How to Read RSI: Formula, Trend Ranges and Divergence

An RSI reading of 72 does not mean an asset is expensive, just as a reading of 28 does not mean it is cheap. It means that upward or downward changes have dominated the indicator’s smoothed lookback. Whether that imbalance precedes continuation, consolidation or reversal depends on trend, location, the price series used and what happens after the reading. Treating 70 and 30 as automatic orders removes precisely the context that makes the Relative Strength Index useful.

This is Part 5 of the Financial Chart Basics series. The previous guide on volume and open interest showed how to separate activity from direction. RSI asks a different question: how have recent gains compared with recent losses? This guide works from the calculation to a repeatable decision process for forex, stocks, gold and futures.

What RSI measures—and what “relative strength” does not mean here

RSI is a bounded momentum oscillator introduced by J. Welles Wilder. Its value runs from 0 to 100 because it converts the ratio between smoothed positive and negative price changes into a fixed scale. A rising value says recent gains are becoming more important relative to recent losses; a falling value says the opposite. It does not measure valuation, order flow, positioning or the performance of one security against a benchmark.

That last distinction prevents a common category error. “Relative strength” can also mean that one asset is outperforming another, such as a stock divided by an equity index. Wilder’s RSI is not that ratio. It normally uses one input series—typically completed closing prices—and compares that series with its own prior values.

The period is a number of bars, not necessarily a number of days. RSI(14) on a five-minute chart processes fourteen five-minute changes after its seed; RSI(14) on a daily chart processes daily changes. The two indicators describe different horizons and cannot be substituted for one another.

The Wilder calculation, from the first value onward

Let Ct be the chosen input for the current bar and Ct-1 the previous one. Split the change into a non-negative gain and a non-negative loss:

Change = Ct − Ct-1

Gain = max(Change, 0)

Loss = max(−Change, 0)

For a period of n, calculate the relative-strength ratio and transform it:

RS = Average Gain / Average Loss

RSI = 100 − 100 / (1 + RS)

The standard fourteen-period version needs fourteen changes, so the first value requires fifteen prices. Its initial average gain is the arithmetic mean of the first fourteen gains; the initial average loss is the arithmetic mean of the first fourteen losses. After that seed, Wilder smoothing is recursive:

New Average Gain = (Previous Average Gain × 13 + Current Gain) / 14

New Average Loss = (Previous Average Loss × 13 + Current Loss) / 14

This is equivalent to exponential smoothing with an alpha of 1/14. It is not a fresh fourteen-bar simple average on every candle. Older observations fade rather than disappearing abruptly, which is why the amount and quality of history loaded before the visible chart can matter.

If average loss is zero while average gain is positive, RSI tends to 100; if average gain is zero while average loss is positive, it is 0. A completely flat sample produces an undefined ratio of zero divided by zero. Charting packages do not all handle that edge case in the same way, so a flat-series output should be checked rather than assumed.

A reproducible example: why 71.43 can become 61.90

Suppose fourteen completed price changes contain 10 points of total gains and 4 points of total absolute losses. The initial average gain is 10/14 = 0.7142857; the initial average loss is 4/14 = 0.2857143. RS is 2.5 and RSI is therefore:

100 − 100 / (1 + 2.5) = 71.4286

Now suppose the next completed bar falls by 2 points. The new gain is zero and the new loss is 2:

New Average Gain = (0.7142857 × 13 + 0) / 14 = 0.6632653

New Average Loss = (0.2857143 × 13 + 2) / 14 = 0.4081633

New RS is 1.625, giving RSI = 61.9048. The oscillator has left the conventional overbought zone after one negative bar, but the earlier gains have not vanished. The example explains the mechanics; it says nothing about where the next price bar must go.

Why two screens can show different RSI(14) values

A disagreement is not automatically a calculation error. Before comparing readings, make the full specification identical:

Seed differences are most visible near the left edge of a dataset or after a gap in history. Because the recursive weight decays, longer clean warm-up history usually brings implementations using the same formula closer together. Rounding only at the displayed value is harmless; rounding average gains and losses at every step can compound into a meaningful difference.

A four-hour RSI is also not the arithmetic average of four one-hour RSI readings. The four-hour chart first constructs its own closes and then applies a nonlinear ratio to those changes. Aggregation and indicator calculation do not commute.

Fourteen, 70 and 30 are defaults, not market laws

Wilder’s fourteen-period setting is a sensible common baseline because it balances responsiveness and smoothing. A shorter length reacts earlier but crosses zones more often; a longer length moves more slowly and may hide changes that matter to a short holding period. Choosing 7, 14 or 21 after seeing which one fitted a historical turning point best is parameter mining, not validation.

The traditional interpretation marks readings above 70 as overbought and below 30 as oversold. Those words describe an unusually one-sided recent path, not an objectively excessive price. In a durable advance, repeated readings above 70 may be evidence of leadership. In a persistent decline, sub-30 readings may describe continuing downside momentum. Selling the first case or buying the second without independent price evidence is an attempt to pick a turning point against the active regime.

Read the zone differently in a trend and in a range

Thresholds become more informative after the price environment is classified. Analysts often observe an RSI regime around 40–80 or higher during an established uptrend, with pullbacks finding momentum support around 40–50. In a downtrend, the working range can shift toward roughly 20–60, with rebounds stalling around 50–60. These are descriptive hypotheses, not fixed borders; the instrument, timeframe and volatility regime must be tested.

The sequence matters: environment first, location second, trigger last. RSI at 28 in the middle of an accelerating downtrend is not the same observation as RSI at 28 after a failed break below a well-tested range.

Divergence is a warning; price structure supplies the clock

A classical bearish divergence occurs when price makes a higher confirmed swing high while RSI makes a lower swing high. A bullish divergence reverses the relationship: price makes a lower low while RSI forms a higher low. The indicator is saying that the latest extension was achieved with a different momentum profile. It is not saying that reversal must begin on the next bar.

Three disciplines make divergence testable. First, define the pivot rule before looking at the outcome—for example, a swing that remains an extreme for a specified number of completed bars on each side. Second, compare pivots of the same scale; a daily price high should not be paired with an arbitrary fifteen-minute RSI wiggle. Third, give the setup an expiry and a price condition.

A bearish divergence may be upgraded after price fails to hold the new high and closes below the reaction low between the two peaks. It is degraded when price spends several completed bars above the breakout, RSI re-expands and the market continues to print higher lows. A bullish setup uses the opposite tests. This approach prevents a divergence from being redrawn indefinitely while the trend keeps moving.

Failure swings require their final pivot break

A failure swing is an RSI pattern rather than a direct comparison between price and indicator slopes. In a bearish version, RSI forms a high in or near its upper zone, retreats, rebounds to a lower RSI high and then breaks the intervening RSI low. In a bullish version, it forms a low in or near the lower zone, rebounds, retests at a higher RSI low and then breaks the intervening RSI high.

The final pivot break completes the pattern. Calling the first retreat from 70 a failure swing creates hindsight bias because the necessary structure does not yet exist. Even after completion, price should confirm: a failed breakout, a close through a relevant swing or loss of a support/resistance zone has more weight than an oscillator pattern occurring in isolation.

Divergence and a failure swing can coexist, but they are not synonyms. Divergence compares two price pivots with two indicator pivots; a failure swing can be identified within RSI alone. Record them as separate conditions so a backtest does not count the same observation twice under different names.

The same reading has different data risks across four markets

Stocks: gaps and corporate actions enter the calculation

A daily earnings gap is a genuine close-to-close change and may dominate a short RSI. A share split, special distribution or badly adjusted history can produce a similar visual extreme for purely mechanical reasons. State whether the series is adjusted, whether extended-hours data are included and whether the signal existed on the actual tradable session. An RSI of 25 created by a data break is not an oversold market.

Spot FX: there is no single global close

Spot foreign exchange is decentralised. Different quote feeds, a small Sunday bar and different New York or UTC day boundaries can change the sequence of daily closes. A four-hour EUR/USD divergence should therefore be checked on the same provider and session used for execution, rather than confirmed by a screenshot built from another clock.

Gold: align the instrument before comparing momentum

XAU/USD spot and exchange-traded gold futures have different venues, trading pauses and contract mechanics. Their broad trend may agree while their intraday RSI differs. Cross-market agreement can be useful context, but one reading cannot be silently substituted for the other. Match timestamps and specify whether the thesis concerns spot exposure or a particular futures contract.

Futures: the continuous chart may manufacture the move

A continuous series joins delivery months. A raw roll gap can create an RSI jump that was not tradable in either contract; a back-adjusted series can remove the gap but rewrite earlier price levels. Check the active contract, rollover method and liquidity migration. A signal should exist in an executable contract before it is treated as market evidence.

Completed bars and multiple timeframes: one hierarchy, not a vote

The current RSI value changes while the source candle is open. A four-hour reading can touch 72 after an early rally and close at 64. If a rule was tested on completed bars, acting on the transient value silently changes the system. Intrabar use is legitimate only if the entry, data frequency and slippage assumptions were built and tested that way.

Multiple timeframes should have assigned jobs. A weekly or daily chart defines trend and major location; a four-hour or hourly chart measures the setup; a lower chart can refine execution. Counting “two bullish and one bearish” readings as a majority vote mixes different horizons. Daily RSI above 40 during an uptrend and hourly RSI below 30 on a pullback can both be correct.

A coherent example would require the daily uptrend and support zone to remain intact, then ask the hourly chart to reclaim its local boundary after an oversold reading. The higher-timeframe close defines thesis invalidation; the lower timeframe controls entry precision. If the intended holding period is hours, waiting for a monthly RSI turn adds impressive-looking but irrelevant information.

Backtest the rule, not the chart annotation

An honest RSI test fixes the symbol, venue, data source, session, input, length, smoothing, warm-up and bar-completion rule before measuring results. Divergence pivots must be detectable using information available at the time. If a swing high needs three future bars to be confirmed, the test cannot enter on the original high.

Report more than win rate. Sample count, median result, maximum favourable and adverse excursion, holding time, failure frequency and out-of-sample performance show how the rule behaves. Include spread, commission, slippage, gaps and futures roll costs. Separate macro-release windows from ordinary sessions; an indicator that appears stable only because event bars were omitted without a rule will not survive live use.

Sensitivity is another defence against overfitting. A thesis that works at RSI 29.7 but fails at 29 or 31, or works with length 14 but collapses at 13 and 15, may have learned historical noise. Results should also be segmented by trend and range, because averaging the two regimes can conceal opposite behaviours.

A seven-step RSI workflow in FastBull

1. Fix the market definition. Record the exact instrument, venue or feed, timeframe, session and whether stock or futures history is adjusted.
2. Fix the indicator definition. Use the intended input, period, Wilder smoothing and sufficient warm-up. Do not compare it with an unspecified implementation.
3. Classify the environment. Mark trend, range and key price zones before viewing the oscillator as a trigger.
4. Name the observation. Distinguish an extreme, a 50-line shift, a divergence and a completed failure swing; do not merge them into one “RSI signal”.
5. Write confirmation and invalidation. Specify the completed price close, swing break or failed breakout that upgrades the idea, plus the price action and time limit that cancel it.
6. Check the clock and event risk. Wait for the required bar to finish and note scheduled releases, earnings, settlement and rollover that can alter execution.
7. Review the evidence. In FastBull charts, save the exact settings and replay both successful and failed cases without revealing future bars.

The most useful question is not “Is RSI overbought?” but “What chain of completed price evidence would make this momentum reading matter here?” RSI provides a compact description of recent directional pressure. Trend, location, data quality, confirmation and risk limits determine whether that description deserves a place in a decision.