Do Fibonacci retracements work? We tested 22,400 pullbacks
7 min read · Updated 2026-10-04
Short answer: on our data, no. Across 22,400 swings on 26 markets, pullbacks did not end at the 38.2%, 50% or 61.8% Fibonacci levels more often than at any other depth nearby, and price did not turn at those levels more often than at arbitrary levels in between. What the same test does show is how deep pullbacks usually go - and that is worth knowing when you place a stop.
Fibonacci retracement is one of the most popular tools on any chart, so the claim deserves a real test rather than a few hand-picked examples. Here is what we did and what came out.
After a market makes a clear move - a swing from a low to a high, say - traders measure how much of that move price gives back before it continues. The Fibonacci tool draws horizontal lines at fixed fractions of the move: 23.6%, 38.2%, 50%, 61.8% and 78.6%. The 61.8% line is the famous one, the so-called golden ratio. The idea is that pullbacks tend to stop at these lines, so they are used as places to enter, and stops are put just beyond them.
A real example, redrawn every hour: a recent gold swing of at least 3 ATR whose pullback stayed inside the move (white line, 4,224.8 → 4,143.1), the Fibonacci lines the tool draws across it, and how far price came back (blue dot). In this one the pullback gave back 65% of the move before price went on. One swing proves nothing either way - that is why the test below counts 22,400 of them.
How we tested it
Data: about two years of 1-hour candles on 26 markets - 10 forex pairs, gold, silver, WTI and Brent oil, natural gas, Nasdaq-100, Nikkei and DAX, the S&P 500, four crypto pairs and three US stocks. Roughly 370,000 candles.
Swings: a swing high or low is the extreme of the five candles either side of it; a move counts when it spans at least three times the average true range (ATR). That gave 22,400 swings.
No hindsight where a decision is made: a level only counts from the candle on which the swing could have been known.
A control: every Fibonacci level is compared with ordinary levels in between (30%, 44%, 56%, 70%). If Fibonacci numbers are special, they should beat their neighbours.
Test 1: do pullbacks end at Fibonacci levels?
If the ratios mattered, the depths at which pullbacks end would pile up around 38.2%, 50% and 61.8%. They do not. The depths form a smooth plateau between about 35% and 65% of the move, with no bump at any of the three levels: the number of pullbacks ending right at 38.2%, 50% and 61.8% was 1.07, 1.00 and 0.97 times the number ending just beside them. For comparison, 400 randomly chosen depths gave between 0.89 and 1.12. The Fibonacci levels sit inside the range of pure chance.
Where pullbacks ended (columns: number of pullbacks per 5% of depth; orange lines: the three Fibonacci levels). If the levels stopped price, the columns under the lines would stand above their neighbours. They do not: it is one flat top from about 35% to 65%. Shown: the 11,616 pullbacks that ended before the whole move was given back.The numbers
Depth
Pullbacks ended
0–5%
3
5–10%
12
10–15%
119
15–20%
235
20–25%
405
25–30%
648
30–35%
754
35–40%
821
40–45%
819
45–50%
873
50–55%
855
55–60%
866
60–65%
840
65–70%
796
70–75%
706
75–80%
691
80–85%
611
85–90%
563
90–95%
541
95–100%
458
Test 2: does price turn when it reaches a level?
Maybe pullbacks do not end there, but price at least reacts? At the first touch of each level we checked whether price moved one ATR back in the direction of the trend before it moved one ATR further against it. At the Fibonacci levels that happened 47.6% of the time (15,158 touches). At the ordinary levels in between: 47.6% (19,410 touches). The difference is -0.1 percentage points, with a margin of 1.1 either way.
How often price turned when it first touched a level (moved 1 ATR back with the trend before 1 ATR further against it). Fibonacci levels 38.2 / 50 / 61.8 together: 47.6% of 15,158 touches. Ordinary levels in between: 47.6% of 19,410. Every column is the same height within the margin of chance.The numbers
Level
Kind
Price turned
Touches
23.6%
Fibonacci
47.9%
1,118
30%
ordinary
47.3%
2,113
38.2%
Fibonacci
47.4%
3,404
44%
ordinary
46.4%
4,338
50%
Fibonacci
47.0%
5,219
56%
ordinary
48.3%
5,933
61.8%
Fibonacci
48.1%
6,535
70%
ordinary
47.9%
7,026
78.6%
Fibonacci
46.7%
7,396
We changed the definition of a swing six ways - tighter and looser pivots, smaller and larger moves - and the difference stayed between -0.7 and +0.2 points, always inside the margin of chance. We also checked whether a Fibonacci level that coincides with one of our own modelled stop clusters does better: 49.4% against 49.8% without. Nothing there either.
What is real: how deep pullbacks go
The useful result is the plain distribution. Of 22,400 swings, this share saw the pullback reach at least this depth before the trend resumed: 23.6% of the move in 97% of swings, 38.2% in 88%, 50% in 79%, 61.8% in 70%, 78.6% in 58%, and the whole move given back in 48%.
How deep pullbacks go before the trend moves on. Read it as stop odds: a stop just beyond the 61.8% line is reached in about seven pullbacks out of ten; almost half of all swings are given back completely.
These numbers were within two or three points of each other in forex, metals and energy, indices, crypto and stocks. So a stop placed just behind the 61.8% line is reached in about seven pullbacks out of ten, and almost half of all swings are fully retraced before the trend continues, if it continues at all. A level that most pullbacks walk through is not much of a wall.
What that means in practice
The Fibonacci lines are a convenient ruler for describing how deep a pullback is. On our data they are not places where price is more likely to turn.
If you use them, know the base rates above: most pullbacks go deeper than 50%, and a stop just beyond 61.8% is hit more often than not.
Any level looks good on a chart chosen after the fact. The only fair test is every swing, with a control beside it.
We run the same kind of test on our own levels and publish the result, including where they show no edge: see the methodology page. How far a stop has to be to survive ordinary movement, measured in ATR, is on each market's stop-loss page.
Educational content, not financial or investment advice. Nothing here is a recommendation to buy or sell anything. Figures describe history and models, not what will happen. Trading leveraged products such as CFDs, forex and crypto derivatives carries a high risk of losing money.
See it on a chart
The Gold liquidation heatmap is free on a 24-hour delay, with no card — and live with a 7-day free trial.