Head and shoulders — how to read the reversal pattern

Head and shoulders describes three peaks after an advance, with the central head above the two shoulders and two intervening troughs defining a neckline. Treat it as a candidate until the chosen break condition is met. This guide distinguishes geometry, confirmation time and hypothetical trade arithmetic; neither the shape nor its projection guarantees reversal.

How is the head and shoulders pattern built?

For the classic reversal interpretation, record a prior uptrend and five selected turning points: left shoulder, trough, head, trough and right shoulder. A zigzag selected from a range does not by itself establish that prior context. Keep the selection rule separate from a later neckline break.

1. An uptrend preceding the pattern. That is a precondition, not decoration — the pattern has to reverse something.
2. The left shoulder — the first peak, after which price pulls back to the level that is about to become the neckline.
3. A pullback to that level, followed by a bounce.
4. The head — the highest peak of the whole formation. If the middle peak is not the highest, you do not have a head and shoulders.
5. A second pullback, again into the neckline area.
6. The right shoulder — a selected peak below the head. Define what roughly similar shoulders means numerically; the price point does not reveal who traded or why.
7. The neckline break — check the defined price and timing condition after the right shoulder is recognisable. Confirmation means that condition passed, not that future reversal is certain.

The inverse version uses three troughs after a decline: the central head is lowest and the neckline joins the two intervening peaks. Change the confirmation direction to upward, rather than reversing the drawing while keeping a bearish closing rule. A qualifying break still does not guarantee subsequent gains.

Why does the neckline decide everything?

A neckline joins the two selected troughs and may be horizontal or sloped. If several trough pairs look plausible, make the selection rule explicit rather than silently choosing the line that fits the eventual outcome. Save any tolerance before reviewing results. An ambiguous human drawing is a reason to refine the definition, not proof of a particular market participant’s action.

The neckline is related to a support level built from two real troughs, but crossing it does not show that buyers disappeared. In a separate fictional horizontal-neck model, shoulders 1.1070 and 1.1080 flank a head at 1.1200, with both troughs at 1.1000. Height is 200 pips at 0.0001 per pip. A chosen 10%-of-height shoulder-price tolerance is 20 pips; the 10-pip shoulder difference passes that one condition.

A sloped neckline changes a comparison. For a separate model, N(t) = 1.1000 + 0.0005 × (t − 2), from troughs at t = 2 and 8. Shoulder prices 1.1070 at t = 0 and 1.1080 at t = 10 differ by 10 pips, but their heights above N are 80 and 40 pips. A head at t = 5, price 1.1150, stands 135 pips above N(5) = 1.1015. A 10% tolerance is 13.5 pips: comparing shoulder prices passes, comparing local heights fails. Specify which metric the detector uses; neither tolerance is a universal standard.

What does a complete setup on EUR/USD look like?

The target comes from the measured move: you take the distance from the top of the head down to the neckline and project it downwards from the neckline. In this example, one hundred pips from 1.1000 gives 1.0900.

The table’s head, shoulders and prices are hypothetical and distinct from the earlier 200-pip geometry example and the diagram’s arbitrary units. Its 100-pip projection begins at the 1.1000 neckline, not the 1.0995 entry. The selected exact-fill outcomes give 95/75 ≈ 1.2667 reward per unit of price risk, excluding commission and funding. The ratio alone does not establish expected profit; realised fills or costs can change both outcomes.

When to enter — three moments and what each one costs

1. Before confirmation. A rule acting while the right shoulder is forming uses incomplete pattern information. State what triggers the order; an earlier decision does not guarantee a better price or tighter realised loss.
2. After a completed close below the neckline. This is the selected closing-rule variant. In the 1.1000-neck example, Close 1.0980 passes; Low 1.0970 with Close 1.1005 does not. The latter can pass a different intrabar rule. The close is not a guarantee of the next executable price.
3. Conditional on a return to the neckline. Define retest tolerance, confirmation and expiry. A return may never occur or may invalidate the scenario. It is not established here as the safest variant or best reward-to-risk.

Compare those protocols separately on the same candidates, recording failed confirmations and no-fill retests. Waiting changes available information and possible prices; it does not supply a universal ranking of entry quality.

What confirms the break?

Additional filters are separate rule choices. State their data, threshold and availability time, then compare a version with and without each one. Do not treat a list of price-derived conditions as automatically independent evidence.

• Activity data. In MT5, your platform’s tick volume counts received ticks; real deal volume is a separate field that may be unavailable for OTC. Record the feed and unit. Neither is a universal global-FX turnover series.
• Break-candle geometry. Define its body, range and completed-close condition. A large body records an Open–Close change; it does not identify supply, hidden protective orders or participant intentions.
• Oscillator comparison. Record exactly which price pivots and RSI or MACD values you compare and when they became available. They may use the same prices as the pattern, so agreement is not automatically independent confirmation.
• Higher-timeframe rule. Define its trend measure and use completed bars already available at the decision. Differing directions across horizons do not by themselves prove that one candidate is weaker or identify its participants.

How much are the pattern statistics really worth?

A performance claim requires a specified instrument, feed, period, detector, entry/exit rule and costs. A book’s stock-market sample cannot be assumed to establish the same result for OTC forex. This article has not reproduced the cited book’s study, so it does not assign a success rate or claim that the pattern beats chance.

A confirmed shape still needs an execution and risk plan. See protective-order and target conditions for the distinction. Ordinary stops are not guaranteed fill prices, and moving a stop farther away changes the modelled loss distance. Define volume and costs as well as chart levels rather than assuming one placement protects every account.

The most common mistakes when trading head and shoulders

1. A pattern in the middle of a range. Without a preceding uptrend, three peaks are just three peaks, not a reversal signal. There is nothing to reverse.
2. A neckline forced onto the chart. If you have to bend the troughs or ignore one of them to make the line look neat, the pattern is not there. A chart is not a test of imagination.
3. Changing the confirmation rule. A completed-close system cannot count a wick breach as its signal after seeing later prices. Keep an anticipatory or intrabar system as a different protocol.
4. Confusing chart invalidation with loss control. A chosen Close above the right shoulder can invalidate a scenario, but it is not necessarily the order trigger or a guaranteed loss bound. Define those rules separately.
5. Assuming an optimal timeframe. M5, M15, daily and weekly bars aggregate different observations. A timeframe label is not a verified success rate; evaluate the specified rule and costs on each chosen dataset.

Keep invalidation separate from projection. In the 200-pip horizontal-neck example, a pre-break high above 1.1200 defeats the selected highest-head candidate. After confirmation, a chosen Close-above-right-shoulder rule is met at 1.1085, but not 1.1008, even though 1.1008 is back above the neckline. Neither rule prescribes waiting for a candle close to place protection; use a separate order and risk plan.

What to do tomorrow

1. Record candidates before outcomes. Save instrument, feed, timeframe, selected points and identification time. Keep rejected candidates and mark any replacement as a new version. A few selected examples are not a performance test.
2. Separate projection and execution. Record actual entry and assumed exit price sides, planned protection, commission and funding. A two-to-one threshold can be a chosen rule, but the ratio alone does not establish positive expectancy.
3. Freeze any additional filters. State their definitions and information time before evaluation. Assess them on later data not used to select the rule; adding two agreeing labels does not prove independent confirmation.