RSI compares the relative magnitudes of smoothed gains and losses in one selected price series. A reading above 70 or below 30 is an observation under a chosen threshold. Check the settings, data and bar status before turning it into a test rule. The worked examples below separate a level, a crossing and an intrabar alert.

RSI75 does not mean three rises out of four

At RSI(4) initialization, compare A=+9, −1, −1, −1 with B=+3, +3, +3, −3. Both have total gains 9 and total absolute losses 3. RSI is identical despite different counts of rising changes.

Scroll sideways to see the full diagram

Two change sequences give RSI75A has one large rise and three small falls; B has three smaller rises and one larger fall. Gain and loss totals match, so initialized RSI(4) equals 75.Same RSI, different changesA:+9, −1, −1, −1+9−1−1−1B:+3, +3, +3, −3+3+3+3−3Total gains 9 • total losses 3RSI(4)=75 in both casesA:1 rise in 4 • B:3 rises in 4
Bar height shows change magnitude; colour shows its sign. Rising changes are 25% of A and 75% of B; RSI is 75 for both.

Average gain is 9/4 and average loss 3/4; their ratio is 3. Thus 100−100/(1+3)=75. RSI describes the relative magnitudes of changes; it does not count candles or give the probability of the next move.

Assumptions: Arbitrary prices: A=100; 109; 108; 107; 106; B=100; 103; 106; 109; 106. The first four changes and arithmetic-mean initialization are shown, with no older history. Formula or data reference: TradingView: RSI formula.

What does RSI actually measure?

RSI (Relative Strength Index) was introduced by J. Welles Wilder in his 1978 book, as described in TradingView’s documentation. It is a momentum oscillator on a 0–100 scale, not a comparison with a different asset or currency basket. RSI 80 does not mean 80% rising bars, an 80% chance of a rise or an 80% overvaluation.

For positive smoothed gain G and loss L, RSI = 100 − 100/(1 + G/L), equivalently 100 G/(G + L). In the Wilder-style construction, averages are initialized from the first N changes and then updated recursively; they are not simply recomputed from an isolated last-N window. N = 14 is a common default, not a best-performing setting. Zero components require the selected implementation’s special-case rule.

What it means in practice:

  • RSI 50: positive G and L are equal. This does not itself classify price as a range.
  • RSI 72: above the chosen traditional 70 zone, without specifying a price peak or its time.
  • RSI 28: below the chosen traditional 30 zone, without establishing an eventual bounce.
  • RSI 72 → 76: a rise of 4 RSI points, not 4% or 4 pips in price.
  • RSI rising from 70 towards 90: the smoothed gain-to-loss ratio is increasing. That does not necessarily mean each new upward price change is larger than the previous one.

The "RSI 70 = sell" myth

Fidelity notes that extreme readings can persist during a strong trend. It gives no fixed number of weeks or guarantee of continuation. Nor does a range make the 70-sell/30-buy shortcut universally profitable. The next calculation is a separate hypothetical loss sequence, not evidence that RSI generates these trades.

Hypothetical counter-trend loss exercise
Assumed sequence, not an observed historical resultsix losses of 50 pips each
One standard lot of USD/JPY; assumed conversion rate100,000 USD; 150 JPY per USD
JPY pip value and USD conversion1,000 JPY / 150 = 6.6667 USD per pip
300 pips at the assumed conversion rate, before costs and execution differencesabout 2,000 USD; 20% of a 10,000 USD account

A high RSI expresses the relation between smoothed gain and loss magnitudes, not the count of bullish candles. It can remain elevated during an uptrend. The reading alone does not establish overvaluation, future returns or a trade direction; inspect the formula and inputs before applying a tested decision rule.

A level, a crossing and an exit from a zone

Suppose a test permits buys only when a closed RSI is above 50. That specifies a filter; it does not prove the market is in an uptrend. Record the symbol/provider, interval, N, applied price, smoothing, available history and closed-bar convention. MetaQuotes iRSI treats several of these as separate parameters. H1 and D1 aggregate different observations.

  • Above-zone observations: assigned closed values 72,76,74 are all above 70. They are not three new crossings.
  • Zone exit: 74 → 69 exits below 70 under a strict current-value rule. It does not specify the next price move.

A crossing needs comparable prior and current observations. Define the rule as previous closed RSI ≤ 70 and current closed RSI > 70. These entirely assigned H1 values teach interpretation; they were not calculated from an undisclosed market-price history.

Assigned H1 readings; times UTC
Observation timeBar and statusRSIConfirmed crossing?
After 09:0008:00–09:00, closed and received67.0No
09:3509:00–10:00, still open70.4No under the closed-bar rule
After 10:0009:00–10:00, closed and received68.5No; prior/current below 70
After 11:0010:00–11:00, closed and received72.0Yes; prior ≤ 70, current > 70

The last two closed readings differ by 72.0 − 68.5 = 3.5 RSI points. The earlier 70.4 was 1.9 points above the same bar’s final 68.5. Do not count its 09:35 alert as a confirmed crossing in a closed-bar test. Intrabar alerts use a different information rule.

Divergence is a defined comparison

Divergence is a mismatch between price and RSI. You will meet two classic kinds:

Bullish divergence

For a regular bullish divergence, compare a lower price low with a higher RSI low using a stated point-selection and pairing rule. Preserve when those lows became recognizable. This comparison does not measure sellers’ motives or establish the probability of an upward reversal.

Bearish divergence

The regular bearish version compares a higher price high with a lower RSI high. Define the same data, pairing and confirmation-time conventions. Geometry and a trade’s net result are separate questions.

The cited educational definitions do not quantify a small edge or the profitability of H4/D1 divergence. Additional conditions must be tested as a complete rule rather than counted as automatically independent evidence.

  1. Price reaches strong support on the daily chart
  2. A reversal candle forms (hammer, bullish engulfing)
  3. RSI shows a bullish divergence
  4. Record whether all declared conditions were available at the decision time; do not treat agreement as proof of a profitable long.

Conditions based on the same prices need not provide independent evidence. Separate the intended entry, exit, size, execution and costs from the indicator observation. Evaluate the complete rule on unused data, retaining failed and no-trade cases.

Same RSI 50, different distance to exactly 70

In a separate controlled Wilder-style RSI(14) state, assume positive G = L and a next change consisting only of a rise x. Then G′ = (13 G + x)/14 and L′ = 13 L/14. Setting RSI′ = 70 gives x = [0.70 × 13(G + L) − 13 G]/0.30.

G = L = 0.0001x = 0.001733… = 17.33… pips
G = L = 0.0010x = 0.017333… = 173.33… pips

Assume one pip = 0.0001 price unit. Both initial RSI readings equal 50, yet required rises differ tenfold. For strictly above 70, x must exceed the exact boundary, not merely its rounded display. These are assigned recursion states, not a price forecast or measured frequency. A different smoothing convention needs a different equation.

Practical takeaway: RSI compares smoothed gains and losses within the same price series; it is not a comparison of two different markets.

What the reading leaves unspecified

Check these separate limitations before interpreting the number:

  1. Future price. An extreme reading does not guarantee reversal or continuation; it does not determine an entry’s loss.
  2. Bar status. A Close-based RSI can change during an open bar. A closed-bar rule can use its final value only after close and receipt.
  3. Timeframe and sampling. M5, H1 and D1 measure different aggregations. No minimum H1 interval or guaranteed superiority of D1 follows from the formula.

For whether indicators can support a profitable rule, separate definitions from tested net outcomes. Combining RSI with an average, a price level or a candle label does not establish independent confirmation or improve returns automatically.

What to do tomorrow with RSI

  1. Write the configuration. Record symbol/provider, interval, period, applied price, smoothing, history and whether bars must be closed.
  2. Define each condition. A price level such as support and resistance, a candle and divergence need their own rules and recognition times. Their agreement is not a profitability test.
  3. Retain information time. Log a transient alert separately from a final closed reading. Do not use future confirmation in an earlier simulated entry.
  4. Evaluate a complete rule. Fix entries, exits, costs and no-trade conditions before judging unused data. Neither four weeks nor an arbitrary indicator count proves an edge.