Cup and Handle: Convention and Failure Cases

Identify the parts without assuming the outcome

The cup is the broader decline and recovery; the handle is a subsequent, usually smaller consolidation near its upper region. Fidelity describes the cup-with-handle pattern. Analysts differ in allowed duration, depth, handle shape and confirmation rules, so the label alone is not a complete test specification.

A completed historical chart can make the rounded shape look obvious. A real-time observer sees an unfinished sequence that may deepen, extend or fail. Record when each part first satisfied your definition rather than dating the entire pattern from a low that was only recognized afterward.

Rounded bottom, rounding top, or cup and handle?

A rounded bottom describes a gradual descent, low region and recovery; it does not by itself require a handle. StockCharts describes the saucer-like bottom convention. A cup-and-handle adds a separately defined pullback after the right rim, as in Fidelity’s guide. A rounding top is the inverted broad-arc description, not an upside cup with a guaranteed opposite result. These labels describe historical prices; none establishes what the next bar will do.

To make the shape test reproducible, consider nine consecutive completed closes of equal-duration bars on one fixed feed. In this invented rule, the center three closes (bars 4–6) must fit within 2.00 units, the fall from bar 1 to bar 3 and the rise from bar 7 to bar 9 must each reach at least 4.00 units, each endpoint must stand at least 8.00 above the center minimum, and the endpoints must be within 1.00 of each other. Center flatness plus the changes on both edges is this screen’s coarse proxy for a rounded arc, not a mathematical curvature estimate. Freeze the window, instrument, timeframe, feed and threshold before looking at later bars. These arbitrary values are teaching choices, not a validated detector.

Fixed-window examples in arbitrary price units.

Sequence	Closes at bars 1–9	Result under this screen
Broad U	100, 96, 93, 90, 89, 90, 93, 96, 100	Bottom candidate: center range 1; each edge change 7; depth 11
Sharp V	100, 100, 100, 100, 89, 100, 100, 100, 100	No bottom candidate: center range 11; each edge change 0
Inverted U	90, 94, 97, 100, 101, 100, 97, 94, 90	Top candidate only under a separately declared mirror rule

The broad U first becomes a candidate at the end of bar 9, not at its bar-5 low. For this teaching example, use the higher endpoint, 100, as the upside boundary and the center low, 89, as the pre-break failure boundary. A close strictly above 100 within the next three bars is an upward boundary event; a close below 89 first is a failure; otherwise the candidate expires unresolved. A wick past 100 with an inside close does not meet this close-based rule.

Following the broad U with closes 96, 98 and 101 records an upward boundary close at bar 12. The dip to 96 is a handle candidate only if a separate duration, depth and timing rule was fixed in advance. A direct bar-10 close of 101 could meet the rounded-bottom rule without forming a handle. A bar-10 close of 87 instead fails the candidate under the declared 89 boundary; keep it in the sample even if later prices make another attractive curve.

For the inverted U, a declared mirror screen would keep the center-three range within 2.00, require a rise of at least 4.00 from bar 1 to bar 3 and a fall of at least 4.00 from bar 7 to bar 9, put the center maximum at least 8.00 above each endpoint, and keep those endpoints within 1.00. The example then qualifies as a top candidate. For this example alone, use the lower endpoint, 90, as a downside boundary: a completed close strictly below 90 within the next three bars is a downward boundary event; a close above the center high of 101 first is a failure; otherwise the candidate expires unresolved. A wick below 90 with a close at or above it is not a close-based break.

Changing nine bars to thirteen, shifting the window until a successful example appears, or silently removing failures changes the sample. A visual shape or a conventional measured target is not a tested probability, executable entry or TradeCopier chart-detection feature.

A hypothetical candidate

Imagine prices forming a left rim near 100, declining over several observations to 84, then recovering toward a right rim of 99.50. A smaller pullback reaches 95. The broad cup depth measured from 100 is 16 units, while the handle depth measured from 99.50 is 4.50.

Suppose an illustrative rule requires a completed close above 100 to classify an upward break. A high of 100.80 followed by a close of 99.70 does not satisfy that rule. Changing the requirement to any intrabar trade above 100 after seeing the wick would create a different hypothesis.

Now suppose a later bar closes at 101 but the next observations fall to 97 and 93. That is a failed continuation under the stated example, even though the drawing was recognizable. It must remain in a test sample. Excluding it because the handle “was not ideal” after the decline introduces selection bias.

Measured targets are projections

A conventional projection might add the cup's depth to a breakout reference. In this example, 100 plus 16 gives 116. That number is geometric arithmetic, not a forecast with a known success probability. It also says nothing about adverse movement before a target, the time required or the actual ability to enter near the reference.

Volume comparisons need consistent sessions and instruments. A volume increase around a proposed break can be recorded objectively, but it does not identify participant intent. Comparing a full-day bar with an incomplete current bar without adjustment can produce a misleading volume statement.

Document a pattern study

Choose a maximum search window, permitted rim difference, minimum recovery and handle limits before scanning the evaluation period. Keep candidates that never break out as well as those that do. State how long a candidate remains eligible and whether overlapping formations can generate more than one trade.

Use a journal to preserve the original endpoints and timestamp. Compare double-top and bottom conventions to see how the same broad swings can receive different labels. Review selection bias in pattern tests before publishing a win rate drawn from hand-picked examples.