Relative Strength Index Explained

Table of Contents

    The Relative Strength Index (RSI) turns a security’s recent gains and losses into a number from 0 to 100. It can help describe whether upward or downward price changes dominate, but an extreme reading does not establish that a reversal is due. Understanding the calculation is especially useful when two charts show different RSI values for the same security.

    This guide works through a 14-period RSI, shows why Wilder’s smoothing differs from a rolling simple average, and compares RSI with the Chande Momentum Oscillator, DeMarker and the disparity index. Each transforms price information differently; none supplies a complete investment decision.

    What Is the Relative Strength Index?

    A Balance of Gains and Losses

    RSI compares the average magnitude of upward price changes with the average magnitude of downward changes in the same series. J. Welles Wilder developed the indicator. Its familiar 14-period setting refers to price bars: 14 daily changes on a daily chart or 14 hourly changes on an hourly chart. [1] [2]

    The word relative can cause confusion. RSI’s gain/loss comparison does not rank a stock against other stocks or a benchmark. Cross-sectional momentum investing compares securities’ past performance within a universe. A high RSI and a high cross-sectional rank answer different questions, even when they occur together.

    The RSI Formula and Wilder’s Smoothing

    Let Ct be the closing price for period t. Split each change into two nonnegative amounts:

    Separate Upward and Downward Changes
    Gt = max(Ct − Ct−1, 0)
    Lt = max(Ct−1 − Ct, 0)
    G is the gain magnitude; L is the loss magnitude. A loss enters as a positive number, and an unchanged close gives zero for both.

    For a fresh 14-period calculation, divide the sum of the first 14 gains by 14 to obtain the initial average gain, AG,14. Do the same for losses to obtain AL,14. Include zero observations; do not divide only by the number of up or down days. Fourteen changes require 15 closing prices.

    Relative Strength and RSI
    RS = AG / AL
    RSI = 100 − 100 / (1 + RS)
    When total average movement is positive, the equivalent form is RSI = 100 × AG / (AG + AL).

    After that seed, Wilder’s method updates both averages recursively. For each new bar in the 14-period version:

    Wilder’s 14-Period Update
    AG,t = (13 × AG,t−1 + Gt) / 14
    AL,t = (13 × AL,t−1 + Lt) / 14
    Each new component receives weight 1/14; the preceding average retains weight 13/14. For n periods, replace 13 with n − 1 and 14 with n.

    Fidelity’s momentum transcript describes the initial averaging and Wilder’s subsequent use of the prior averages. [2] This recursion retains a diminishing influence from older changes. A rolling simple RSI instead recomputes equal-weight averages from only the latest 14 changes, dropping the oldest observation completely. Both start with the same seed under this setup, then can diverge.

    Wilder’s smoothing also differs from the usual exponential-moving-average setting of 2/(n + 1): for n = 14, those new-observation weights are 1/14 and 2/15, respectively. A platform’s label “14-period RSI” therefore does not, by itself, establish its exact averaging and initialization convention.

    A Complete 14-Period RSI Example

    Consider a hypothetical share with an initial close of $100. The table supplies 14 subsequent daily closes, including an unchanged day. All prices and changes are invented for arithmetic; they are not historical prices or trading results. The first row establishes the starting price and supplies no change.

    Hypothetical daily prices and the 14 changes used to seed RSI
    Period Close (USD/share) Change (USD/share) Gain G (USD/share) Loss L (USD/share)
    0 100 — — —
    1 102 +2 2 0
    2 101 −1 0 1
    3 104 +3 3 0
    4 102 −2 0 2
    5 103 +1 1 0
    6 103 0 0 0
    7 102 −1 0 1
    8 104 +2 2 0
    9 102 −2 0 2
    10 105 +3 3 0
    11 104 −1 0 1
    12 105 +1 1 0
    13 104 −1 0 1
    14 106 +2 2 0
    Totals for periods 1–14 — +6 14 8

    The seed averages are AG,14 = 14/14 = 1 and AL,14 = 8/14 = 4/7. Relative strength is 1 / (4/7) = 1.75. Therefore:

    Initial RSI

    RSI14 = 100 − 100/(1 + 1.75) = 63.6364, or 63.64 rounded to two decimals. Gains account for 14/(14 + 8) of the total absolute movement; this is not the percentage of days that rose.

    The Next Day: Wilder vs Rolling Simple RSI

    Now the hypothetical share falls from $106 to $102 at period 15. The new gain is zero and the new loss is $4. Keep the unrounded seed averages for the update.

    • Wilder average gain: (13 × 1 + 0)/14 = 13/14 = 0.928571.
    • Wilder average loss: [13 × (4/7) + 4]/14 = 40/49 = 0.816327.
    • Wilder RSI: 100 × (13/14) / [(13/14) + (40/49)] = 53.22.

    The rolling alternative removes period 1’s $2 gain and adds the new $4 loss. Its latest 14 changes have total gains of 14 − 2 = 12 and total losses of 8 + 4 = 12. Equal simple averages give RSI = 100 × 12/(12 + 12) = 50.00. The difference is caused by the weighting rule, even though the new price and nominal lookback match.

    Zero Changes, Zero Losses and Flat Prices

    The gain-share form makes the boundary cases clear. If average gain is positive and average loss is zero, RSI is 100. If average gain is zero and average loss is positive, it is 0. If both are zero, the equation is 0/0 and undefined; check how the software handles that case rather than interpreting a displayed substitute as a directional signal.

    A single unchanged close does not normally erase prior history. If period 16 is flat after the example’s decline, Wilder’s two averages both multiply by 13/14. Their ratio stays the same, so RSI remains 53.22 in exact arithmetic. A newly initialized all-flat series and a flat bar following earlier movement are different cases.

    Interpreting RSI: Thresholds and Divergence

    Above 70 and below 30 are traditional “overbought” and “oversold” labels. Strong trends can leave RSI in these zones for extended periods. They describe the recent price-change balance; they do not say that an asset is fundamentally expensive or cheap. [1]

    For positive average loss, RSI of 70 implies RS = 70/30, about 2.33: average gains are about 2.33 times average losses under the chosen method. RSI of 50 means those averages are equal. Neither equation contains tomorrow’s return, an entry price, a holding period or trading costs. Those additional choices are necessary before a threshold becomes a testable trading rule.

    A bearish divergence means price makes a higher high while RSI makes a lower high; a bullish divergence pairs a lower price low with a higher RSI low. [1] Such a mismatch can prompt closer examination, but price can keep trending afterward. It supplies no automatic deadline or magnitude for a reversal.

    The market regime changes the question you ask. In a trading range, an analyst might investigate whether an extreme accompanies a return toward the range’s interior. In a persistent trend, the same extreme may accompany continued directional movement. Define those hypotheses before evaluating results; retrospectively choosing the explanation that fits the chart does not test forecasting skill.

    The Chande Momentum Oscillator and Its RSI Connection

    The Chande Momentum Oscillator (CMO) uses the difference between summed gains and losses, divided by their total. The unsmoothed window version below follows Fidelity’s documented formula. Its guide describes 20 periods as usual; this example uses 14 solely to match the RSI comparison. [3]

    Unsmoothed Window CMO
    CMO = 100 × (U − D)/(U + D)
    U and D are sums of gain and loss magnitudes over the same chosen window. With positive total movement, the range is −100 to +100.

    Using the seed window, CMO = 100 × (14 − 8)/(14 + 8) = 27.27. A positive value means gains exceed losses in the specified window. If their sums are equal, CMO is zero; if both sums are zero, its denominator is undefined.

    There is a useful identity, but its conditions matter. When RSI uses the same unsmoothed sums, same observations and same lookback, RSIsimple = 100U/(U + D). Therefore:

    Identity for Matching Unsmoothed Inputs
    CMO = 2 × RSIsimple − 100
    Requires U + D > 0. It is not a universal identity between an unsmoothed CMO and Wilder’s smoothed RSI.

    Period 15 proves the distinction. Its rolling RSI is 50.00, so the matching CMO is 2 × 50 − 100 = 0. Applying the same transformation to Wilder’s unrounded RSI instead gives 2 × 53.216374 − 100 = 6.43. That is a different result because the inputs have different weights.

    Do not carry thresholds across scales unchanged. For matching unsmoothed inputs, CMO +50 maps to RSI 75, and CMO −50 maps to RSI 25. The commonly discussed CMO ±50 guides therefore do not equal RSI’s 70/30 guides. No such threshold establishes profitability.

    DeMarker: Changes in Highs and Lows

    DeMarker shifts the input question from closing-price changes to the expansion of successive highs and lows. The following convention uses simple moving averages of the two components, as documented by MetaTrader 5. [4]

    DeMarker Components and Ratio
    DeMaxt = max(Ht − Ht−1, 0)
    DeMint = max(Lowt−1 − Lowt, 0)
    DeM = SMAn(DeMax) / [SMAn(DeMax) + SMAn(DeMin)]
    H is the bar’s high and Low is its low. This unmultiplied formula ranges from 0 to 1 when the denominator is positive.

    For a compact hypothetical example, use three comparisons from four bars. Highs are 101, 103, 102 and 104; lows are 98, 99, 97 and 98, all in dollars per share. DeMax is 2, 0, 2, while DeMin is 0, 2, 0. Thus DeM = (4/3)/[(4/3) + (2/3)] = 0.6667. The shared division by three cancels, giving the same result as 4/(4 + 2).

    Check the display convention before reading any levels. Here 0.30 and 0.70 are the counterparts of 30 and 70 on a version multiplied by 100. Lookback, smoothing and scale must match when comparing providers; the name alone is insufficient. If both component averages are zero, the displayed ratio needs a defined software convention. DeMarker’s high/low inputs also mean its value need not agree with RSI, even on identical bars.

    The Disparity Index: Distance From an Average

    The disparity index asks how far the latest price lies above or below a chosen moving average. We use the moving-average-denominator convention documented by Definedge’s indicator library: [5]

    Percentage Gap From a Moving Average
    Disparity = 100 × (C − MA)/MA
    C is the current close; MA is the selected moving average. The example assumes a positive MA. A zero denominator is undefined.

    If a hypothetical share closes at $105 against a chosen average of $100, its disparity is 100 × (105 − 100)/100 = +5%. At $95 against the same $100 average, it is −5%. Zero means price equals that average; it does not mean gains and losses are balanced as they are at RSI 50.

    Record the average type and length. For example, thinkorswim’s DisparityIndex documentation specifies an exponential moving average. [6] A simple-average version can produce another number, and similarly named studies should be checked for their exact percentage denominator. It uses a zero-centered percentage scale rather than RSI’s 0–100 range. A +5% reading measures distance, not a 5% forecast decline or proof of overvaluation.

    RSI vs CMO, DeMarker and Disparity

    What the four price indicators summarize under the conventions used here
    Indicator Inputs and averaging Scale and balance point Key distinction
    Wilder RSI Close-to-close gains/losses; simple seed, recursive updates 0–100; equal averages at 50 Retains diminishing influence from older changes
    Window CMO Unsmoothed sums of close-to-close gains/losses −100 to +100; equal sums at 0 A rescaling of matching simple RSI, not generally Wilder RSI
    DeMarker High increases and low decreases; same-window simple averages 0–1 here; equal components at 0.5 Uses high/low expansion rather than changes in the close
    Disparity index Current close relative to a specified moving average Percentage gap; price equals average at 0% Measures distance from a trend reference

    These tools often reuse much of the same underlying price information. Agreement among several transformed versions of one series is not automatically several independent pieces of evidence. In the matched unsmoothed RSI/CMO case, the relationship is exactly algebraic.

    How to Evaluate an Indicator Reading

    1. Specify the data. Record the instrument, bar interval, session boundaries, price field and treatment of splits or distributions. An artificial discontinuity in an inconsistent price series changes the indicator too.
    2. Specify the calculation. Record the lookback, smoothing rule, initial seed, available warmup history and zero-denominator handling. Match those before investigating numerical differences.
    3. Separate closed bars from an unfinished bar. A reading based on a currently changing price can move before the bar closes. Compare values at the same observation time.
    4. State the hypothesis. Define what qualifies as an extreme, trend or divergence, and what subsequent outcome would count as support. A shorter Wilder lookback gives each new component more weight; test sensitivity to settings without choosing them solely for the best historical result.
    5. Evaluate implementation separately. Specify entry/exit timing, position limits and realistic trading execution costs. Test on observations not used to select the rule. Describing a chart and establishing a repeatable net-return advantage are separate tasks.

    MarketXLS documents an RSI retrieval function with a default 14-period parameter and a historical-date input. That documentation does not establish its exact seed/warmup parity with this worked series, access under every plan or unrestricted real-time data; its data footer includes delay and exchange-subscription qualifications. [7] Apply the same method check to any vendor-supplied indicator.

    Common Mistakes

    • Counting 14 prices as 14 changes. A fresh seed needs the prior close plus 14 new closes.
    • Averaging only winning or losing days. Divide each seed total by all 14 periods, including zero components.
    • Entering negative average losses. The loss component is a nonnegative magnitude.
    • Treating 70 as an automatic sale or 30 as an automatic purchase. Define and evaluate a complete decision rule rather than inferring a reversal from a label.
    • Comparing indicators without their conventions. Wilder and rolling averages, different timeframes and DeMarker scales can all produce legitimate differences.
    • Assuming CMO always equals twice Wilder RSI minus 100. The identity shown here requires matching unsmoothed inputs.
    • Confusing RSI with cross-sectional strength. A single security’s gain/loss balance does not establish its rank among other investments.

    Limitations of Momentum Oscillators

    A Description Is Not a Return Forecast

    The formulas compress historical price information. They do not incorporate every new business development, investor objective or implementation cost. A precise calculation can coexist with an unreliable trading rule.

    Changing the lookback changes the question; changing the smoothing changes the weights. Extreme readings can persist, divergences can appear repeatedly before a reversal, and a moving average can catch up with price without a price decline. No fixed threshold is demonstrated here to earn a profit across assets, periods and regimes.

    Assessing whether past-price information produces an investable advantage belongs to the broader market-efficiency question. The worked examples establish arithmetic and interpretation, not empirical evidence against market efficiency or proof of forecasting skill.

    Frequently Asked Questions

    The seed uses 14 price changes, and the first change requires a starting close. Fifteen closes provide those 14 differences. A running platform may use additional earlier history to initialize its smoothed averages.

    No. After a Wilder seed with positive total movement, a flat bar reduces both component averages by the same factor and leaves RSI unchanged in exact arithmetic. An all-flat fresh seed produces a zero denominator, which requires a software handling rule.

    It does not establish a sale decision. The reading summarizes the balance of recent changes under a particular method. A decision also needs a defined objective, timeframe, risk limits and evidence for the proposed rule after costs.

    The unsmoothed CMO defined here is exactly 2 × simple RSI − 100 when both use the same gain/loss observations and window and the total movement is positive. That does not generally hold when the RSI uses Wilder’s recursive averages instead.

    Check prices, adjustments, bar intervals, observation times, smoothing, initial seed, earlier history and rounding. Matching the ticker and the number 14 does not establish that all of those choices match.

    A higher RSI means a larger gain share under matching calculation conventions. It does not directly measure the stock’s cumulative return or rank it against peers. Cross-sectional relative strength requires a separate comparative return measure.

    Sources

    1. Fidelity: Relative Strength Index (RSI). Definition, traditional levels, persistent extremes and divergence.
    2. Fidelity: Trading with momentum, transcript, pages 8–9. Initial averaging, Wilder’s updates and the role of calculation settings.
    3. Fidelity: Chande Momentum Oscillator. Gain/loss-sum definition, scale and period conventions.
    4. MetaQuotes: MetaTrader 5 Help, DeMarker. Component and simple-average equations; this article explicitly uses the unmultiplied 0–1 formula.
    5. Definedge Securities: Disparity Index. Percentage difference using the chosen moving average as denominator.
    6. Charles Schwab / thinkorswim: DisparityIndex. A documented exponential-moving-average implementation.
    7. MarketXLS: Relative Strength Index (RSI) function documentation. Documented retrieval parameters and data qualifications; not verification of identical initialization or plan access.

    Sources accessed November 14, 2026. All worked prices and indicator results are hypothetical and independently calculated.

    Disclaimer

    This article is for educational and informational purposes only and does not constitute investment advice. Hypothetical indicator readings are not recommendations to buy or sell securities. Technical signals can fail, and actual results depend on market conditions, data, execution, costs and risk decisions. Consult a qualified financial professional about your circumstances.