Head and shoulders pattern — how to trade trend reversals

Recognizing three peaks does not yet establish a profitable method. This lesson turns head and shoulders into a test: define when the pattern becomes knowable, calculate a sloping neckline at the correct candle, distinguish a breakout from a retest, and record execution and costs. The GBP/USD geometry example and the separate EUR/USD timing model below are hypothetical, not reported trades or performance evidence.

What the head and shoulders pattern is

Head and shoulders is a chart pattern associated with a possible reversal after an uptrend. It has a higher central peak between two shoulders and a neckline through the intervening troughs. The break completes the geometric condition; it does not establish the subsequent price path or profitability of a trading rule.

The sequence is a first peak, a pullback, a higher peak, another pullback and a third peak below the head. This describes observed prices. It does not identify which participants traded, whether buyers lost conviction, or whether the next move will reverse the trend.

The mirror variant, the inverse head and shoulders, appears after a downtrend. Three troughs instead of three peaks, the middle trough deeper than its neighbours, and a line connecting two local highs that serves as a neckline signalling a possible upside reversal. In practical trading, both variants are treated with the same toolkit: identification, waiting for a confirmed breakout, entry, stop loss, target projection.

Structure of the pattern: LSH, HEAD, RSH and what happens between them

The anatomy of a classic head and shoulders consists of five distinct elements that are worth listing one by one. I will keep the shorthand that naturally circulates among analysts: LSH (left shoulder), HEAD, RSH (right shoulder).

A first peak alone is not a completed pattern. Record the trend rule and your method for finding local highs and lows before marking a sample. In the hypothetical EUR/USD model used here, a high requires two strictly lower highs on each side; a low requires two strictly higher lows on each side. Equal values are rejected. This chosen 2+2 rule makes a pivot on candle 13 knowable only after candle 15 closes. Unlisted bars are assumed to satisfy the inequalities; the selected points are not complete OHLC data or a backtest.

The shoulders need not be exactly equal. Set a numerical tolerance if you want repeatable classification, and keep the head above both shoulders. Distinguish this geometry from a range or triple top. Changing the tolerance, pivot rule or timeframe creates a different test version rather than evidence that the original method improved.

The neckline — the single most important line of the pattern

The neckline connects the trough after the left shoulder to the trough after the head. For points (t₁,p₁) and (t₂,p₂), use N(t) = p₁ + (p₂ − p₁) × (t − t₁)/(t₂ − t₁). Here t indexes consecutive candles, not calendar hours including a weekend gap. A close below the line is one possible confirmation rule; record any required buffer and when the right shoulder becomes knowable.

In a separate hypothetical EUR/USD Bid-price model, the shoulders are High 1.1080 at t=2 and High 1.1090 at t=13; the head is High 1.1140 at t=7. The two neckline troughs are Low 1.1000 at t=4 and Low 1.1024 at t=10. Their slope is 0.0004, or 4 pips per candle for a pip of 0.0001, so N(t) = 1.1000 + 0.0004 × (t − 4). Neckline orientation changes the reference price; it does not by itself establish comparative reliability.

After the right shoulder becomes known at t=15, our chosen breakout rule requires a close at least 2 pips below N(t), with the prior close not qualifying. N(15)=1.1044 and Close 1.1048 is 4 pips above it. N(16)=1.1048 and Close 1.1042 is 6 pips below it, so the rule passes. Using the second trough price 1.1024 instead would miss this event. In a separate control path, both closes are 1.1045: the gaps become +1 and −3 pips because the line rises while Close is unchanged. Adding C(t) < C(t−1) would reject that control path and define another variant.

Volume as a separately defined filter

A volume filter needs its own definition and data source. Record whether you use exchange-reported trade volume or tick volume from a particular feed. A plotted decline in volume does not identify the motivation of buyers or prove an imminent reversal.

For example, a test could require breakout volume greater than twice the mean of the preceding 20 completed candles. This is an illustrative rule, not a validated threshold. State whether the current candle is excluded from the mean, and compare the filter against the same unfiltered candidates instead of changing the threshold after seeing the result.

On MT5, tick volume counts ticks received during the bar; it is not market-wide FX turnover. Real volume represents trade volume where the feed provides it and may be absent for OTC symbols. Neither measure automatically confirms a profitable neckline break, and results from one feed need not match another.

Entry rules — aggressive, classic, conservative

Compare entry variants as different rules. A candle-close decision, a next-quote fill and a retest entry are separate events. In TradingView’s default strategy calculation, an order created at a bar close fills no earlier than a subsequent tick, often the next bar open in a historical test; settings can change that behavior. This software convention is not a guarantee of your broker’s price.

1. Before the neckline break. This variant trades an incomplete pattern. Specify the information available at entry; do not label the eventual shoulder high as known earlier when your pivot rule needs later candles. Record failed and uncompleted candidates as well as completed patterns.
2. After a qualifying close. The EUR/USD model recognizes the event after t=16. An order is then created for later execution; the signal close 1.1042 is not automatically the sale price. Record the actual hypothetical Bid fill separately.
3. Within a retest window. A second variant waits at most three subsequent candles, 17–19, for High to return to N(t) ±2 pips and Close to finish at least 2 pips below N(t). At t=18, N=1.1056, High 1.1057 and Close 1.1053 satisfy this chosen rule. A later qualifying touch creates no entry after expiry.

Define cancellation separately. In another path, Close 1.1059 at t=18 is above N(18)=1.1056 and cancels waiting; if already short, this example requests an exit at the next available quote. Close 1.1055 is only 1 pip below the current line and does not trigger that cancellation. Comparing it with the old N(16)=1.1048 would incorrectly mark it as 7 pips above. An Ask-based protective stop and a close-based exit are different rules.

Price projection target — the head’s geometry subtracted from the neckline

Measure the vertical head-to-neckline distance on the head candle, then project it from the neckline on the breakout candle and freeze that reference. In our EUR/USD model, N(7)=1.1012, height = 1.1140 − 1.1012 = 0.0128 = 128 pips, and N(16) − height = 1.0920. Do not keep moving the target as the line extends. The following GBP/USD example is a separate horizontal-neckline model.

The measured move is a geometric reference, not a promise. A trader may test a partial exit and a trailing stop, but their effect depends on the subsequent path and costs. A stop above the right-shoulder high defines planned invalidation risk; slippage or gaps can produce a larger realized loss.

Execution and costs: separate alternatives

Return to the separate EUR/USD model: sell 10,000 EUR at actual Bid 1.1040 on t=17, use a USD account, and assume total commission of USD 2 with no funding or other fees. A 0.0001 pip is USD 1. Each closing path below is an alternative, not a sequence of transactions. Short-position price P&L is 10,000 × (1.1040 − actual Ask repurchase).

If Bid reaches projection 1.0920 with a 2-pip spread, assumed Ask 1.0922 gives +USD 118 before commission and +USD 116 after it. If cancellation at Close 1.1059 on t=18 is followed by an Ask fill of 1.1062 on t=19, the results are −USD 22 and −USD 24. Exact stop execution at Ask 1.1095 gives −USD 55 and −USD 57; a 3-pip worse fill at 1.1098 gives −USD 58 and −USD 60. Spread is already in these execution prices; do not deduct it again. Convert for another account currency and add applicable funding. None is a maximum possible loss.

Practical takeaway: The neckline defines a possible confirmation condition; the head-to-neckline distance gives a geometric reference, not a guaranteed result.

Five test-design mistakes to check

Keep geometry, timing and execution consistent. These checks identify mismatched assumptions; passing them does not establish a profitable strategy.

• Using future pivot information. A 2+2 pivot is unavailable until the second later candle closes. A close-confirmed rule cannot take an earlier intrabar entry and still be evaluated as the same rule. Record the recognition and execution candles separately.
• Mixing volume definitions. Broker-feed tick counts and exchange trade volume are different fields. A volume condition must identify the field, baseline window and missing-data policy; otherwise it is not reproducible.
• Treating a stop as a loss ceiling. Record its price side, trigger rule and hypothetical fill. A stop beyond a shoulder and a close above the neckline are different invalidation choices. Gaps or slippage can make execution worse; no universal pip buffer follows from the shape.
• Adding context after observing the result. If higher-timeframe trend, support or a second indicator is part of the test, define it in advance using available information. Several price-derived conditions do not automatically provide independent evidence or a better outcome.
• Inferring intrabar order from OHLC alone. When both a target and a stop are inside one candle, the OHLC values may not reveal which was touched first. Use more granular available data or state a conservative assumption. A tester’s inferred path is a simulation convention, not observed tick history.

For each candidate, save its recognition time, neckline anchors, rejection rule, entry or no-entry decision, simulated fills and costs. Report all tested variants, the chosen sample and any parameter changes. A few attractive charts are not a performance estimate.

What to do tomorrow to get comfortable with this pattern

1. Choose a fixed sample. Hide the later chart while identifying candidates. Apply one recorded definition to every bar rather than selecting only places where an uptrend subsequently ended.
2. Measure and freeze each reference. Compute head height at the head candle and project it from the neckline at the recognized breakout. Record failed projections and unfilled orders as well as reached targets; three selected cases cannot estimate reliability.
3. Separate planned and realized risk. Record the entry condition, price-side stop, geometric reference and execution costs. Calculate potential reward divided by planned price risk; the ratio alone does not decide profitability or guarantee that a target will be filled.
4. Make an alert follow the line convention. A fixed price alert can become stale when N(t) changes. An alert only prompts a check; it does not prove the recorded close condition, surviving retest window or an executable fill.

Related reading: how to draw support and resistance — compare anchor conventions; the most important candlestick patterns — distinguish geometry from confirmation; multi-timeframe analysis — record additional context without treating it as a guaranteed advantage.