Harmonic Patterns

Harmonic patterns are chart formations defined by the Fibonacci relationships between consecutive price swings rather than by shape alone. Most are drawn as five points — X, A, B, C and D — joined by four legs, where each leg is measured as a retracement or extension of an earlier one. A pattern qualifies only when every one of those measurements falls inside a defined ratio window. That makes the family unusually explicit for something that looks discretionary on a chart: given a set of swing points and a tolerance, the ratios either qualify or they do not.

The underlying thesis is that a corrective sequence ends where several independent measurements converge. XA acts as the reference leg while AB, BC and CD trace the correction; the area where CD completes and projections from different legs overlap is usually called the Potential Reversal Zone (PRZ). Trades are taken against the CD leg inside that zone, invalidation sits just beyond the level that would break the geometry (commonly past X, or past D by a fixed buffer), and targets are expressed as retracements of the AD leg or as prior structural points such as C and A.

## How the geometry becomes a trade

Every implementation has to answer four mechanical questions before it can place an order. First, how swings are identified — a ZigZag, a fractal, or an N-bar pivot, each with its own depth setting. Second, how much ratio tolerance is allowed: a narrow band around 0.786 finds few patterns, a wide one finds many, most of them noise. Third, whether the PRZ is treated as a single price (a resting limit order at D) or as a zone entered in stages. Fourth, what confirms the reversal — nothing at all, a candlestick trigger, a momentum divergence, or an independent read of participation such as volume, delta or order flow. Those four choices produce very different trade populations from the same textbook definition.

## Main variants

The classic set builds on AB=CD, the simplest two-leg symmetry, extended into five-point structures. Gartley keeps D inside the XA range, around a 0.786 retracement with B near 0.618. Bat uses a shallower B and completes deeper, near 0.886 of XA. Butterfly and Crab are extension patterns whose D finishes beyond X — moderately for the Butterfly, far beyond for the Crab — which makes their stop geometry structurally different from the retracement types. Cypher, Shark and Three Drives use their own labelling and their own ratio tables. Exact numbers differ between sources, so a version's stated ratio table matters more than the pattern name it uses.

A second axis of variation is what the geometry is paired with. Purely geometric versions trade the PRZ on its own; hybrid versions treat the pattern as a location filter and require separate confirmation before entering. The decoded version linked from this page belongs to that second group, combining harmonic structure with an order-flow confirmation layer.

## What typically differentiates implementations

Beyond the ratio table: swing-detection sensitivity, whether patterns are validated at bar close or intrabar, minimum and maximum leg duration, whether overlapping or nested patterns may trade simultaneously, higher-timeframe direction filters, session and instrument restrictions, stop placement (fixed buffer past X versus ATR-scaled), and target logic (fixed Fibonacci levels of AD versus trailing). Position sizing also diverges, because extension and retracement patterns produce very different stop distances.

## Common mistakes

The most damaging is repainting. Swing detectors that need future bars to confirm a pivot will mark patterns that were not visible in real time, and a backtest built on them is measuring hindsight. Related errors: entering at D with no invalidation rule, so a failed pattern becomes an open-ended loss; tightening tolerances until only the historically successful patterns qualify; treating the PRZ as an exact price when it is a range; and ignoring that one swing sequence can host several overlapping patterns, which concentrates risk into a single move. Judging a version from a handful of screenshots is also unreliable, since valid patterns are rare enough that visual review is heavily selection-biased.

## How to evaluate and backtest a version

Code the detector deterministically and confirm patterns only with information available at the time. Log every pattern the rules detect, including the ones that fail — the failure population is the point of the exercise. Because valid patterns are infrequent, expect small samples: test across several instruments and a long history instead of optimising on one symbol, and break results down by pattern type, since a Gartley trade and a Crab trade are not the same trade. Run sensitivity checks on the two parameters that dominate outcomes, swing depth and ratio tolerance, and treat an edge that disappears when tolerance shifts slightly as a fitting artefact. Include realistic costs and slippage, particularly for versions resting limit orders at D. Finally, compare against a baseline — the same entries without the harmonic filter, for instance — so you can see whether the geometry is contributing anything beyond the trend or location it happens to select.

Strategies in this concept (4)

Frequently asked questions

What makes a chart pattern "harmonic" rather than just a shape?

The qualifying criteria are numeric, not visual. Each leg must fall inside a defined Fibonacci retracement or extension range relative to a previous leg, so two formations that look identical on a chart can be one valid pattern and one non-pattern. This is also why harmonic rules translate into code more directly than most price-action concepts.

What is the Potential Reversal Zone (PRZ), and how does it relate to the stop?

The PRZ is the area where projections from several legs converge and where the pattern is considered complete at point D. It is a range rather than a single price, which is why implementations differ on whether to enter with one order or in stages. The stop is normally placed beyond the level that would invalidate the geometry — often past X for retracement patterns, or past D plus a buffer for extension patterns — so the invalidation level, not a fixed risk figure, defines the trade's structure.

Why do two traders identify different patterns on the same chart?

Because pattern detection depends on swing detection, and there is no universal definition of a swing. A ZigZag with a different depth setting, or a pivot rule with a different lookback, produces a different set of X-A-B-C-D candidates from the same price series. Ratio tolerance adds a second source of divergence: what one version accepts as a 0.886 completion, another rejects.

Does a harmonic pattern need external confirmation, or is completion at D enough?

Both approaches exist and they are genuinely different strategies. Geometry-only versions enter at the PRZ and rely on the invalidation level to cap the cost of failures; confirmation-based versions treat the pattern as a location filter and wait for a separate signal — candlestick structure, momentum, volume or order flow — before committing. Confirmation reduces the number of trades and shifts the entry price, so the two variants should be evaluated separately rather than as one concept.

Why do harmonic backtests often look different from live behaviour?

The most common cause is a detector that confirms swing points using bars that had not printed yet, so patterns appear in the historical record that no one could have traded in real time. Two other causes are ratio tolerances tuned on the same data used to evaluate them, and limit entries at D that fill in a backtest but would be skipped or slipped in practice.

How much data is needed to assess a harmonic version fairly?

More than most concepts, because valid patterns are infrequent by construction — tight ratio requirements filter out the large majority of swing sequences. A realistic assessment usually means multiple instruments and several years of history, with results broken down by pattern type and by market regime, since a small overall sample can hide the fact that one pattern type contributed almost all of the trades.

All strategy concepts · Explore all strategies