Home › Learn › Lesson 15

RSI — it measures a ratio, and the ratio is not linear

By TradingPrimer · Published 30 Aug 2026

RSI is the first indicator almost everyone adds to a chart, and the first one almost everyone loses money on — usually by selling something because a number went above 70. The reason is not that the indicator is unreliable. It is that the 0–100 scale hides what is being counted, and once you can convert a reading back into the quantity it stands for, several of the standard pieces of RSI advice stop making sense. This lesson does that conversion, then uses it to answer three questions with exact arithmetic: how much force a reading represents, why an extreme reading does not cool off on its own, and how much selling it actually takes to bring one down.

What is RSI actually measuring?

The ratio between the average size of recent up moves and the average size of recent down moves. That is the whole quantity. Everything else — the 0–100 scale, the shaded bands, the word “overbought” — is packaging wrapped around that one ratio.

Here is the recipe, in the form J. Welles Wilder published it in 1978, with the default lookback of 14 bars. For each bar, compare the close with the previous close and record the move in one of two buckets:

• If the bar closed higher, the difference goes in the gain bucket and a zero goes in the loss bucket.
• If the bar closed lower, the difference goes in the loss bucket and a zero goes in the gain bucket.

Each bucket is then smoothed, so the running average carries thirteen parts of yesterday and one part of today: average_today = (average_yesterday × 13 + today) ÷ 14. Divide one average by the other to get RS, the relative strength, and squash it onto a 0–100 scale:

RS = average gain ÷ average loss     RSI = 100 − 100 ÷ (1 + RS)

Two things are worth noticing straight away, because almost every misuse of RSI traces back to missing one of them.

It reads closes only. Wicks are ignored entirely. A bar that spiked violently in both directions and closed unchanged contributes exactly nothing to either bucket — a fact we will use in a moment, because it has a surprising consequence.

It looks backwards, at 14 bars, and no further. RSI contains no information that was not already on your chart. Like the moving averages in Lesson 14, it is a filter: prices go in, a smoother number comes out. It cannot know anything about the next bar, and any sentence beginning “RSI is telling us price will…” has smuggled in a forecast that the arithmetic does not contain.

Does RSI 80 mean 80% of the market is buying?

No, and it cannot — not in any market, at any price, ever. You will often read that RSI 80 means “80% of the market is buying”. It is a vivid image and it is impossible, because every executed trade has a buyer and a seller in equal measure. If someone bought a bitcoin, someone sold one. The market is never 80% buyers in any literal sense — it is always exactly 50/50 by construction, in every market, at every price, forever.

What actually differs is urgency, and urgency shows up as movement. When buyers are the impatient side, they lift offers and price travels upward to find willing sellers; the up bars get larger and the down bars get smaller. RSI measures precisely that asymmetry in bar sizes. So the honest translation of RSI 80 is not “four buyers for every seller” but:

“Over the last 14 bars, the average up move has been four times the size of the average down move.”

That sentence is longer and much less exciting, and it is the one that will keep you out of trouble. It is a description of price behaviour that has already finished happening. It contains no claim about what anyone intends to do next.

What does an RSI number convert to in plain terms?

Exactly and reversibly, by rearranging the formula. If RSI = 100 − 100 ÷ (1 + RS), then:

RS = RSI ÷ (100 − RSI)

Every RSI value you have ever looked at converts to a specific ratio in one division, and the results are not what the coloured bands on the chart imply.

Try the conversion on the number that is on your chart right now. If it says 64, the market has been producing up moves about 1.8 times the size of its down moves for fourteen bars. That is a real, specific, interesting fact. It is also plainly not the same kind of statement as “the market is nearly overbought”, which is the sentence most people reach for instead.

Why is a move from 80 to 90 bigger than a move from 50 to 60?

Because the 0–100 scale is not a ruler. It is a compressed rendering of a ratio that runs from zero to infinity, and squeezing an unbounded quantity into a hundred boxes means the boxes near the ends have to hold much more than the boxes in the middle.

Ten points of dial buys you a very different amount of force depending on where you spend it:

This has a practical consequence that gets overlooked constantly. When RSI climbs from 72 to 78, most commentary treats it as “a bit more overbought” — six points, barely worth mentioning. In ratio terms it is a move from 2.57 : 1 to 3.55 : 1, an increase of nearly 40% in the imbalance. The dial understated it because the dial is compressed up there.

The same compression runs the other way, and it explains something that puzzles beginners: why RSI seems to move so slowly in the middle of the range and then race at the extremes. It is not the indicator being erratic. Near 50, a large change in the underlying ratio barely nudges the number; near 90, a small change in the ratio throws it several points.

Why doesn’t an overbought reading cool off on its own?

Because time does not enter the formula. Only closes do — and a bar that closes unchanged contributes nothing to either bucket.

Watch what happens on a genuinely flat bar. The gain bucket gets a zero, so the average gain becomes (average gain × 13 + 0) ÷ 14, which is the old average multiplied by 13 ÷ 14. The loss bucket gets a zero too, so the average loss is also multiplied by 13 ÷ 14. Both halves of the fraction shrink by exactly the same factor, and a fraction whose top and bottom are both multiplied by 13 ÷ 14 is the same fraction it was before.

RSI 80 stays at 80.00 through one flat bar, eight flat bars, or eighty. The two averages get smaller and smaller, but their ratio never budges. That is a strange and slightly beautiful property, and it demolishes one of the most common intuitions about the indicator: that an extreme reading somehow “works itself off” if the market just pauses. It does not. A pause changes nothing at all.

Perfectly flat closes are of course an idealisation; real ranges chop up and down rather than printing identical closes. Choppy bars do drag RSI back toward 50, because alternating small gains and losses feed both buckets and the smaller one grows proportionally faster. But the drift is slow. Running the same model with alternating small up and down bars, an RSI of 80 is still above 70 after ten bars of chop and only reaches the low sixties after twenty.

So the honest version of “wait for it to cool off” is: the reading will not come down until price actually comes down. Which raises the obvious next question.

How much selling does it take to bring RSI back down?

More than most people picture. We can put a number on it, because the decay is fully determined once you fix the size of the down bars.

Take a market that has run RSI up to 80 — a ratio of 4.00 : 1 — and then start feeding it down bars, each one as large as the average up bar that built the reading in the first place. That is a decisive turn, not a wobble: every bar is a full-sized move against the trend.

Read the second row twice. A full-sized red bar arriving straight after the peak moves RSI from 80.00 to 75.36. The chart still shows a reading inside the shaded band. Whatever you think about the market at that point, the indicator has not confirmed anything: it takes three such bars merely to print a number below 70, and seven to get back to balance.

Halve the size of the down bars — a more typical pullback — and the numbers stretch out further: five bars to break below 70 and twelve to reach 50. Both models say the same thing in different accents. An RSI that has reached an extreme has bought itself a substantial amount of time.

This is the arithmetic behind an old piece of trading advice that usually gets passed on as folklore: markets can stay overbought far longer than you can stay solvent shorting them. It is not folklore. It is a property of a smoothed 14-bar ratio, and you can compute exactly how much longer.

Are 70 and 30 the right lines to watch?

They are the conventional ones, and there is a decent case that 60 and 40 are the more informative pair. Not because the tradition is wrong, but because the two pairs of lines answer different questions.

Start from the ratios rather than the habit. Between RSI 45 and RSI 55 the pressure ratio runs from 0.82 : 1 to 1.22 : 1 — everything in that band is within about 25% of dead level, which on a noisy chart is indistinguishable from balance. RSI 60 is the first reading where up moves are a full 50% larger than down moves, and RSI 40 is its mirror. So there is a real, computable frontier there: below 60, a rising RSI is compatible with sellers simply easing off; above 60, something has to be actively lifting price.

That distinction is more useful day to day than the overbought line, because it separates two situations that look identical on a price chart:

A reading stuck between 40 and 60 for a long stretch is itself informative: it says neither side has been able to make its bars bigger than the other’s for fourteen bars running, which is a reasonable working definition of a range. You can often tell a market is going sideways from the RSI panel alone, without looking at price.

Where this framing is wrong. It is a reading, not a law, and it fails in two identifiable places. First, in a genuinely range-bound market the older 70/30 mean-reversion logic works better than the 60/40 momentum logic — inside a range, extremes really do tend to revert, which is close to what “range” means. Second, in a sustained downtrend RSI can sit under 40 for weeks; treating every dip below 40 as “sellers arrived, therefore a bounce is due” is the mirror image of the mistake this whole lesson is about. The lines tell you about the ratio. They never tell you what happens next, and no threshold ever will.

Should you trade RSI divergence?

Divergence — price making a higher high while RSI makes a lower high — is the most popular RSI technique and the one that deserves the most caution. It is worth being precise about what it does and does not say.

What it says is true and modest: the second push was made with a smaller size advantage than the first. Price went further, but it did so with up bars that were less dominant relative to the down bars. That is a genuine observation about the ratio weakening.

What it does not say is that the trend is ending. A trend can weaken and then re-accelerate; the ratio can shrink from 4 : 1 to 3 : 1 and stay at 3 : 1 for months, which prints a textbook divergence at every step while price keeps climbing. Divergence has no built-in mechanism that makes the move stop.

It also has a timing problem that no amount of skill removes. A divergence only exists once the second high is in place, which means you can only identify it after the fact — and the second high is not confirmed as a high until price has already fallen away from it. If you act before that, you are not reading a divergence, you are predicting one.

The reasonable position is to treat divergence as context rather than as a trigger: a reason to tighten risk on a position you already hold, or to demote a setup you were about to take, rather than a reason to trade against a trend that is still intact. If you want to test that for yourself, the honest experiment is to mark every divergence over a year on one chart — including the ones that failed, which are the ones nobody remembers — and count.

How do you use RSI without pretending it predicts?

Give it one job: describing the balance of the last 14 bars in a number you can compare across time. Then read it in a fixed order, so the number does not get to jump straight to a conclusion.

Step 3 deserves a note, because it is the one that catches errors. RSI is computed from price, so an RSI trend that has no matching price move is a contradiction: if up moves really were dominating for fourteen bars, the chart would show an advance. When the panel looks decisive and the chart looks flat, the usual explanation is not that RSI has seen something hidden — it is that one side simply stopped trading, which shrinks one bucket without filling the other. Check the price chart before you believe the panel.

Two habits make the whole thing more reliable, and neither involves a setting. Read the reading only on closed bars: an RSI value on a bar still in progress is a number about a bar that has not happened yet, and it can move several points before the close. And fix your period and leave it alone. Shortening the lookback from 14 to 7 does not make RSI faster in any useful sense; it makes it noisier, which feels faster because it produces more readings, most of them meaningless.

When is RSI simply the wrong tool?

Three situations, and knowing them is worth more than any threshold.

When you want to know where price will stop. RSI has no price axis. It cannot tell you a level, only a balance, so it can never answer “where do I put my stop” or “where does this run out”. Those are questions for the zones in Lesson 11, where trading demonstrably happened at a price, and for the resting size in the book that Lesson 8 measures.

On the first bars after a listing or a gap. The 14-bar average needs 14 bars. A reading built from four bars of a newly listed token is arithmetic performed on almost nothing, and it will swing to both extremes for no reason at all.

When you are stacking it with two more oscillators. RSI, stochastics and MACD are all computed from the same closes, so agreement between them is close to guaranteed and tells you nothing new. Three views of one number is one piece of evidence wearing three hats. If you want a second opinion, it has to come from a different measurement — participation, as in Lesson 13, or a different timeframe, as in Lesson 10.

There is also a crypto-specific wrinkle worth knowing. Because crypto trades 24/7 with no session close, what counts as a “daily bar” depends on each venue’s UTC cut-off. Two exchanges can display slightly different daily RSI values for the same asset on the same day, and neither is broken. If a rule of yours turns on RSI crossing a precise number, it will fire on one venue and not the other.

What are the most common mistakes with RSI?

Six of those eight are mistakes about the instrument rather than about the market, which is the same pattern the volume lesson and the moving average lesson both ran into. Indicators very rarely fail. They report exactly what they were built to report, to people who never checked what that was.

What else do people ask about RSI?

What is a good RSI setting for crypto?

The default 14 is a reasonable choice and the case for changing it is weaker than it looks. A shorter lookback does not detect turns earlier in any reliable sense; it reacts to a smaller sample, so it crosses your thresholds more often, and most of those extra crossings are noise. There is also a practical argument for the default: because so many people use 14, readings at 14 are at least comparable between traders and across articles. If you do change it, change it once, write down why, and leave it for months — adjusting the period after seeing the chart guarantees you will find the setting that would have worked on the past.

Does RSI above 70 mean I should sell?

No. RSI 70 means up moves have averaged 2.33 times the size of down moves over the last 14 bars, which is a description of an advance rather than a warning about one. Readings can stay above 70 for long stretches, and the arithmetic explains why: starting from 80, it takes three consecutive down bars the size of the average up bar merely to print a value below 70, and seven to return to balance. A high reading tells you the trend has been strong. It contains no information about when it will stop.

Why is RSI still at 80 when price has stopped going up?

Because a bar that closes unchanged adds nothing to either the gain bucket or the loss bucket, so Wilder smoothing multiplies both averages by 13 ÷ 14 and leaves their ratio — and therefore RSI — exactly where it was. Time does not appear in the formula anywhere. In real markets a choppy range does drag the reading toward 50, because alternating bars feed both buckets and the smaller one grows proportionally faster, but the drift is slower than most people expect: in our model an RSI of 80 is still above 70 after ten bars of chop. Only lower closes bring the number down meaningfully.

Is RSI a leading or a lagging indicator?

Lagging, in the only sense that matters: every input is a close that has already printed. RSI is sometimes called leading because it can turn before price does, but that is a statement about the shape of a smoothed ratio, not about foresight — the ratio can shrink while price still rises, and it can shrink for a long time without the trend ending. The useful framing is that RSI is a measuring instrument for something that has already happened, in the same family as a moving average. Use it to describe the balance you are trading inside, and take your entries and exits from prices on the chart.

Where does this sit in the course?

Lesson 15 follows Lesson 14 on moving averages, which measured the cost of smoothing price; this one measures the balance between up moves and down moves and puts an exact ratio on every reading. It leans on the closes defined in Lesson 9, since RSI reads nothing else, and on the timeframe choice from Lesson 10, because “14 bars” means fourteen hours or fourteen weeks depending on which chart you opened. Next comes Lesson 16 on chart patterns.

Educational content only — not financial advice, and not a trade recommendation. RSI cannot guarantee direction, execution quality or profit. Every figure on this page comes from Wilder’s published formula and from worked models built for this lesson — the conversion RS = RSI ÷ (100 − RSI), a smoothed state of average gain 4 against average loss 1, and down bars of a stated size — and each can be reproduced in a spreadsheet; no market data is quoted. Sources: our own arithmetic, stated inline. Published 30 Aug 2026.