Golden Ratio: The Number, and the Claim

Michael Whitman · Trading since 2017, full-time since 2020 · Reviewed 2026-09-06

How it works

The golden ratio is a specific number: 1.6180339887, and onwards. It is defined by a simple property — a line divided so that the whole is to the larger part as the larger part is to the smaller.

The Fibonacci sequence is where traders meet it. 1, 1, 2, 3, 5, 8, 13, 21, and each term after that is the sum of the two before it.

Divide any term by the one before it and the answer converges. 21 divided by 13 is 1.615; 89 divided by 55 is 1.618. Its reciprocal is 0.618, which is where the retracement level comes from.

The mathematics and the claim are different things

Nothing above is controversial. The dispute is entirely about a further claim: that financial prices retrace by these particular fractions more often than by others.

That claim has a shape you can test. Take every pullback in a series, express its depth as a fraction of the move it interrupts, and see whether the depths cluster near 0.618 or spread out evenly.

Run on this site’s shared history, the band from 0.593 to 0.643 contained 2 of 43 pullbacks — 4.7%. A band that wide covers 5% of the range from 0 to 1, so 4.7% is almost exactly the share an even spread would put there.

The bands at 0.382, 0.500 and 0.786 each contained 2 as well. Four levels, four identical counts, no level standing out from its neighbours.

What that number does and does not mean

The series it was measured on was generated, and no Fibonacci rule went into generating it. So the flat result is what it should be — this is a control, showing what the measurement looks like when the effect is absent by construction. It is not evidence about the S&P 500, and 43 pullbacks is too small a sample to settle anything even if it were.

What it is good for is calibration. Most people have never seen the null version of this test, so they have nothing to compare a real result against. If you run the same measurement on your own instrument and get 5%, you have found nothing — and knowing that in advance is worth more than the test itself.

In practice

Participation is the check that costs nothing. If a level matters, unusual volume should appear when price reaches it. That is a testable prediction and a stronger one than the level itself.

The ratio is scale-free, which cuts both ways. It applies identically on every timeframe, and so does any absence of effect.

A gap crosses the level without a single trade occurring at it, which is a reminder that a level is a coordinate rather than a barrier.

Stops belong under structure, not under a fraction. The swing low is a place where something actually happened; 0.618 of the way back is a place where nothing has.

Costs do not care which theory produced the entry. 2% of a median bar’s range per round trip on this history, whatever line you drew to justify it.

One argument for the ratio survives all of this, and it is the reflexive one. The levels are on every charting platform by default, drawn from the same obvious swings, so a large number of traders place orders at the same prices. That is a real mechanism for a level to matter, and it has nothing to do with mathematics or nature.

It also predicts exactly where the effect should be strongest. On liquid instruments with one unambiguous swing that everybody would measure the same way — and weakest where the swing is arguable, which is most of the time. Stated that way the claim becomes checkable, which is more than can be said for the version about shells.

What the golden ratio is not

It is not a law of markets. The mathematics does not extend to price by itself.

It is not 50%. Half is not a Fibonacci ratio and is on the tool anyway.

It is not disproved either. A control on a generated series settles nothing about real ones.

And it is not useless. A shared reference point is a real mechanism.

When it fails

In a range every fraction of every swing gets touched repeatedly, so the level appears to work and is simply being crossed by a market going nowhere.

The second failure is the nature argument. Sunflowers and shells are used to make the market claim feel established, and they are unrelated to it.

A third is retrofitting the swing. Given a turn, a swing can usually be found that puts 0.618 near it.

A fourth is the confidence transfer. The certainty belongs to the arithmetic and gets spent on the prediction.

And a fifth is arguing about it instead of measuring it. The test takes an afternoon and produces a number.

The original data

On this site’s shared 576-bar history, 43 pullbacks that held above their origin were measured on a one per cent zigzag. The bands at 0.382, 0.500, 0.618 and 0.786, each 0.05 wide, contained 2 pullbacks apiece — 4.7% each, against the 5% an even spread implies. The median depth was 0.708. The figures are in research/series-measurements.json, produced by site/measure_series.py.

The median depth is the more interesting figure. Pullbacks on this series ran deeper than the level most often quoted, which means a trader waiting at 0.618 was frequently waiting above where price actually went. Measure your own market’s median pullback depth before deciding which level to wait at — the number is specific to the instrument and the timeframe, and it is the one piece of this that is genuinely worth knowing.

Related

Fibonacci is the tool the ratio lives inside. Retracement is the move being measured. And technical analysis is the wider frame this argument sits in.

This page is educational, not financial advice. Test every idea on your own charts before risking money.