RSI Indicator Explained: Formula, Settings and Why Overbought Is Not a Sell Signal

The Relative Strength Index, or RSI, is a momentum indicator that compares smoothed upward and downward price changes on a scale from zero to 100. Readings above 70 are commonly called overbought and readings below 30 oversold. Those labels describe recent price behavior; they do not guarantee the next move.

A stock can become overbought at $108 and remain overbought as it climbs to $116. Selling merely because the indicator reached 70 confuses strong upward momentum with evidence that the momentum must immediately end.

This guide builds the RSI calculation from individual price changes, explains the common 14-period setting and shows how to evaluate a trading rule without quietly borrowing information from the future.

Reviewed September 2026. Numerical price paths and trade examples are hypothetical and are not performance claims or historical backtests. Visit the Knowledge hub for related market mechanics.

What does the RSI indicator actually measure?

RSI asks how large recent upward changes have been relative to recent downward changes, under a specified smoothing method. It uses price changes, usually between successive closing prices. It does not directly incorporate revenue, valuation, trading volume or the prices of competing stocks.

Despite its name, RSI is not the same as relative performance against a benchmark. A stock can have a high RSI while still underperforming a stronger market. Conversely, it can outperform a falling index while displaying weak absolute momentum.

Fidelity’s technical-indicator guide identifies J. Welles Wilder as RSI’s developer and explains the familiar 70/30 interpretation. It also notes that strong trends can sustain extreme readings.

That limitation is central. RSI compresses part of a price history into one number. Before treating the number as an instruction, you need a hypothesis about why that particular pattern might precede an attractive trade.

The RSI formula

For each period, calculate the closing-price change. A positive change goes into the gain series, with zero in the loss series. A negative change becomes a positive magnitude in the loss series, with zero in the gain series. An unchanged close contributes zero to both.

The losses in this formula are positive magnitudes. Entering negative values in the denominator creates nonsense. Also use changes consistently: substituting percentage changes for dollar changes produces a different indicator from the standard price-change version.

If the two smoothed averages are equal and positive, RS equals one and RSI equals 50. If average gains are three times average losses, RSI equals 75. If average gains are one-third of average losses, RSI equals 25.

For positive total movement, the equivalent expression is RSI = 100 × Average gain ÷ (Average gain + Average loss). This version makes the scale easier to understand and avoids explicitly calculating the ratio RS.

How Wilder’s 14-period smoothing works

A common initialization uses the first fourteen changes: add the gains and divide by fourteen, then do the same with the loss magnitudes. Fourteen changes require fifteen closing-price observations.

After initialization, update each average recursively. New average gain equals (previous average gain × 13 + latest gain) ÷ 14. New average loss uses the corresponding loss values. This gives the newest observation a weight of 1/14 while carrying forward information from earlier prices.

It is not a fresh simple average of only the latest fourteen changes. It is also not the standard exponential moving average with a 2/(n + 1) coefficient. Using either substitution can produce a different line.

TradingView’s RSI documentation shows its RMA-based implementation. When matching a platform calculation, check the smoothing, source price, initialization history and treatment of special cases rather than assuming every RSI label means identical inputs.

A complete example you can reproduce

Start with a closing price of $100 and use these fourteen successive changes: +2, −1, +3, −2, +1, +2, −1, +1, +2, −1, +2, −1, +1, 0. The final close is $108.

The positive changes total $14. The negative magnitudes total $6. Initial average gain is $14 ÷ 14 = $1. Initial average loss is $6 ÷ 14, approximately $0.428571.

RS is therefore 14 ÷ 6, or approximately 2.333333. RSI is 100 − 100 ÷ 3.333333 = 70. The price has risen $8 overall, but RSI is not 8%, and it does not measure an 8% probability of anything.

Now let the next close rise by $1 to $109. Average gain stays at $1 because (1 × 13 + 1) ÷ 14 = 1. Average loss falls to approximately $0.397959 because the latest loss is zero. RSI rises to approximately 71.53.

The decline in smoothed losses helps lift RSI even though the new gain is modest. This is a useful reminder that the indicator depends on the balance of both series, not simply the size of the latest green candle.

Why overbought does not automatically mean sell

Continue the same example with seven additional $1 gains. Price reaches $116 after eight consecutive rising closes beyond the initial calculation. RSI remains above 70 throughout that extension.

The chart shows the actual Wilder-smoothed RSI for this invented sequence. The accompanying closing prices make the interpretation explicit: strong readings coexist with higher prices. This is a mathematical counterexample to the claim that every overbought reading must immediately reverse, not evidence that overbought readings always predict continuation.

Now imagine selling short at $108 solely because the initial RSI is 70. At $116, the position has moved approximately 7.41% against the entry before borrow costs or other expenses. The indicator never promised that 70 was a price ceiling.

The opposite error occurs below 30. Persistent selling pressure can produce a low RSI while the price continues to fall. An oscillator bounded at zero does not create a floor under the underlying security.

Schwab’s discussion of overbought and oversold markets similarly cautions that these conditions can last. Treat an extreme reading as information to investigate, not as a standalone prediction.

The same reading can have different meanings in different markets

Consider a stock moving between well-established price boundaries and another steadily making higher highs. In the range, a high RSI near the upper boundary might fit a mean-reversion hypothesis. In the trend, the same reading might simply accompany a breakout.

That distinction is a starting hypothesis, not a free trading edge. A backtest needs an objective way to classify the market before entering the trade. Labeling a chart a range only after it reverses imports the answer into the rule.

Define any filter using data available at the decision time. It might involve past highs and lows, a moving-average condition or a volatility measure. Then test whether the filter improves results after costs and across periods outside the development sample.

Our VWAP strategy guide discusses another price-context tool. Adding it to RSI changes the hypothesis; it does not automatically make two weak signals into one strong signal.

What are the best RSI settings?

Fourteen periods is a widely used baseline. A shorter lookback responds more quickly to recent changes and generally produces more volatile readings. A longer lookback responds more gradually. Neither is universally superior.

The word period matters. Fourteen daily bars and fourteen five-minute bars summarize very different horizons. On an intraday chart, session boundaries, overnight gaps and extended-hours settings can also change the sequence being measured.

Changing a threshold from 70 to 80 will usually reduce the number of extreme observations in a given dataset. It does not establish that the remaining observations have better net outcomes. Higher selectivity and higher profitability are separate claims.

Choose settings to express a specific hypothesis, then reserve unseen data for evaluation. Trying dozens of lookbacks and thresholds until one historical equity curve looks attractive makes accidental patterns easier to mistake for a repeatable advantage.

RSI divergence: a warning with a timing problem

A bearish divergence typically means price makes a higher high while RSI makes a lower high. A bullish divergence reverses the pattern: price makes a lower low while RSI makes a higher low.

The observation says the price extreme is not matched by the selected momentum measure. It does not specify when a reversal begins, how far it travels or where a trade should exit if the trend continues.

Imagine highs of $100 and $105 paired with RSI readings of 78 and 72. That is the conventional bearish pattern. Price could still rise to $115 while RSI stays elevated. A short trade needs a separate entry condition and invalidation rule.

Many charting tools identify swing highs only after several later bars confirm a pivot. A marker drawn retrospectively on the high can make the signal look available earlier than it actually was. For a valid test, record the confirmation bar and the first executable entry after confirmation.

Manual chart review has a related problem: successful divergences are visually memorable, while failed ones blend into the trend. Define every qualifying signal before counting winners and losers.

RSI versus price, volume and valuation

RSI does not know whether a price move came from broad participation or a few thin trades. Volume can provide a different dimension, but it requires a consistent benchmark. Our relative-volume guide explains why an intraday comparison should account for time of day.

Nor does RSI know whether a company is cheap. An oversold stock can still be expensive relative to deteriorating earnings, and an overbought stock can still trade below a defensible valuation estimate.

A fundamental investor and an intraday trader may use the same indicator for different purposes. The first might investigate the timing of an existing thesis. The second might test a short-horizon price pattern. Neither should mistake the shared label for a shared investment process.

Also avoid counting several transformations of closing prices as fully independent evidence. RSI, moving averages and price momentum all reuse overlapping information. A more elaborate chart can create confidence without adding much new information.

Turn an RSI idea into a testable rule

“Buy when oversold” leaves too much unspecified. A test needs a universe, timeframe, data source, entry condition, execution assumption, exit, position size and cost model.

A purely illustrative rule might observe daily RSI crossing back above 30 after a close below it, then enter at the next session’s opening price. That is a different rule from buying the first close below 30, buying during an unfinished bar or assuming a fill at the signal bar’s closing price.

Specify whether the rule can hold multiple positions, how it treats missing observations and what happens around stock splits or delistings. These details affect the result, especially when testing distressed or illiquid stocks.

Use a chronological holdout period and realistic transaction assumptions. A signal generated at the close is not automatically executable at that same closing price. Limit orders require a defensible fill model; touching a price does not guarantee that every order fills.

Compare the proposed rule with simple alternatives using similar market exposure. A strategy should not receive credit for a rising market without examining whether its timing and risk control added anything beyond holding the exposure.

Win rate does not determine whether the rule makes money

Suppose a hypothetical strategy wins 60% of its trades. Its average winner is $1 per share and its average loser is $2. Before costs, expected profit per trade is 0.60 × $1 − 0.40 × $2 = −$0.20 per share.

The attractive win rate hides a negative payoff balance. Adding commissions, spreads or adverse fills makes the expectation worse. An RSI strategy needs the distribution of gains and losses, not just the percentage of successful signals.

Position sizing translates the trade hypothesis into account exposure. Our position-sizing guide explains how a planned exit distance and risk budget interact. The distance must reflect an executable trade plan, not an arbitrary amount chosen to produce a larger position.

A stop order also does not guarantee the intended price through a gap. Read stop orders, gaps and slippage before treating a spreadsheet’s maximum planned loss as a contractual limit.

RSI FAQ

Is RSI above 70 a sell signal?

It is a conventional overbought reading. Whether selling has a useful expected outcome depends on the broader rule and evidence. The worked price sequence demonstrates why the threshold alone cannot guarantee a reversal.

Is RSI below 30 a buy signal?

It identifies relatively strong recent downward movement under the chosen settings. That can occur in a temporary pullback or a continuing collapse. A low number does not establish a bargain or limit further losses.

What happens when average loss is zero?

If average gain is positive and average loss is zero, the limiting RSI value is 100. If gains are zero and losses positive, it is zero. If both are zero, the expression is undefined; platforms may use different conventions for that flat-series case.

Why does my RSI differ between platforms?

Check adjusted prices, trading sessions, bar timing, selected price source, smoothing and the amount of initialization history. Different data can produce different values even when both indicators display the same period length.

The useful question comes after the indicator reading

Once you know RSI, ask what observation would change your trade thesis, where execution is plausible and what evidence supports the proposed response. A rule without those answers remains a chart interpretation.

RSI is useful because it summarizes the balance of recent price movement. Its limits follow from that same compression: it cannot supply a valuation, an execution plan or a tested edge on its own. Use the calculation to sharpen a question, then test the complete decision.