Williams Fractals

A five-bar swing pattern that marks local highs and lows — always confirmed two bars after the fact, which is where most implementations quietly cheat.

What it is

Williams Fractals, popularized by Bill Williams, are a simple pattern definition for local swing points. An up (bearish) fractal is a bar whose high is higher than the highs of the two bars on either side of it; a down (bullish) fractal is a bar whose low is lower than the lows of the two bars on either side. Fractals are used to mark swing structure, place reference levels for breakouts and stops, and feed other Bill Williams tools such as the Alligator. The critical mechanical fact — and the reason fractal signals appear 'late' — is that a fractal cannot exist until two more bars have closed after the candidate bar. The pattern is defined by its neighbors, so it is unknowable in real time. Most charting platforms then draw the arrow back on the fractal bar itself, two bars in the past, which looks clean but misrepresents when the information became available. Fractals are descriptive markers of past swings, not predictions; this is educational content only, not financial advice, and all trading carries risk of loss.

How it works

The standard fractal uses a five-bar window. For an up fractal at bar N, the condition is: high[N] > high[N−1], high[N] > high[N−2], high[N] > high[N+1], and high[N] > high[N+2] (variants allow equal highs on one side). The down fractal mirrors this with lows. Because the condition references the two bars after the candidate, the earliest moment the pattern can be evaluated is at the close of bar N+2 — a fixed two-bar confirmation delay. In Pine Script this is exactly what ta.pivothigh(high, 2, 2) computes: it returns the pivot value on the confirmation bar, two bars after the pivot printed. The trap is in the plotting: the conventional display offsets the marker two bars back so the arrow sits on the swing itself. Visually that is fine, but any backtest, alert, or strategy that reads the offset series as if it were available on the fractal bar is looking two bars into the future. The correct, non-lookahead approach evaluates and acts on the confirmation bar, accepting that price may already have moved meaningfully away from the swing by then.

How traders read it

• Up fractal above a bar: that bar's high exceeded the two highs on each side — a confirmed local swing high, known two bars after it formed.
• Down fractal below a bar: a confirmed local swing low under the same two-sided rule.
• Reference levels: some traders track the most recent confirmed fractal high/low as breakout or invalidation levels — these are descriptions of past structure, not instructions to act.
• Timing honesty: whatever you do with a fractal, the earliest bar you could have done it on is the confirmation bar (fractal bar + 2), never the arrow's visual position.
• Density: fractals appear constantly on every chart; most mark minor noise, and there is no built-in ranking of importance.
• Context: fractals are usually read together with trend context (for example the Alligator or a moving average) rather than in isolation, and always alongside a risk plan.

Common settings

The classic definition is fixed at two bars on each side (a five-bar pattern) and has no adjustable inputs on most platforms. Generalized pivot versions let you widen the window — for example 3–5 bars per side — which yields fewer, larger swings at the cost of a proportionally longer confirmation delay: n bars per side always means an n-bar wait. Applies to any market and timeframe.

Strengths

• Completely objective and deterministic: the pattern either holds or it does not, with no subjective judgment
• Simple five-bar logic that is easy to verify by eye and trivial to code correctly
• Provides concrete swing-high/swing-low reference levels for structure analysis
• Fixed, known confirmation delay (two bars) — unlike ZigZag, the lag never grows with the size of the move
• Available as a built-in on MetaTrader and easily reproduced in Pine Script via ta.pivothigh/ta.pivotlow

Pitfalls to watch

• Two-bar confirmation delay: by the time a fractal exists, price has printed two full bars beyond the swing — the marker's position and its availability are different moments in time.
• Lookahead in naive backtests: reading the visually offset fractal series as tradeable on the fractal bar peeks two bars into the future and inflates results that cannot be achieved live.
• Signal density: fractals mark every minor five-bar wiggle, so most of them are noise without an additional filter.
• The unconfirmed candidate repaints: a forming fractal on the last two bars can vanish if a neighbor exceeds it — only evaluate on confirmed bars.
• Equal-high/equal-low ambiguity: implementations differ on whether ties count, so two platforms can mark different fractals on identical data.
• No predictive power: a fractal says a local extreme occurred, nothing about whether it will hold; no indicator removes the risk of loss.

Pine v6 example

//@version=6
indicator("Williams Fractals (Non-Lookahead)", overlay = true)

n = input.int(2, "Bars Each Side", minval = 1)

// ta.pivothigh/ta.pivotlow only return a value once the pattern is
// confirmed n bars after the candidate bar — the same 2-bar delay the
// classic 5-bar fractal has by definition.
upFractal   = not na(ta.pivothigh(high, n, n))
downFractal = not na(ta.pivotlow(low, n, n))

// Marks are placed on the CONFIRMATION bar, not offset back onto the
// fractal bar. Less pretty, but it shows when the signal truly existed —
// a live trader could not have acted any earlier.
plotshape(upFractal and barstate.isconfirmed, "Up Fractal Confirmed", style = shape.triangledown, location = location.abovebar, color = color.red, size = size.tiny)
plotshape(downFractal and barstate.isconfirmed, "Down Fractal Confirmed", style = shape.triangleup, location = location.belowbar, color = color.green, size = size.tiny)

alertcondition((upFractal or downFractal) and barstate.isconfirmed, "Fractal Confirmed", "A Williams fractal was confirmed 2 bars after the swing.")

Pro tip: When testing anything built on fractals, compare two versions of your logic: one that acts on the confirmation bar and one that (incorrectly) acts on the fractal bar itself. The performance gap between them is pure lookahead — information you never had. If a published fractal strategy only works with the offset version, it does not work. Educational information only, not financial advice; all trading involves risk of loss.

Educational only — not financial advice, not a recommendation to trade. No indicator is predictive; trading involves substantial risk of loss.