RSI: What It Says and What It Doesn't

RSI is the most widely misread number in retail trading, and the misreading is remarkably consistent. Someone sees 72 on the indicator, concludes the asset is overbought, and treats that as a statement about value. It is not a statement about value. It is not really a statement about the future either.

What it is, precisely, is a measure of how one-sided recent price movement has been. That is a genuinely useful thing to know. It is also a much smaller thing than most people assume they are being told.

This guide is about what the calculation does, and where the standard interpretation breaks. It does not tell you what to do with the number, because nobody honestly can.

What is RSI actually measuring?

Take the last fourteen periods. For each one, note whether price closed up or down, and by how much. Average all the up moves. Average all the down moves. Divide the first by the second, and you get the relative strength: how much bigger the typical gain has been than the typical loss.

That ratio is then squeezed onto a 0-to-100 scale. A reading of 50 means up moves and down moves have been about equal in average size. Above 50 means gains have dominated. Below 50 means losses have. RSI of 70 corresponds to average gains running roughly 2.3 times average losses. RSI of 30 is the mirror image.

Two things follow immediately, and both are routinely forgotten.

First, RSI knows nothing about price level. It has no idea whether an asset is at an all-time high or eighty percent below one. Feed it the same sequence of percentage moves at any price and it returns the same number. It is not a valuation tool and it cannot be made into one.

Second, the "fourteen periods" is doing enormous work. Fourteen periods on an hourly chart is fourteen hours. On a weekly chart it is a quarter of a year. The indicator gives completely different readings on the same asset at the same moment depending on which chart it sits under, and neither is wrong. Timeframes covers why this happens and how to stop it from confusing you. A stated RSI value without a stated timeframe is not information.

Are 70 and 30 buy and sell signals?

No, and this is where most of the damage gets done.

The words "overbought" and "oversold" are unfortunate. They sound like verdicts. They are shorthand for "recent movement has been lopsided in one direction," which is a description of the past fourteen periods, not a prediction about the next one.

Consider what a high RSI actually means about an asset. It means buying pressure has been consistently stronger than selling pressure. Restated that way, the naive conclusion inverts: an asset whose buyers have been dominating for two weeks is an asset in a strong uptrend. Wilder himself, who created the thing, described readings above 70 as a condition to watch rather than an instruction, and the popular reading has drifted a long way from that.

The empirical version of the problem is stark. In a genuine trend, RSI does not oscillate politely around 50 and touch 70 occasionally. It pins. Bitcoin's larger advances have repeatedly held daily RSI above 70 for stretches of two to six weeks, and the same behaviour shows up on the way down below 30. During those stretches, the indicator is telling you something true and useless: that the trend is strong, over and over, every day. Anyone treating each new reading above 70 as a fresh signal is generating one wrong call after another and blaming the market for it.

The structural reason is that RSI is a bounded oscillator applied to an unbounded quantity. Price can keep going. The indicator cannot go above 100. Once it is compressed against its ceiling, it loses almost all of its ability to distinguish between a strong move and an extraordinary one.

When is RSI actually informative?

In range-bound conditions, mostly. When an asset is oscillating within a band rather than trending, RSI oscillates with it, and its extremes line up with the edges of the range reasonably often. That is not a coincidence and it is not magic; it is the same information the range itself already showed you, expressed differently.

The more interesting use is watching where RSI fails to reach. In a healthy uptrend, pullbacks tend to bottom out around 40 to 50 rather than dropping to 30. When a pullback pushes RSI down to 25 for the first time in months, something about the character of the move has changed. That is a genuine observation about momentum, and it does not require you to believe any threshold is a signal.

None of this tells you what will happen next. Momentum shifts precede reversals and they also precede nothing at all, frequently. The indicator describes the recent past accurately and forecasts the future badly, and a lot of trouble comes from people who have quietly stopped distinguishing between those two jobs.

What is divergence, and how often does it lie?

Divergence is when price and RSI disagree about the shape of a move.

Price makes a higher high; RSI makes a lower high. The new price extreme was reached with less momentum behind it than the previous one, meaning the up moves that got there were smaller relative to the down moves along the way. Bearish divergence is the usual name. Reverse it for the bullish case: a lower low in price that RSI does not confirm.

The mechanism is real. A move made on weakening momentum is genuinely different from one made on strengthening momentum, and reading that difference is the most defensible thing RSI does.

The problem is base rates. Divergence appears constantly. In a long trend it can appear five or six times consecutively while price continues in the original direction, and each instance looks exactly as convincing as the one that eventually coincides with a turn. There is no property visible in the divergence itself that separates the ones followed by a reversal from the ones followed by nothing. People who report high success rates with divergence are almost always applying additional filters they have not articulated, and looking backwards at charts where the answer is already visible.

Divergence also depends heavily on where you decide the peaks are, which is a judgement call, and on a phone screen it is a judgement made from a compressed line in a panel maybe fifty pixels tall. Honest advice for mobile: if you cannot see the divergence clearly without zooming, you are pattern-matching on rendering artefacts. Reading a candlestick chart on a phone covers what does and does not survive at that scale.

How do you use RSI without letting it use you?

Treat it as one description of recent behaviour, attached to a specific timeframe, sitting alongside price and volume rather than above them. An RSI reading that contradicts what volume and price structure show is not an override; it is a disagreement to sit with.

Keep the period at 14 unless you have a specific reason to change it, because shorter settings mostly produce more extremes rather than more information. And be sceptical of momentum readings that arrive at the same moment as a headline, since news and price moves frequently explain a spike more directly than any oscillator does. Keep one indicator on the chart at a time on mobile. Two panels under a price chart on a 6-inch screen leaves each of them too short to read a crossing accurately, and moving averages have the same constraint.

And notice when you are consulting the indicator versus when you are shopping for permission. RSI at 68 on the 4-hour and 34 on the daily is not a contradiction to resolve by picking the one you like. It is two accurate answers to two different questions, and which one is relevant depends on a holding period you should have decided before you opened the chart.

The next thing worth learning is not another oscillator. It is what happens underneath the price: whether a move carried real participation or not, which is a question RSI structurally cannot answer.