A Fibonacci retracement is a set of horizontal lines drawn across one finished price move, marking how much of that move a pullback has given back. The ratios are fixed arithmetic, so every number can be checked with a calculator.

The one decision you make is where to put the two ends, and it changes everything downstream. Two traders who anchor the same leg slightly differently get different prices on every line, which is why no published success rate for this tool transfers to your chart.

What a Fibonacci retracement measures

It is a measuring tool. You pick a move that has finished, mark where it started and where it ended, and the tool divides the distance between them at fixed fractions.

LuxAlgo's library page puts the mechanism plainly: "The grid is only as good as the leg it is anchored to, so most of the work is choosing anchors."

StockCharts is just as direct about what the lines are: "These retracement levels are not hard reversal points. Instead, they serve as alert zones for a potential reversal. At this point, traders should employ other aspects of technical analysis to identify or confirm a reversal."

Which Fibonacci retracement levels are standard?

Four of them: 23.6%, 38.2%, 50% and 61.8%. After that the published lists disagree. LuxAlgo, Schwab and OANDA all carry 78.6%, though OANDA files it under less commonly used and LuxAlgo notes that some platforms substitute 76.4% for it. Schwab also publishes 161.8%. OANDA puts 100% among its most-used levels, and it is alone in listing 11.6% at all. StockCharts' guide works with the core four and never mentions 78.6%.

So if you copy a level set from a guide and your platform draws a different one, your chart and the guide's chart are not showing the same thing. Decide which lines you want, then keep that set.

Why is 50% on the tool when it is not a Fibonacci number?

Because it arrived from somewhere else and stayed.

StockCharts traces it: "The 50% retracement is not based on a Fibonacci number. Instead, this number stems from Dow Theory's assertion that the Averages often retrace half their prior move."

LuxAlgo says the same in its own words: "The 50% level is not a Fibonacci ratio at all; it persists by convention from older halfway-back rules." Schwab makes the same split and puts 100% on the same side of it, describing both as levels many traders consider psychologically significant.

Draw the 50% line if you want it. Plenty of people watch it, which is a reason to have it on the chart. Just do not claim arithmetic on its behalf.

What is the golden ratio, and what is the golden zone?

The golden ratio is roughly 1.618. Its inverse, 0.618, is where the 61.8% level comes from, and that is the line usually called golden. StockCharts derives it from the sequence: divide a number by the next one along and you approach 0.618, with the approximation tightening as the numbers grow. Square 0.618 for 0.382, cube 0.618 for 0.236, and the square root of 0.618 is 0.786.

The golden zone, or golden pocket, is a band rather than a line, and its width depends on who you read. LuxAlgo puts it between the 61.8% and roughly 65% retracements, a slice of the same tool that many strategies single out as the highest-interest entry area. StockCharts draws a wider box, calling 38.2% to 61.8% the zone that covers the most possibilities. Pick one width and stay with it.

How do you draw a Fibonacci retracement, step by step?

Find a move that has already finished, put one anchor at each end, and read the lines between them.

OANDA gives the direction rule for an upmove: the 0 line goes at the swing low, the trough of the move, and the 1 line at the swing high, the peak. For a downmove you reverse it, anchoring at the high first and the low second.

Round numbers make the arithmetic visible. The move below is an illustration, not a real chart and not a trade. It runs from a low of 1.1000 to a high of 1.2000, so the leg is 0.1000 tall, and each level sits at the high minus the ratio times the leg.

Level	Arithmetic	Price
23.6%	1.2000 - 0.0236	1.1764
38.2%	1.2000 - 0.0382	1.1618
50%	1.2000 - 0.0500	1.1500
61.8%	1.2000 - 0.0618	1.1382
78.6%	1.2000 - 0.0786	1.1214

For a downmove from 1.2000 to 1.1000 the arithmetic flips: you add the ratio times the leg to the low, so the 61.8% level sits at 1.1000 plus 0.0618, which is 1.1618.

Wick or body: where do the two anchors go?

Wherever you decide, as long as you decide once. Anchor the same move to candle bodies instead of wicks. Say the body low is 1.1050 and the body high is 1.1950, so the leg is 0.0900 rather than 0.1000.

Level	Wick anchors	Body anchors	Gap
23.6%	1.1764	1.1738	26 pips
38.2%	1.1618	1.1606	12 pips
50%	1.1500	1.1500	none
61.8%	1.1382	1.1394	12 pips
78.6%	1.1214	1.1243	29 pips

Every line except one has moved, and the lines furthest from the middle move most. The 50% line sits still here only because this example trims both ends by the same amount, which will rarely be true on a real chart.

LuxAlgo states the rule and the reason: "Wick-to-wick is the most common convention and body-to-body the main alternative; mixing them shifts every level." Its FAQ adds: "Both are defensible; consistency is what matters, because switching conventions moves every level." It offers a third option, drawing both grids and treating the gap between them as a tolerance band around each level, and it gives permission to walk away: "If the swings are ambiguous and overlapping, the honest move is to skip the tool rather than force a grid."

Treat anchoring as a convention you adopt and apply the same way every time. That is worth saving as a default rather than deciding fresh on each chart.

What is the difference between a retracement and an extension?

A retracement divides the move you already have. An extension projects past the end of it. LuxAlgo draws the boundary: "Ratios above 1 (1.272, 1.618) project beyond the leg and belong to the extension tool, not retracement."

OANDA says the extension uses the same ratios with 100% added: "For example, Fibonacci extension levels will include 123.6%, 138.2%, 161.8%, etc." Which one to use it leaves to the trader's discretion, and its remark that 161.8% tends to see stronger reactions is labelled a rule of thumb and illustrated with chart examples rather than a measurement.

Test the retracement first. The extension adds a target choice on top of the anchors, which is a second decision to get wrong.

Does Fibonacci retracement work, and what is its success rate?

There is no rate worth printing, and you should be wary of any page that gives you one. Published tests pick their anchors differently, so they are not measuring the same thing.

One test drew anchors by eye across the Nasdaq 100 stocks and the Nasdaq 100 and S&P 500 indices, over charts covering the 2022 crash and recovery, and concluded against the tool. A second handed the swing choice to a zig zag indicator over 100 trades, counted a reaction as anything within 5 percent either side of a level, never named its instrument or timeframe, and also came out against. A third used a session rule almost nobody else uses, where the previous day's closing colour sets the orientation, on one US index contract over a year, and concluded in favour.

tradingrush, which concluded against, names the reason in passing: "Some swing lows and swing highs according to the zig zag indicator were not accurate, but since swings of a trend can differ from trader to trader, this small number of less accurate swings were fine."

On whether the levels have any hold on markets at all, LuxAlgo's FAQ is blunt: "It is contested. There is no accepted evidence that markets obey golden-ratio mathematics, but the levels are watched widely enough that reactions near them are common and partly self-fulfilling."

How do you test Fibonacci retracement on your own chart?

By drawing it on your instrument, on your timeframe, stepping the chart forward one bar at a time without knowing what comes next.

One argument for doing it by hand comes from a source that concluded the tool does not work. liberatedstocktrader writes: "Backtesting Fibonacci in trading is still a manual process today. The best backtesting software, like TrendSpider, TradingView, and Trade Idea, has not incorporated this Fibonacci into its backtesting engines. Why? Because the rules for Fibonacci trading are so loose and open to individual interpretation, it defies systemization. Like Elliott Waves, Fibonacci is more of a drawing tool than a tested indicator and should be used as such when backtesting." The same page recommends TrendSpider elsewhere as its author's own pick, so a page promoting the software is telling you the software cannot do this.

The procedure:

Write the rule down before you draw anything. Which leg counts as a leg, wick or body, which levels you will watch, what has to happen at a level before you act, and where the trade is wrong.
Open a historical session on the instrument you actually trade, and start it somewhere you do not remember.
Step forward bar by bar. Draw the grid when the rule says a leg has completed, rather than when you can see a bounce coming.
Take the trade with a real order, a stop and a target, so the result includes your exit as well as your entry.
Write down why you drew the grid where you drew it, before you know the outcome.
Repeat until you have enough trades to say something. The beginner's guide to backtesting a trading strategy covers the general method.

ChartLabs replays historical charts bar by bar in a browser with your own trades on them, which is the loop above. The Fibonacci retracement tool sits on Alt+F, or Option+F on a Mac, which sounds minor until you are drawing a grid on every leg across two hundred bars. Drawing style is reusable, the practical answer to the consistency problem: right-click a drawing for Apply defaults to reset it to your saved look, or switch on Remember last used so new drawings start from the style you used last.

Our markets are forex, indices, commodities including gold and oil, and crypto. Worth noting against the tests above: the first ran on US equities and equity indices, the third on a US index contract, and the second never said what it traded. Our data comes from Dukascopy, tick-level where it is available, cleaned to remove anomalies, and depth is per symbol, up to around 22 years on the oldest forex pairs, with other instruments starting later.

Record why you drew each grid where you drew it, because a log of wins and losses alone cannot tell you whether your anchoring was the problem. Note the leg, wicks or bodies, the level you acted at, what confirmed it, and how you exited. ChartLabs keeps that on the trade: a Notes column in the Orders panel, a note button on the chart while a trade is open, and notes in the CSV export. The Orders panel shows up to five tags per trade, so you can tag by level or by convention and split the results later. Set commission and slippage per symbol too, which feed into P&L, balances and a Trading Costs card in analytics. Level-to-level targets are often short, and a test that ignores costs flatters short trades most. Backtesting Accuracy goes through that properly.

What goes wrong when you test it by hand?

You already know where the leg ended.

On a chart you have scrolled through, the swing high is obvious because it has already happened. You place the anchor knowing the answer, the grid lands where the bounce was, and the test measures your hindsight instead of your rule.

tradingrush, testing against the tool, says exactly this: "Furthermore, backtesting the fibonacci tool can be really difficult because fibonacci requires you to draw from the swing low to the swing high and vice versa. When you try to backtest, this swing low and swing high will become very biased."

Start somewhere you did not choose: a session can pick random dates by length and period, or you can hand-pick a range, and a period you picked yourself is a period you may remember. Write the anchoring rule while there is no chart in front of you, because a rule you cannot state without looking at one is an opinion that will move. And take the losing trades, since the bias is in which grids you count as well as in where you draw them.

Doing this by hand trains the judgement the tool requires. Manual Backtesting Builds Better Traders makes that case in full.

Try it on your own chart

Fibonacci retracement is fixed arithmetic wrapped around one judgement call, and nobody else's number tells you whether your version of that judgement holds up.

ChartLabs is one plan at $20 a month or $200 a year, with a 7-day free trial and no credit card required to start.

Start your free trial

Fibonacci Retracement Examples: Levels, Anchors, Testing