Harmonic patterns take the retracement tool one step further: instead of measuring a single pullback, they demand that a whole sequence of swings land on specific Fibonacci ratios, and in exchange they promise something rare — a reversal zone defined to the pip, with the stop already implied.
The XABCD skeleton
Every harmonic pattern is drawn on five points: X, A, B, C and D. XA is the initial move. AB retraces it. BC retraces AB. CD extends to the Potential Reversal Zone (PRZ) — the area where the pattern says the sequence completes and price should turn. The trade is always the same shape: enter at D against the direction of CD, stop beyond the zone, first target back near point C.
What distinguishes the named patterns is only which ratios each leg must hit:
- Gartley — B retraces 61.8% of XA; D completes at 78.6% of XA. The original, from H.M. Gartley's 1935 book.
- Bat — B is shallower (38.2–50%), D completes deep at 88.6% of XA.
- Butterfly — D extends beyond X, typically 127% of XA: you are fading a fresh high or low.
- Crab — the extreme case, D at 161.8% of XA; the deepest stretch and, its advocates argue, the sharpest reactions.
Bullish versions complete below, bearish versions above; the geometry is symmetric. The ratios come from the same Fibonacci family as ordinary retracements — 0.382, 0.618, 0.786 (√0.618), 0.886 (√0.786), 1.272 (√1.618), 1.618.
The idea underneath the geometry
Strip the exotic names away and each harmonic is a structured way of buying a pullback or fading an overextension at a confluence of measured levels. Point D is powerful — when it is — because several measurements land in one small area: the XA retracement, a BC extension, and often an old structure level. Confluence is a real phenomenon: more independent reasons for orders to rest in one place means a bigger reaction when price arrives. Harmonics are a formalized confluence detector, and that is the defensible core of the method.
The honest critique
The precision is more marketed than proven. There is no rigorous public evidence that price reverses at 88.6% retracements more often than chance once you account for how much tolerance pattern-finders allow — and the tolerance is the trick. Scanners and educators routinely accept legs within a few percent of the ideal ratio, and with five points and four named patterns (plus Shark, Cypher and friends), some harmonic can be fitted to a large fraction of all price paths. Add the fractal argument — patterns exist on every timeframe — and you have a method that can never run out of setups, which should make you suspicious, not excited.
There is also a practical hazard: every harmonic entry is counter-trend by construction. Point D of a bearish Butterfly is, by definition, a new high — you are shorting strength because a ratio says so. When the pattern fails, it fails into a trend, and the losses are fast. The stop beyond the PRZ is not optional decoration; it is the only thing separating this method from "fade every breakout."
If you trade them anyway
Three habits keep harmonics honest. Demand structure agreement: take a D-point only where it lands on something that would matter anyway — a prior swing, a weekly level, a round number. Wait for a trigger: let the zone produce a rejection candle or a lower-timeframe structure break before entering, rather than catching the knife at the ratio. Log the misses: every pattern your scanner finds, not just the pretty ones that worked — the completion rate you measure yourself is the only statistic about harmonics you should trust.
Treated as a precision oracle, harmonics will disappoint you exactly as often as any other tool that promises certainty. Treated as a disciplined map of where measured pullbacks cluster — with entries triggered by price behaviour, not geometry alone — they are a workable, if elaborate, way to buy low and sell high with defined risk.