The RSI indicator: why overbought is about speed, not price

RSI compares the average size of gains with the average size of losses inside a chosen window and expresses that ratio on a normalised scale with hard bounds. That is why it measures neither how "expensive" a security is nor its fair value, but the pace at which a move absorbs the pressure coming against it. A high reading means that almost all the changes over the latest window went in one direction — and it says nothing about where the price will stop.

What exactly is calculated inside

The mechanics are simple, and that is their strength. For each bar the change in the closing price against the previous close is taken: a positive change goes into the gains bucket, a negative change into the losses bucket, and an unchanged close goes nowhere. Both buckets are then smoothed over the window and divided by each other. The result is forced into a bounded scale — this is how the indicator becomes comparable across securities with different prices and different volatility, which was exactly the task set by Welles Wilder, who described the calculation in 1978.

An important detail of the smoothing: Wilder used exponential averaging, not simple averaging. This means that every new value rests on the entire preceding history, albeit with a declining weight. The practical consequence is that two platforms with the "same" RSI can draw slightly different lines if one of them calculates with a simple average. The gap is small, but near the thresholds it decides whether a signal counts or not.

The bounded scale and what it hides

Normalisation is the main convenience of RSI and its main trap. Because the value cannot leave its bounds, a strong trend runs into the ceiling and stays there until the pressure eases. The indicator has not "broken" in this case: it is honestly reporting that there were almost no pullbacks inside the window. A trader who reads the upper zone as an order to sell ends up, in such a regime, standing against the move at exactly the moment when the move is at its strongest.

Hence the working distinction: in a sideways market the bounds of the scale act as mean-reversion zones, while in a trend they act as confirmation of strength. The regime has to be identified separately; RSI itself does not do that. What actually stands behind the word "overbought" is examined in more detail in the article on overbought indicators.

The period is not a cosmetic setting

The indicator period changes the meaning of the readings, not just the smoothness of the line. A short window produces a line that hits the bounds almost every week: there are many signals and even more noise. A long window reaches the extreme zones rarely, but every visit deserves attention. Choosing the window means choosing how often you are prepared to be wrong, and it cannot be chosen "correctly" in general — only for a specific holding horizon. The formal definition of the parameter is set out in the glossary entry on the indicator period.

A common mistake is to keep fitting the window to past history until the picture looks attractive. What you get that way is a description of the past, not a tool for the future.

Divergence and the failure swing

The most meaningful part of RSI is its divergence from price. The price sets a new extreme, but the indicator does not make a new extreme: this means the new high cost the market less speed than the previous high did. That is a statement about the character of the move, and a testable statement at that.

Wilder regarded not divergence itself but the failure swing as the more reliable signal: the indicator leaves the extreme zone, pulls back, tries to return — and does not return, after which it breaks the level of its own intermediate reversal. Here the signal is the behaviour of the line itself, with no reference to price. What counts as a completed signal, and how it differs from "looks like a reversal", is covered in the glossary.

Lag and repainting

RSI is built on bars that have already closed, which places it among the lagging indicators by its method of calculation, although divergence gives it features of a leading indicator. Inside an unclosed bar the line moves together with the price and may shift by the close — this is not indicator repainting in the strict sense, but a decision taken on a flickering value cannot be reproduced on historical data. Record the signal at the close of the bar, otherwise testing and trading drift apart.

How to test it on your own market

The test is built not on attractive examples but on your own history. Step 1 — fix the window and the entry rule in writing. Step 2 — run it over the available history of liquid securities from the list of stocks. Step 3 — separate trending periods from sideways periods and look at the result for each regime on its own: a figure averaged over the whole period will hide exactly the effect you are calculating for.

It is also worth stripping out the dates of corporate events from the calendar and the days of major news: a gap on the ex-dividend date or on an earnings report breaks the calculation of price changes, and the indicator shows a speed that never existed in actual trading.

This article contains no ready-made statistics on the success rate of RSI signals for Russian securities — such figures depend on the window, the thresholds, the testing period and the exit rules, and transferring someone else's numbers to your own system is pointless. Run the calculation on your own history.