What Is the RSI Indicator, and What Is It Actually Counting?

The RSI takes the average size of a market's up closes over a set number of bars, divides it by the average size of its down closes over the same bars, and rescales that ratio onto a line running from 0 to 100. It was published by J. W. Wilder Jr. in 1978, in a book called New Concepts in Technical Trading Systems, and the specification below is the one set out in Park and Irwin's review of the technical analysis literature.

Everything else attached to the indicator — the 14-bar setting, the 70 and 30 lines, divergence, failure swings — is a convention layered on top of that one ratio, and the two are worth keeping apart.

What is the RSI indicator?

Wilder's RSI belongs to the momentum oscillators, a family of indicators whose output swings either side of a neutral point — zero, in the usual case. Park and Irwin's review places Wilder's RSI in that family and calls it widely known. Quantower's documentation for the study files it under the same family, describing it as a measure of the speed of price movement and of the change in that movement, and states that its output runs between 0 and 100.

The plainest member of the family is a difference: today's price against the price n days back. The CFTC's futures glossary defines momentum in similar terms, as relative price change across a stated interval, linking the idea to speed and to relative strength. The review quotes Wilder's own 1978 description of the momentum oscillator as measuring "the velocity of directional price movement".

Wilder's addition was a scale. According to the review, the RSI was built to deal with two weaknesses in raw momentum readings: erratic movement in the output, and the absence of any objective scale for how large a reading was. A raw difference is denominated in the instrument's own points; a ratio mapped onto 0 to 100 is not.

All of this sits inside what the CFTC glossary calls technical analysis — forecasting from price-change patterns, rates of change, volume and open interest, with fundamentals excluded. That entry attaches a condition worth keeping in view: for the approach to work with any consistency, the random walk theory of price movement has to be wrong.

How is the RSI calculated?

The RSI needs two running averages and one division. Park and Irwin's review sets out the definitions. For each bar, an up close is the amount by which the close exceeded the previous close, and a down close is the amount by which it fell short; a bar that goes the other way contributes nothing to that side. Sierra Chart's documentation writes the same two quantities as upward and downward change, and makes the zero explicit: whichever of the two did not happen is set to zero at that point in the series.

Those two series are then averaged over n bars. The review gives the seed as the plain n-bar mean, and every subsequent value by the recursion (previous average × (n − 1) + newest value) / n. Relative strength is the ratio of the two averages, and the index is 100 − 100 / (1 + RS). Quantower's page prints the identical algebra, with the ratio written out in words.

Reading the formula backwards explains both ends of the scale. Take the simple-mean version of it: if none of the last n bars closed down, the down-close average is zero, the ratio has nothing to divide by, and the line pins at the top. Sierra Chart's page handles that case explicitly, returning 100 when the downward average is zero and the upward one is not, and 0 in the reverse case. A reading of 100 on that version therefore says the window held no down closes, and says nothing at all about how large the up closes were.

Equal averages give a ratio of 1, and 100 − 100 / (1 + 1) is 50, so the midpoint of the scale is the point where the up and down averages match.

What do overbought and oversold mean on the RSI?

Overbought and oversold are opinions about price, and the CFTC glossary frames both of them as exactly that — a technical opinion. Its overbought entry describes a price judged to have gone both too far and too fast against the fundamentals; oversold is the mirror for a fall. Neither entry mentions a number, a formula or an indicator.

The numbers come from the system Wilder specified. The review records his original parameter values as a lookback of 14 and an entry threshold of 30, and it writes the thresholds as a symmetric pair — the threshold itself and 100 minus the threshold, so 30 and 70. The trading rules in the review are crossings rather than levels: a long when the index drops under the lower threshold and then climbs back through it, a short when it rises above the upper one and then falls back through. The signal is the return, and the two parameters are the lookback and the threshold.

Quantower's documentation describes 70 and 30 as the traditional levels and says they can be moved to suit the instrument, giving 80 as its example of a raised upper line. It also carries a caveat: a persistent trend can hold the reading inside the overbought or oversold zone for a long stretch. Sierra Chart's study simply draws two horizontal lines wherever two numeric inputs put them.

None of the Sierra Chart study pages read here prints a default lookback or a default line value, and neither does Quantower's RSI page. The 14 and the 70/30 trace back to Wilder's own choices as recorded in the review, which makes them a convention with an author rather than a property of the indicator.

Why do two platforms show a different RSI for the same market?

One reason is that the averaging inside the formula is a setting, and the default differs. Sierra Chart's studies reference lists eight studies with RSI in the name — a Stochastic RSI and a Laguerre RSI among them — which shows how loosely the label names one thing. Four of the eight are compared below.

The first two are the live issue. The base RSI study takes an Average Type input whose six alternatives to the simple mean are Wilder's, exponential, smoothed, weighted, linear regression and a simple mean that skips zeros, and that one input sets every average in the calculation at once. RSI - W is the same study with Wilder's as the default instead of a simple mean. Sierra Chart's page for the Welles Wilders moving average gives it as a recursion adding 1/n of the gap between the new value and the previous average — the averaging the review attributes to Wilder. Two different functions of the same data, so the two studies draw two lines from one chart.

RSI - TS looks like a third formula and is not. Its page builds the study from a net change average and a total change average, which resembles nothing in the up/down pair. The algebra collapses: net change is the up series minus the down series, total change is their sum, the recursion is linear, so 50 × (net/total + 1) rearranges to exactly 100 − 100 / (1 + up/down) — Wilder's index in different notation.

RSI - Connors is the genuine departure. Sierra Chart describes it as the mean of three components, only one an RSI computed on price; the others are an RSI of a streak counter and a rate-of-change percentile.

What does a 14-period RSI count on a chart that is not built from time?

A 14-period RSI counts 14 bars, and what a bar is depends entirely on the chart's aggregation. Quantower's documentation draws the line clearly: a time bar closes when a set period has elapsed — its example is five minutes — while a tick bar closes when a set number of trades has printed, its example being 500.

That makes "14 periods" a different window on different chart types, on the same instrument at the same moment. Fourteen five-minute bars is seventy minutes of clock. Fourteen 500-tick bars is 7,000 trades, and how much clock time that spans depends on how fast the market happens to be trading. The RSI formula does not know or care which it was handed; it is computed from the closes of whatever fourteen bars the chart built.

Quantower adds a second wrinkle for anyone using tick bars as the input. Its page states that trade size plays no part in the count: a tick can be a single contract or ten thousand of them, and each counts as one. So a tick-bar RSI is computed over a fixed number of transactions of unknown aggregate size. Our post on tick charts and time charts works through what else changes when the bar stops being a unit of time.

What is RSI divergence, and what does the documentation claim for it?

Divergence is the condition where price makes a new high or low and the RSI does not confirm it, and Quantower's documentation is explicit that this "can signal a price reversal". That is the vendor's claim on the vendor's page, and one of the two signal sources it names.

The same page names failure swings as the second signal source, and both are read on the indicator rather than on price. A Top Swing Failure, in its description, is the RSI making a lower high and then moving below a previous low on its own plot; a Bottom Swing Failure is the reverse.

Quantower's page goes further. It states that chart patterns such as double tops, double bottoms and trend lines can appear on the RSI plot while being absent from the price chart, and directs the reader to look for support and resistance on the RSI itself. It puts that resistance at 50-60 in a bear market, where it gives the reading's usual range as 10 to 60, and the support at 40-50 in a bull market, where the range it gives is 40 to 90.

Treat all of that as one platform's published reading conventions. The CFTC glossary carries no entry for divergence and none for oscillator, so it offers nothing to check these against, and the page itself cites no test behind the ranges or the reversal claim. Our post on double tops and double bottoms covers what formal testing has found when the same shapes are read on price.

What has formal testing found about the RSI?

Formal testing of the RSI is thinner than its popularity would suggest, at least in the review read for this post. Park and Irwin's 2004 survey covers both what they call early studies and modern ones, and tabulates each by market, period, system tested, benchmark and transaction-cost treatment.

Searching that report for the token RSI returns five occurrences. Four of them are in the specification on page 19 — the index formula, the entry-threshold definition, and the two trading rules. One is in the study tables: a single row, Wong, Manzur and Chew (2003), in Table 9, the catch-all table for studies dated 1988 to 2004 that fall outside the review's named methodology categories.

What that row covers is narrow. The market is the Singapore Straits Times Industrial Index, on daily data, over 1974 to 1994 split into three seven-year subperiods, with moving averages and the RSI as the systems tested. The conclusion cell reports statistical significance for every system across each subperiod and the whole span, with single moving average rules ranked first, dual crossover rules second and RSI rules third.

Two columns in that same row limit how far it travels. Under benchmark strategies and optimisation the cell reads "Not considered", and under transaction costs it reads "Not adjusted". A result with no benchmark to beat and no costs deducted is a weaker thing than one that survives both. The review's abstract lists three objections it holds against the wider literature: data snooping, rules and search methods picked after the fact, and risk and transaction costs that are hard to estimate. The first is the mechanism curve fitting works through with numbers. For a test of moving average crossovers across twelve US futures markets, see that post, which draws on a different Park and Irwin report.

What does the RSI change on a simulated funded account?

The RSI changes nothing about position size, which is the part that decides whether an evaluation survives. An RSI reading is a number between 0 and 100 computed from the closes behind it; it says nothing about how many contracts fit inside your remaining room, and a funded account's remaining room is a published figure. The arithmetic that does govern size is in our post on futures position sizing, and the line it measures against is set out in the help center article on how drawdown works.

Where the indicator does intersect with a funded account is in the temptation to tune it. The lookback and the thresholds are both free parameters, and both are easy to adjust after the fact until the signals line up with a chart you have already seen. Two of the three deficiencies the review's abstract lays against the literature are exactly that — data snooping, and choosing the trading rules after the event — and on an account with a hard loss line it costs more than a bad backtest does.

The practical version is narrow and checkable. Know which study your platform loaded, know what the averaging type is set to, know how many bars your lookback actually spans on the chart type you are using, and treat 70 and 30 as one author's choices from 1978 rather than as thresholds the market observes.

FAQ

What is the best RSI setting?

No source cited here recommends one. Park and Irwin's review records Wilder's original parameters as a lookback of 14 and an entry threshold of 30, which is where the familiar 14 and the 70/30 pair come from. Neither Sierra Chart's nor Quantower's reference page prints a default length or line value, and Quantower says the levels can be adjusted to suit the instrument.

Does an RSI above 70 mean it is time to sell?

Not in the system as specified. The trading rules in Park and Irwin's review are crossings: a short signal is the index rising above the upper threshold and then falling back through it, so the signal is the return. Quantower's documentation adds that a strong trend can hold the reading inside the overbought zone for a long stretch, and the CFTC glossary calls overbought a technical opinion.

Why does my RSI differ from another platform's on the same chart?

One documented reason is the averaging type. Sierra Chart's base RSI study defaults to a simple moving average and offers six alternatives through one input that changes every average at once; its RSI - W study is the same calculation with Wilder's average as the default. Wilder's specification, as the review sets it out, uses that recursive average rather than a simple mean, so two studies running different defaults draw different lines from identical data.

Has the RSI been tested on futures markets?

In the review read for this post, no. Park and Irwin's report contains one study row naming the RSI — Wong, Manzur and Chew (2003) — and the market tested there is the Singapore Straits Times Industrial Index on daily data from 1974 to 1994, which is a stock index. Where the report summarises futures markets, the two it singles out for consistent profitability are the channel rule and the moving average, and it adds that the profitability of technical trading across futures markets after the mid-1980s had not been investigated comprehensively.

Is the RSI the same thing as relative strength?

No, and the review's own tables carry both senses under the one name. Wilder's RSI compares one market against its own recent history: up closes against down closes, on the same instrument. The review's early-studies table also lists a system called simply "relative strength", tested by Akemann and Keller on industry groups from 1967 to 1975, whose rule is to hold whichever stock group has been strongest across a thirteen-week stretch and to close the position 52 weeks later. Ranking instruments against each other is a different calculation answering a different question.