In India
Three Indian specifics change how these shapes appear and how they should be read.
Gaps disrupt the shapes. Because Indian markets have a single session with no long pre-market window, overnight news arrives all at once at 9:15 am. A neckline is frequently not broken during trading at all. Price simply opens below it. When that happens, the volume on the break tells you much less, because the opening auction concentrates a whole night of orders into a few minutes. Wait for the close, and preferably the next day's close.
Price bands cap the completion. A stock that hits a 5%, 10% or 20% band cannot move further that day. A neckline break that coincides with a band is not a measurement of selling pressure. It is the rule stopping the auction. the current band structure on the NSE circulars page.
Index shapes are crowded with option positions. Because Indian index options carry very large open interest at round strikes with weekly expiry, a neckline that sits near a heavily traded strike will behave differently from one that does not. Check the option chain before assuming the shape is doing the work.
Volume data in India has a useful feature with no direct US equivalent: the NSE publishes the delivery percentage, the share of each day's volume actually taken into demat accounts rather than squared off intraday. A neckline break on high volume but low delivery is largely intraday traders. A break on high volume and high delivery means people took real positions. That is a genuinely better confirmation filter than raw volume.
In the United States
Extended hours blur the break. A neckline can break at 6:00 am Eastern time on very thin volume and be recovered before the 9:30 am open. Decide your rule in advance. Most professionals count only regular-session closes, and that is the defensible choice, because pre-market volume is a small fraction of the day's trading.
Volume is fragmented across venues. US shares trade on many exchanges and in private venues at once. The volume figure your platform shows is a consolidated number, and a large share of it may have executed away from any public exchange. This does not make volume useless, but it does mean the fine detail of volume within a pattern is noisier than it looks.
Earnings dates dominate the calendar. A very large proportion of US reversal patterns complete on or immediately after a quarterly result. When that happens, the shape did not cause anything. New information arrived at a known time. Before attributing a break to a pattern, check whether the company reported that day. This single check removes a great many false conclusions.
Index reconstitution matters. Quarterly index rebalancing days produce enormous volume with no informational content at all, because index funds must trade regardless of opinion. A volume-confirmed break on a rebalancing day is not confirmed. the current rebalancing schedule for the major US indices.
Where they differ, and what that tells you
The instructive difference is in how a level is broken.
In the United States, a neckline is usually broken through trading. Price walks down to the level, trades at it, and goes through it while the market is open. There is a record of how many shares changed hands at each price on the way. That record is the evidence.
In India, a neckline is very often broken by a gap. The market closes above it and opens below it, and no trading occurred in between. The exchange ran a call auction and produced a single clearing price. The gap may also be capped by a price band, so the size of the move is partly set by regulation.
What that tells you is a rule specific to Indian charts. A gap through a neckline is a much weaker signal than a traded break through the same neckline, and you should treat it as unconfirmed until the market has traded through the level during a normal session. Wait for the second day. Indian traders who apply US-written pattern rules directly treat both breaks as identical, and they are not: one is a full day of people agreeing on a new price, and the other is a single auction reacting to one piece of news.
The reverse warning applies to US readers using Indian or other single-session markets: the absence of gaps in your home market has trained you to trust breaks more than the data supports elsewhere.
Carry this
- A reversal pattern is a break of structure plus a broken level. Trade those.
- The failures are never named, so your sense of the odds is built from a filtered sample.
- The measured move target is a convention, not a calculation.