Home › Learn › Lesson 21

Polarity and Fibonacci — old resistance as new support, and why Fibonacci is never used alone

By TradingPrimer · Published 1 Sep 2026

Fibonacci is the most-drawn and least-examined tool on a crypto chart. Almost every guide says the same three things: the market retraces to 0.382, 0.5, 0.618 or 0.786; a shallow retracement means a strong trend; and you should never use Fibonacci on its own. All three are repeated as folklore, and the third is repeated most often by people who cannot say what the tool is missing. This lesson does the arithmetic on all three. It turns out the grid quietly prices every level it draws, that the level with the strongest evidence is the worst-priced entry on the chart, and that the reason to never use Fibonacci alone is measurable to the point of being embarrassing: on the worked example below, no possible pullback price could land more than one average candle away from some Fibonacci line.

What does polarity actually claim, and why would a level flip?

That a price which was hard to get through in one direction becomes hard to get back through in the other — and the reason is people, not geometry.

You already met the mechanism in support and resistance. A level exists because a lot of decisions happened there. What polarity adds is what happens to those decisions after the level breaks. Three groups are sitting at a resistance that has just given way. Sellers who were short into it are now wrong and want out near their entry. Buyers who bought the breakout have a natural place to add. And the largest group — people who watched it hold three times and did nothing — now has an obvious price at which to finally act. All three want to transact at roughly the same number, and all three want to do it on the same side. That is what an old ceiling turning into a floor looks like from underneath.

The classical course material states the principle in one line for each direction: old resistance becomes new support, and old support becomes new resistance. It then attaches two conditions that are far more important than the slogan.

The market has to be trending. Polarity is a claim about a level that was overcome. Inside a range, the same level is being traded from both sides on purpose — nothing has been overcome, and there is no flip to trade. This is why diagnosing trend or range before anything else is not preparation, it is the thing that decides whether the setup exists.

The first retest usually works. Not the fourth. This is the same erosion you priced in lesson 11, seen from the other side: every additional visit spends more of the resting interest that made the level a level. And as the next section shows, repeated testing does not only weaken the level in some vague sense — it also makes the trade measurably more expensive, because the zone gets wider and the zone is your stop.

Why does the polarity trade fail so often, and what does a safe stop cost?

Because the two most natural things to do — buy at the level, and put the stop just under it — are both mistakes, and they are the same mistake twice.

The course material names four reasons polarity trades fail. They are worth listing exactly, because each has a number attached to it later on this page:

• entering in a rush, with limit orders set in advance at the level;
• not patiently waiting for a reversal structure on a lower timeframe before entering;
• not first deciding whether the market is trending or not;
• placing the stop close to the sensitive level.

Take the last one and price it. A resistance zone ran from $63,800 to $64,200 — a zone rather than a line, because three separate reactions turned at slightly different prices, which is how real levels look. Price broke it and ran to $70,000. Now it comes back. You wait, a reversal appears on a lower frame, and you enter at $64,300, just above the zone. Your target is the prior high at $70,000, so the reward is $5,700 whatever you do next. Only the stop is a decision:

The average candle range on the frame is assumed at $900 throughout this lesson — the same stated modelling assumption used in lesson 20, not a measurement of any particular day. Substitute your own ATR reading and every row moves.

Now the honest way to read that table. The tight stop is not wrong; it is a bet, and you can see the exact size of the bet. Clearing the zone costs 7.0 percentage points of break-even win rate. So the wider stop pays for itself if more than roughly one trade in fourteen of your tight-stop retests would have been wicked out and then gone on to the target. That is a question your own journal can answer and nobody else’s can, and it is a far better question than “how tight should my stop be”.

There is a second, quieter cost hiding in the same table. The zone’s width is your stop. Reward-to-risk on any retest entry is distance to target ÷ (zone width + buffer). So the fourth touch of a level does not merely weaken it — it widens it. Push the zone floor down from $63,800 to $63,600 as one more sloppy reaction gets added to it, keep the same $400 buffer, and the stop widens from $900 to $1,100 — the same trade drops from 6.33R to 5.18R without a single thing changing about your analysis. Levels are priced by their tightness, not by their popularity.

What is a Fibonacci retracement actually measuring?

One number: what fraction of a single impulse has been given back. Everything else about the tool is presentation.

You anchor it at the start and end of a move — low to high for an up-impulse, high to low for a down one — and it divides that distance into fixed fractions. The classical set is 0.236, 0.382, 0.5, 0.618 and 0.786. The course material gives the tool exactly two jobs, and it is worth noticing how modest they are:

• Finding the likely support of an up-impulse, or resistance of a down-impulse — that is, where a pullback might stop.
• Finding a target for the next impulse once the correction has finished.

Neither job is “tell you when to buy”. The tool is a coordinate system for one wave. It has nothing to say about whether that wave belongs to your timeframe, whether the pullback is finished, or whether anyone else is looking at your anchors. What it does have — and what nobody prints — is a price list.

What does each Fibonacci level actually pay?

Exactly f ÷ (1 − f), and the break-even win rate is exactly 1 − f. The grid quotes its own odds.

Set up the plainest possible retracement trade. An impulse runs from $60,000 to $70,000 — call the size A = $10,000. Price pulls back to level f, so your entry is H − fA. Your stop goes below the origin of the impulse at $60,000, because that is what invalidates the whole idea. Your target is the prior high at $70,000, which is the level that stopped price last time. Then:

• Risk = entry − $60,000 = (1 − f)A
• Reward = $70,000 − entry = fA
• R:R = f ÷ (1 − f)  ·  break-even win rate = (1 − f) ÷ 1 = 1 − f

A cancels. The size of the impulse, the coin, the account — none of it appears. What you get paid at a Fibonacci level is a property of the level itself:

Three things fall out of that table, and none of them appear in the usual Fibonacci guide.

The break-even column is the level, subtracted from one. Not approximately — exactly. Buy the 0.382 and you must win 61.8% of the time to stand still. Buy the 0.618 and you must win 38.2%. You do not have to compute anything at the chart: whatever fraction the tool is telling you has been retraced, one minus that number is the hit rate the trade requires.

The two golden levels are each other’s break-even. The 0.618 entry pays 1.618R and breaks even at 38.2%. The 0.382 entry pays 0.618R and breaks even at 61.8%. This is a property of the ratio itself — φ is the number whose reciprocal is itself minus one — not a fact about markets. But it makes the two rows you actually use free to remember, and it is a small piece of arithmetic poetry that sits alongside the one from lesson 20, where a flag retracing 0.618 broke even at exactly 38.2% too.

The 0.382 entry is a sub-1R trade. That is the finding that should stop you. The shallow retracement is the one every guide teaches first, the one that looks strongest on the chart, the one that means the trend is healthy. Priced against the prior high with a structural stop, it risks $6,180 to make $3,820. Under 1R. It loses money at anything short of a 61.8% hit rate, which is a rate very few strategies sustain.

Why is the level with the best price the one with the worst evidence?

Because depth means two opposite things at once, and the grid only prices one of them.

The classical reading of retracement depth is the mirror image of the table above. A pullback that stops shallow — around 0.382, in the upper half of the impulse — says the force behind the impulse was preserved: the buyers only paused. A pullback that runs deep — 0.618, 0.786, 0.8 — says that force was nearly cancelled, and the case for a continuation has largely gone with it. A commonly taught line is that a genuine uptrend’s corrections do not give back more than half the previous wave.

So here is the whole problem in one sentence: evidence improves as you move up the grid, and price improves as you move down it. They are in direct opposition, and no arrangement of the tool resolves that. Anyone telling you that a particular level is the level to buy has simply picked a side of this trade-off without saying so.

The right-hand column is the most usable thing on this page, so it is worth spelling out. Targeting the prior high is conservative; most people want a new high. If your target is e beyond the old high as a fraction of the impulse, then R:R = (e + f) ÷ (1 − f). Set that to 2.00 and solve for e. Buying the 0.382 requires the market to make a new high 85.4% of the impulse above the old one just to be an ordinary 2R trade — on our numbers, $70,000 has to become $78,540. Buying the 0.618 requires 14.6%, which is $71,460. You are asking the market for 5.85 times more to justify the shallower entry.

And you are asking for it in the situation where you were also told the move had further to run, which is exactly how these two errors hide each other. The shallow entry feels safe because the evidence is good, and the extension needed to make it pay never gets calculated because nobody publishes this column.

Does price really “respect” Fibonacci levels?

On a chart with the standard grid on it, the question cannot be answered, because the grid is too dense to miss.

Measure the gaps between adjacent lines: 0.236 to 0.382 is 0.146, then 0.118, then 0.118, then 0.618 to 0.786 is 0.168. The widest is 0.168, so the furthest any price inside the grid can sit from a line is half of that: 0.084 of the impulse.

On our $10,000 impulse, 0.084 is 840 points. The average candle on the frame is $900. So every possible pullback price between 0.236 and 0.786 lands within 0.93 of one average candle of some Fibonacci line. There is no price the market could have chosen that would have looked like a miss.

That is what “never use Fibonacci on its own” actually means, and it is stronger than the way it is usually said. The problem is not that Fibonacci is unreliable. The problem is that a claim which cannot fail carries no information. When someone shows you a chart where price turned at the 0.618, they have shown you that price turned somewhere — the line was always going to be there.

What can carry information is a level that existed before you drew the tool. A prior swing high or low. A zone that produced three reactions. A broken resistance now trading as support. Those are decisions the market made without your input, so when one of them coincides with a Fibonacci line, the coincidence is at least not manufactured. The Fibonacci line is doing the same work here that a moving average does when price bounces off it in a trend: it is a convenient description of where structure already was. That is the honest version of the word confluence, and everything else sold under that name is a way of drawing more lines until one of them is near the price.

How much does the anchor you clicked change the trade?

By f × your anchoring error — which matters exactly in proportion to how tight your stop is, and not at all if you are internally consistent.

Two people can look at the same impulse and disagree about where it started. One anchors at the absolute wick low; another at the candle body; another at a small higher low a bit further along. Say you anchor d too high, as a fraction of the impulse. The new impulse is A(1 − d), and the line at level f moves up by f · d · A.

Read the third column again. With a stop tucked $200 under the line, a 5% disagreement about where the swing began moves the entry by more than the entire stop distance. Two traders with the same idea and the same rules get opposite outcomes because one of them clicked on a wick and the other on a close. Against a stop placed under structure, the same error is a third of the risk and rarely decisive.

There is a genuinely useful subtlety here, and it is the same lesson lesson 17 found about mixing timeframes. If you take the entry and the stop from the same grid — entry at f, stop at the anchored low — the anchoring error changes nothing at all, because f ÷ (1 − f) has no A in it. Move both anchors and the reward-to-risk is untouched. Anchoring only starts costing money when the entry comes from Fibonacci and the stop comes from somewhere else. The tool is not fragile; mixing two coordinate systems is.

How do polarity and Fibonacci combine into one entry?

The polarity level decides whether there is a trade. Fibonacci decides what it is worth. Neither does the other’s job.

Put the whole lesson on one chart. The old resistance zone at $63,800–$64,200 was broken during the impulse that ran $60,000 to $70,000. Price is now pulling back into it. And the 0.618 retracement of that impulse computes to $63,820 — inside the zone. That is real confluence: a level the market built earlier, and a ruler that happens to agree with it.

Two people take this trade. Here is what separates them, and it is not the analysis:

The paper number is 4.88 times the honest one, and the entire difference is a stop measuring 0.22 of one average candle. An entirely unremarkable candle covers that stop four and a half times over. The 30.90R trade is not really a 30.90R idea; it is a bet that the next candle happens to be small, dressed up as an edge. This is the exact arithmetic behind the classical warning to keep the stop clear of the sensitive levels — the levels are sensitive because that is where the fighting happens, which is the worst possible place to stand.

Note also which trade the two people are in. The patient entry is only taken if the reversal appears. On the occasions when price slices through $63,800 and keeps going, the patient trader simply has no trade, while the limit order was filled automatically on the way past. That is the second half of why pre-set limits underperform: they are guaranteed to fill in exactly the case where you were wrong.

So the sequence, in order, and the order matters:

1. Is the market trending? No trend, no polarity, no trade — regardless of how good the level looks.
2. Is this the first retest of the broken level? The first usually works; by the fourth you are trading a used-up level with a wider zone and a worse price.
3. Does a Fibonacci line of the impulse land on the zone? If yes, note it. If no, you still have a polarity trade — the fib was never the reason.
4. Compute the price before deciding. Distance to the prior high, divided by zone width plus buffer. Break-even is 1 ÷ (R + 1).
5. Wait for a reversal on a lower frame before entering, and put the stop clear of the zone, not inside it.

When is everything on this page wrong?

Four conditions, and the first two are common enough that you will meet them this week.

When the market is ranging. Every formula above assumes a directional impulse with a start and an end. In a range there is no impulse to measure, so the Fibonacci anchors are arbitrary and the polarity flip never happened. Drawing the tool across a swing inside a box produces perfectly convincing lines that describe nothing. Diagnose the state first.

When the impulse belongs to a different timeframe than the one you are trading. The retracement of a daily impulse is a daily-sized pullback, which on a fifteen-minute chart is an entire trend in the opposite direction. The grid does not know which frame you are on; you have to. This is the frame-mixing cost in a new costume.

When the target is not the prior high. The clean identity break-even = 1 − f holds for the specific trade defined above: stop at the impulse origin, target at the prior high. Change either leg and you have to redo the division. That is a two-second calculation, and the point of having a formula is that redoing it is cheap.

When the level is thin. Everything here assumes a level with enough resting interest to actually stop price. On a low-liquidity pair, or into a scheduled event, price can travel through a beautifully drawn zone without registering it. Confluence between two lines on an illiquid chart is confluence between two opinions.

Common mistakes

Frequently asked questions

Which Fibonacci level is the best one to buy?

There is no best level, because the level decides the price and the price decides what you need to be right about — and those move in opposite directions. With the stop at the swing low and the target at the prior high, reward-to-risk at level f is f ÷ (1 − f) and break-even win rate is exactly 1 − f. The 0.382 entry pays 0.618R and needs 61.8%; the 0.618 entry pays 1.618R and needs 38.2%. Deeper is always better priced. The classical reading of the same chart says the reverse: a shallow pullback preserves the impulse, a deep one has mostly cancelled it. That opposition is the lesson. The grid does not tell you where to buy — it tells you the exchange rate between price and evidence at every depth, and you have to choose knowingly which one you are paying for.

Why do people say never use Fibonacci on its own?

Because the grid is too dense to be capable of failing. The widest gap between adjacent standard levels is 0.168 of the impulse, so any pullback inside the grid lands within 0.084 of some line. On the $10,000 impulse used throughout this lesson that is 840 points, against an assumed average candle of $900 — so no possible pullback price could sit more than 0.93 of one candle away from a Fibonacci level. “Price respected the 0.618” is therefore arithmetic, not evidence. What can be evidence is a level that existed before you drew the tool: a prior swing, a zone that produced repeated reactions, an old resistance now acting as support. When one of those coincides with a Fibonacci line, the coincidence was not manufactured by you. That is the only version of “confluence” worth the word.

Does it matter exactly where I anchor the Fibonacci tool?

It matters in exact proportion to how tight your stop is. Anchor the low d too high as a fraction of the impulse and the line at level f moves by f × d of the impulse: a 5% anchoring error shifts the 0.618 line by 3.09%, which is 309 points on a 10,000-point swing. Against a 200-point stop tucked under the line that is 1.54 times the entire stop, so the click decides the outcome. Against a 900-point structural stop it is 0.34 of the risk and rarely decisive. There is one subtlety that surprises people: if entry and stop both come from the same grid, an anchoring error does not change reward-to-risk at all, because f ÷ (1 − f) contains no measure of size. Anchoring only starts costing money when the entry comes from Fibonacci and the stop comes from somewhere else — the same failure mode as mixing timeframes.

Should I place a limit order at the Fibonacci level in advance?

The order is not the problem; the stop it forces on you is. A pre-set limit at $63,820 with a stop $200 under prices at 30.90R — but that stop is 0.22 of one average candle, so an ordinary candle covers it four and a half times over. The same idea entered at $64,300 after a reversal appears on a lower frame, with the stop clear of the zone at $63,400, prices at 6.33R on a stop of exactly one average candle. The paper number is 4.88 times bigger and the real one is the one you can hold. There is a second cost too: the limit order fills automatically on every occasion price slices straight through the level, which is precisely the case where you were wrong and would rather have had no trade at all. The classical course material lists both halves as named reasons polarity trades fail.

Educational content only — not financial advice, and not a trade recommendation. Every figure on this page comes from worked models built for this lesson and each can be reproduced from the numbers given: an impulse from $60,000 to $70,000, an old resistance zone of $63,800 to $64,200, entries at $63,820 and $64,300, and stops at $63,620, $63,900 and $63,400. Two identities do all the work. For a retracement entry at level f with the stop at the impulse origin and the target at the prior high, risk = (1 − f)A and reward = fA, so R:R = f ÷ (1 − f) and break-even win rate = 1 − f. Extending the target by e beyond the old high gives R:R = (e + f) ÷ (1 − f), which set to 2.00 yields e = 2 − 3f. Break-even win rate is 1 ÷ (R + 1) throughout. The $900 average candle range is a stated modelling assumption carried over from lesson 20, not a measurement of any particular day or pair, and readers should substitute their own ATR reading. No historical hit rate for polarity retests or Fibonacci levels is quoted anywhere, because published figures for both vary too much with test design to repeat responsibly. The principles — old resistance becoming new support and the reverse, the requirement that the market be trending, the first retest usually working, the stop needing to sit clear of sensitive levels, the four named failure causes, the tool’s two stated jobs, and the reading of shallow versus deep retracement — follow the classical rules taught in our own slide course. Sources: our own arithmetic, stated inline. Published 1 Sep 2026.