A Fibonacci retracement is a grid of levels placed between a move's origin and its extreme, to locate how far price can come back before moving on. Its most followed levels, 0.382, 0.5 and 0.618, are computed as a share of the move, and two of them come from the Fibonacci sequence.
The pullback measured on the move
Contents
The Fibonacci retracement starts from an ordinary observation: a rising market does not rise in a straight line, it regularly retraces before continuing. The tool gives a way to put numbers on that pullback. You take the move's origin and its extreme, and draw between them levels expressed as a share of the move: 0.382 means price has given back 38.2% of what it took.
Those shares come from the Fibonacci sequence: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144 and so on, each term being the sum of the two before it. Leonardo of Pisa, known as Fibonacci, introduced it to Europe in 1202 in his Liber Abaci, through a problem about breeding rabbits, quoted in his biography by the University of St Andrews. It was only tied to markets in the twentieth century, notably by Ralph Nelson Elliott and his wave theory.
What trading keeps from it is the ratios between its terms. Divide a term by the next one and the result settles around 0.618: 55 / 89 = 0.618, 89 / 144 = 0.618. That is the inverse of the golden ratio, 1.618. Divide a term by the one two places further on and you get 0.382 (55 / 144); three places further on, 0.236 (34 / 144). These are the 61.8%, 38.2% and 23.6% that platforms display, and the larger the terms, the closer the ratio between two neighbours gets to the golden ratio, a property set out on Wolfram MathWorld's Fibonacci Number page.
The grid's other two levels are not ratios of the sequence. 0.786 is the square root of 0.618, a derivative of the golden ratio. And 0.5 has nothing to do with Fibonacci: it is half the move, a marker inherited from Dow theory, for which a correction commonly gives back between one third and two thirds of the previous move, with half as the typical amount.
Retracement = extreme − (extreme − origin) × ratio
Extension = origin + (extreme − origin) × ratio
Drawing the grid means designating those two points. In an uptrend, you click on the low and drag to the high: level 0 sits on the high, level 1 on the low, and pullbacks are read going down. In a downtrend, you start from the high and drag to the low, and the levels are read going up.
The two points are usually taken on the wicks, that is on the highest and lowest prices actually traded. Some measure on closes: both conventions are defensible, provided you do not switch from one to the other between charts.
The formula is the same in both directions. On a bearish move, the origin is the high and the extreme is the low: the difference turns negative, and the level comes out above the low.
A retracement's ratio is below 1, an extension's is above it: 1.272, which is the square root of 1.618, and 1.618 itself.
| Level reached | How it reads |
|---|---|
| 0.236 | Shallow pullback: the move resumes almost without pausing. Rarely an entry area, more the sign of a strong trend. |
| 0.382 | Light pullback, typical of sustained trends. Entering here leaves a long stop, since it stays beyond the move's origin. |
| 0.5 | Half the move. In ICT vocabulary it is equilibrium: premium zone above, discount zone below. |
| 0.618 | Deep pullback: price has given back close to two thirds of the move, and the stop placed beyond the origin becomes short. |
| 0.786 | The last level before the origin. Short risk, but the scenario is close to being invalidated. |
| beyond 1 | The origin is broken: the move is erased. That is no longer a retracement, it is a reversal. |
The table says something the tool's presentations almost always leave out: a deep level is not better than a shallow one, it is different. The deeper the pullback, the shorter the risk per trade, and the less often the level is reached. It is a trade-off, not a hierarchy, and it is settled with numbers rather than with a preference.
A level is never read on its own. It is read first with market structure: a pullback is only worth buying in a trend that is still making higher highs and higher lows, and the move to measure is the one that has just broken the previous high, not any rally.
It is then read with whatever sits at the same price. A level that coincides with a former high turned support, a fair value gap or an order block forms a confluence: two independent reasons to expect a reaction in the same place.
A buy at 1.0938 (the 0.618 level) with a stop at 1.0890, ten pips below the low, risks 48 pips. With the 1.272 extension at 1.1027 as the target, the gain aimed for is 89 pips, or 1.85 R. The same idea executed at 1.0921 (the 0.786 level) risks 31 pips to aim for 106, or 3.4 R, but price has to drop 17 pips further to fill you, which happens less often.
This is the original use. Rather than entering by feel during a pullback, you decide in advance at what level you will act, and the level is numeric before price gets there. That removes from the decision the part which is taken badly, the part taken while price is moving.
The side effect matters as much as the main one: a level fixed in advance makes risk calculable, so the trade becomes comparable with the others. An entry decided in real time produces no usable statistic.
Level 1 is the move's origin: if price breaks it, there is no pullback left to buy and the scenario is invalidated. The stop therefore goes beyond that point, however deep the entry. The grid thus answers the question of where to place a stop loss with a price tied to the scenario, not with a number of pips picked out of habit.
The distance between the entry and that stop then gives the size: that is the arithmetic a position size calculator does, and it is what lets you risk the same amount at 0.382 and at 0.786.
An extension projects the move's range past its extreme. Measured from the origin, the 1.272 extension sits 27.2% of the range beyond the extreme, and the 1.618 extension 61.8% beyond it. On a retracement grid whose 0 is at the extreme, those are the −0.272 and −0.618 levels.
Two precautions. The three-point extension tool that platforms offer projects the range from the end of the pullback, not from the origin: it gives different prices for the same ratios, so two targets only compare when drawn with the same tool. And a target is judged by what it pays relative to the risk: that is the trade's risk reward ratio, known before entry.
The grid displays prices to the pip, and that precision is misleading. Between 0.236 and 0.786, two neighbouring levels are never more than 16.8% of the range apart: no price in that span is therefore more than 8.4% of the range away from a level. On a 100-pip move, a reversal occurring anywhere between 1.0921 and 1.0976 sits less than 9 pips from a line. Seen after the fact, almost any pullback appears to have 'respected Fibonacci'.
The second trap is the talk about natural proportions. That the golden ratio shows up in a seashell says nothing about markets. If the levels trigger reactions, it is because many traders display the same grid and place their orders on it: a self-fulfilling prophecy, whose strength depends on how many people follow it. A statistical test run in 2006 by Roy Batchelor and Richard Ramyar of Cass Business School looked for these ratios in successive Dow Jones trends, on daily prices from 1915 to 2003: it concludes, in 'Magic numbers in the Dow', that they appear no more often than chance would have it.
Other published work does not overturn that finding. On S&P 500 closes from April 1998 to March 2003, Sukanto Bhattacharya and Kuldeep Kumar report a favourable result in the Annals of Economics and Finance, which they themselves describe as merely indicative and not conclusive. On three equity markets, a study published in Expert Systems with Applications in 2022 by Tsinaslanidis, Guijarro and Voukelatos finds that the wider the zone drawn around a level, the more likely a bounce inside it, and adds that this does not necessarily make a profitable strategy. None of these three studies establishes a success rate for any level.
The third has to do with real time. On a past chart the extreme is known and the grid fits. Live, as long as the move keeps printing new highs, the extreme moves and every level moves with it: the grid you are watching during a pullback is only final if the high is, and nobody knows that at the time.
The classic mistake. Picking the move after deciding where you wanted to enter. It is the most frequent and most invisible mistake: the grid is arithmetic, so it looks objective, while its two points have just been chosen to produce the convenient level.
| Indicator | Formula | What it measures | When to use it |
|---|---|---|---|
| Fibonacci retracement | Levels between a move's origin and extreme | How deep a pullback is, as a share of the move | Deciding an entry level in advance |
| Fibonacci extension | The same two points, ratios above 1 | A projection past the extreme | Setting a target before entering |
| OTE zone | The 0.62-0.79 band of that same grid | A deep pullback, named in ICT vocabulary | Looking for the shortest-risk entry |
| Fair value gap | 3 candles, the 1st and the 3rd do not overlap | A price range that barely traded | Locating a return on an absolute level |
| Moving average | Average of the last N prices | A smoothed trend, which moves | Following a direction, not fixing a level |
It is the tool's only subjective decision, so the only place to make it reliable. One sentence is enough: 'the last move that breaks structure', or 'the session's last high and last low'. Written down, it stops you picking the convenient high afterwards.
Overlaying retracements of several moves guarantees a price will always land on a level, and empties the tool of any selective power. Two Fibonacci levels at the same price are not two reasons, they are the same reason counted twice. Keep one grid, the one from the timeframe that commands your plan, and ask something else for the second reason: structure, an area price has already worked, a round number.
This is the trade-off nobody settles in general, and it settles very well in particular. Two tags, thirty or so trades each, and the average R of each group says which of the two levels pays you.
It is a grid of levels drawn between a move's origin and its extreme, to put numbers on how far price comes back before moving on. A 0.618 level means price has given back 61.8% of what it took.
It is the sequence of numbers in which each term is the sum of the two before it: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34 and so on. It is named after Leonardo of Pisa, known as Fibonacci, who published it in 1202, although Indian mathematicians had described it several centuries earlier. The ratio between two consecutive terms tends towards 1.618, the golden ratio, and that ratio is what trading borrows from it.
0.382, 0.5 and 0.618 for pullbacks, with 0.236 added for very shallow ones and 0.786 for deep ones. For targets, the 1.272 and 1.618 extensions. In MetaTrader, 0.786 is not in the default list: you add it in the tool's properties.
Level = extreme − (extreme − origin) × ratio. On a move from 1.0900 to 1.1000, the 0.618 level is 1.1000 − 0.0100 × 0.618 = 1.0938. A Fibonacci calculator does the same arithmetic, and a platform's drawing tool applies it as soon as the two points are set.
No. The drawing is included free in every charting platform, and the calculation can be redone by hand in one line. A paid Fibonacci tool cannot add anything to a subtraction and a multiplication.
They do produce reactions, but not for the reason usually given. Nothing shows that markets obey these ratios by nature: looked for in the Dow Jones from 1915 to 2003, they turn up no more often than chance would have it. The levels matter because many traders display the same grid and concentrate their orders on it. The only answer that counts for you is the one from your own trades, measured level by level.
The trade-off between a shallow level and a deep one is a question of R, and that is exactly what Tradoshi computes. A trade's R is its result divided by the risk taken, so by the distance between your entry and your stop: entering at 0.79 rather than 0.38 shortens that distance and changes the same trade's R, without the market having done anything different.
The Reports screen shows average R alongside win rate and expectancy. By tagging your entries by level, you get both halves of the trade-off on your own data: the deep level should show a higher average R and a lower win rate, and it is the product of the two, not either one, that designates the better choice.
To go deeper on this: read the full guide. Every term is gathered in the glossary.