How to Use Fibonacci Retracement

Michael Whitman · Trading since 2017, full-time since 2020 · Published 2026-09-08 · Updated 2026-10-22 · Reviewed 2026-10-22

The tool divides a move you selected into fixed percentages, so everything contentious about it lives in the word “selected”. Tested on 33.7 years of SPY swings further down this page, the levels showed no clear pull of their own, while two reasonable swing rules picked different lows under the same high in 32 of 43 up-swings and moved the 61.8% line a median 6.03% of price.

Before you start

A swing you can point at, with a high and a low you would defend to someone else. If the swing is ambiguous, every level derived from it is ambiguous by the same amount.

A decision about which levels you use, made before the tool is drawn. Three levels is a framework; seven levels means price is always near one.

The instrument’s ordinary bar range, so a reaction can be told apart from noise. Write down what an ordinary bar spans on this instrument and what a large one spans, because a bounce smaller than an ordinary bar is not a reaction to anything.

The steps

1. Identify one swing and write down its two ends

The high and the low of a move that is obvious enough that someone else would pick the same two points. Write the prices down before drawing anything.

2. Draw from the start of the move to the end of it

In an uptrend, low to high. In a downtrend, high to low. Reversing the direction produces a different set of numbers from the same two prices.

3. Use three levels and hide the rest

The 38.2, 50 and 61.8 lines are the conventional three. Adding more does not add information; it adds places for price to be near, which makes the tool unfalsifiable.

4. Look for confluence before treating a level as real

A retracement level sitting on a prior high, a moving average or a session level is a different event from one sitting in open space. The confluence is what carries the weight.

5. Wait for a reaction rather than anticipating one

Price reaching a level is not information. Price reaching it and then rejecting it over 2 or 3 bars is, and waiting costs a little of the move in exchange for evidence.

6. Refuse to redraw the swing after price moves

If the levels did not hold, the answer is that they did not hold. Redrawing from a different swing until one lines up produces a tool that has never once been wrong, which is the same as useless.

7. Take the stop beyond the swing, not beyond the level

The 61.8 line is a percentage. The swing low is a price where something happened. Put the stop beyond the structural point and let the distance set the size.

How to tell it worked

You can name the two prices the tool was drawn from without looking. If you cannot, the swing was not deliberate.

The number of levels displayed is 3. More than that and the tool cannot be wrong.

No swing was redrawn after price reached a level — 0 times across the period. Redrawing is the tell that the tool is being fitted to the answer.

And every entry waited at least 2 bars for a reaction before being taken, with a non-Fibonacci reason — a structure, a prior level, a session boundary — sitting at the same price.

Pick the swing with a rule, not by eye

A percentage reversal rule chooses the swing before any level is visible. Mark a swing high only once price has fallen a set percentage from it, and a swing low only once price has risen that much from it. The zigzag indicator draws exactly this, and the swing high and low page covers the bar-count versions.

The percentage decides the scale of the tool. On SPY daily bars from 29 Jan 1993 to 25 Sep 2026, a 5% rule marks 402 swings, about 11.9 a year. A 10% rule marks 87, about 2.6 a year. Neither is correct; they answer different questions, and the retracement levels they produce can sit far apart.

Write the rule down next to the chart. “5% swings on the daily” is a choice someone else can check. “The obvious low” is a choice made after looking, which is the opening the redrawing failure walks through.

A real swing, drawn two ways

SPY set a high of $760.40 on 2 Jun 2026. Under the 5% rule, the swing low before it was $629.28 on 30 Mar 2026. Under the 10% rule, it was $481.80 on 7 Apr 2025, because that rule marked no swing between the two dates.

The two drawings give two sets of levels from the same top. The 5% swing puts 38.2% at $710.31, 50% at $694.84 and 61.8% at $679.37. The 10% swing puts them at $653.97, $621.10 and $588.23. The two 61.8% lines are $91.14 apart.

The pullback that followed ended at $716.58 on 26 Jun 2026. That is a 33.4% retracement of the smaller swing, stopping $6.27 above its 38.2% line, and a 15.7% retracement of the larger one, nowhere near any of its levels. Same chart, same bars, two rules that would each have been defended.

What it does not claim

The ratios are not derived from anything about markets. They come from a numerical sequence, and the claim that price respects them is empirical rather than mathematical — which means it has to earn its keep against the cost of testing it.

Every level acted on costs a full round trip, and a strategy trading seven levels on every swing pays that repeatedly. Say $0.04 per round trip across seven levels on each of 10 swings a month: that is $2.80 a share before any level has been tested by price.

A swing formed on very little activity is a weak reference. The high and the low were set by a handful of participants, so the move being divided up was not a move many people took part in.

Why the swing choice matters so much

Two swings that both look reasonable produce levels several bar-ranges apart. For example, anchoring the same retracement at two nearby lows $2.00 apart moves the 61.8% level by $1.236, more than three bars on a stock whose ordinary bar spans $0.40, so the disagreement between two defensible drawings is bigger than the thing being measured.

Which is why the tool is a framework rather than a measurement. It organizes a view you already had; it does not supply one.

The original data

Across every 5% swing on SPY since 1993, the Fibonacci levels caught about as many pullbacks as any other lines of the same width. Using the rule above on daily highs and lows from 29 Jan 1993 to 25 Sep 2026, there are 400 pullbacks, each measured as a share of the move before it.

Exactly 200 of the 400, 50.0%, went past 100%, erasing the whole move before the next 5% swing, so every level drawn on those moves broke.

Part of that is built into the rule. A pullback only counts once it reaches 5% of price, so on the 233 moves smaller than 10% it had to retrace at least half the move to count at all, and 168 of those 233 went past 100%. Shallow pullbacks of small moves never show up.

Of the 200 that did not go past 100%, 38 ended within 2.5 percentage points of 38.2%, 50% or 61.8%. That is 19.0%; 2 of the 200 ended exactly at 100%. A 5-point window centered on any whole percent from 25% to 75% caught 12.1 on average, so three ordinary windows would catch about 36.3.

The 50% line did best with 17, and 6 of those 51 windows caught more. The 38.2% line caught 10 and the 61.8% line 11, and the busiest window of all was centered on 64%, with 22.

The rule has one choice worth stating. A bar that sets a new high cannot also confirm it; letting it do so finds 464 swings instead of 402, and 49 pullbacks near a level against about 42 for ordinary windows, a similar result. Prices are as traded, not adjusted for dividends.

The 10% rule leans a little further toward the levels, on far fewer cases. It finds 85 pullbacks, 44 of which went past 100%. Of the 41 that stopped inside, 12 ended near a level, against 9.0 for three average windows, a gap too small on 41 cases to separate from chance. Every pullback, both rules, is in the published CSV.

What this means for the steps above: on this test the levels caught little or nothing beyond what ordinary lines caught, so a reaction at one needs a second reason at the same price, which is step 4.

The June 2026 disagreement between the two rules was not unusual. Both rules were run over every completed 10% up-swing in SPY from 1993 to 25 Sep 2026, 43 of them, each running from a 10%-rule low to a 10%-rule high. The 5% rule marked a swing high on the same bar every time, so the only question is which low each rule anchors the tool to: the 10% rule’s low, or the 5% rule’s last swing low before that high.

In 11 of the 43 the two rules picked the same low and the drawings matched; in the other 32, 74.4%, they did not, and the two 61.8% lines sat a median 6.03% of the high apart. The gap was more than 5% of price in 17 of the 32 and more than 10% in 11.

The widest came under the 16 Jul 2007 high of $155.53, where the 10% rule anchored to the low of 12 Mar 2003 and the 5% rule to 14 Mar 2007, and the two 61.8% lines were $35.45 apart, 22.80% of the high.

The June 2026 pair is not one of the 43, but its gap was wider than all but 9 of the 32. Its top on 2 Jun is a 5%-rule high only, because the 10% rule’s up-swing from April 2025 was still open on 25 Sep 2026. Its 61.8% lines were $91.14 apart, 11.99% of the $760.40 high, and the chart below draws both sets of levels on the real daily closes, with the swing points at their intraday prices. Yahoo’s closes for the chart matched Nasdaq’s own price history on 433 of its 434 sessions. Every up-swing is in the two-rule table and every close in the chart file.

Of the 24,971 unique videos collected for this site’s search study, 381 have Fibonacci or fib in the title, at a median of 8,105 views across 306 channels. 93 of them name the retracement itself, and 189 of the 381 are instruction-shaped, at a median of 11,441 views. The relative strength index appears in 339 instruction-shaped titles at 7,043 and support and resistance in 77 at 41,998.

Among seven tool names measured the same way (RSI, MACD, moving averages, Fibonacci, VWAP, Bollinger Bands, support and resistance), only RSI (393 channels), MACD (337) and moving averages (334) are taught by more channels. A tool taught by that many separate people is exactly where a measured count like the one above is worth more than another tutorial.

When it fails

The failure is a tool redrawn until it fits, and it never feels dishonest. Price ignores the levels, a different swing is selected — equally defensible on its own terms — and the new levels line up beautifully with what already happened. On SPY in June 2026, switching from the 5% rule’s low to the 10% rule’s moved the three lines by $56.34, $73.74 and $91.14 at once. Repeated a few times, the tool has an unbroken record and has never made a prediction, because the input was chosen after the outcome was visible.

The second failure is displaying every level. Price is then always near one.

A third is trading a level with no confluence. The arithmetic alone carries no weight.

A fourth is placing the stop just beyond a level. It is a percentage, not a price anyone defended.

A fifth is drawing on an ambiguous swing. The ambiguity propagates into every level.

And a sixth is anticipating the reaction. Reaching the level is not the evidence.

Related

Fibonacci covers where the ratios come from and what they do not claim. Fibonacci extension is the same tool pointed at targets rather than pullbacks. And retracement is the move itself, independent of any tool. The golden ratio page covers the 61.8% number on its own, and support and resistance is the structure a level needs beside it.

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