RSI Divergence: The Rule, the Four Types, and What Every Signal Did Next
Price makes a new high or low and the RSI does not. How we mark RSI divergences by a fixed rule, the regular and hidden types, and what each divergence on Bitcoin and the S&P 500 ETF was followed by.

On this page
An RSI divergence is a disagreement between price and the Relative Strength Index: price makes a higher high while the RSI makes a lower high (bearish), or price makes a lower low while the RSI makes a higher low (bullish). J. Welles Wilder, who published the RSI in 1978, read it as a sign that momentum no longer confirms the move. Below: the exact rule we use to mark divergences on the chart, the four types, and what every divergence on Bitcoin and the S&P 500 ETF did next, including the ones that failed.
RSI(14) divergences on Bitcoin, marked by rule

- Price, daily candles (hollow up, filled down)
- RSI(14) (solid line)
- Regular bullish: lower low in price, higher low in the oscillator
- Hidden bearish: lower high in price, higher high in the oscillator (dashed)
- Hidden bullish: higher low in price, lower low in the oscillator (dashed)
What an RSI divergence is
StockCharts, following Wilder, gives the two classic definitions. “A bullish divergence occurs when the underlying security makes a lower low, and RSI forms a higher low.” “A bearish divergence forms when the security records a higher high and RSI forms a lower high.” According to Wilder, divergences signal a potential reversal point “because directional momentum does not confirm price.”
Two warnings come from the same page. Divergences “tend to be more robust when they form after an overbought or oversold reading,” that is above 70 or below 30. And they “are misleading in a strong trend”: a strong uptrend can show numerous bearish divergences before a top. Both claims can be checked on real data, which is what the rest of this page does. To judge how strong a trend is, Wilder also built the average directional index, published in the same 1978 book as the RSI.
How we mark a divergence: the rule
A divergence line drawn by hand, after the fact, is easy to fit to what happened. Ours are drawn by a fixed rule, the same on every bar:
- A swing high is a bar whose high is strictly above the highs of the 5 bars before and the 5 bars after it. A swing low is the same with lows.
- Each swing high is compared with the swing high just before it, if the two are at most 60 bars apart. Same for swing lows.
- The RSI(14) is read on the two swing bars, and the four comparisons below decide the type.
The rule has a cost that hand-drawn charts hide: a swing high is only known 5 bars after it happens, since the next 5 highs must be lower. A divergence can therefore be acted on 5 bars after its second swing at the earliest. All the returns below are measured from the close of that confirmation bar, not from the swing itself.
The June 2026 bullish divergences on daily Bitcoin, up close

- Price, daily candles (hollow up, filled down)
- RSI(14) (solid line)
- Regular bullish: lower low in price, higher low in the oscillator
- Hidden bearish: lower high in price, higher high in the oscillator (dashed)
Regular and hidden divergences
Besides Wilder’s two regular divergences, the chart marks two hidden ones (dashed). StockCharts credits them to Andrew Cardwell under another name: a “positive reversal forms when RSI forges a lower low, and the security forms a higher low”, and a negative reversal is the opposite. Cardwell read divergences differently from Wilder: he “considered bearish divergences to be bull market phenomena,” more likely to form in uptrends.
The four RSI divergences

| Type | Price | RSI | Usual reading |
|---|---|---|---|
| Regular bearish | Higher high | Lower high | Uptrend losing momentum (Wilder) |
| Regular bullish | Lower low | Higher low | Downtrend losing momentum (Wilder) |
| Hidden bearish | Lower high | Higher high | Negative reversal (Cardwell): downtrend continuing |
| Hidden bullish | Higher low | Lower low | Positive reversal (Cardwell): uptrend continuing |
Every regular divergence on Bitcoin, and what followed
We ran the rule on all our daily Bitcoin bars. The rule waits 43 days for the RSI(14) to settle, so the first usable bar is April 20, 2025, and the last is October 28, 2026. The rule found 7 regular divergences and 8 hidden ones (one more hidden one is discarded: its first swing falls before April 20). Here are all 7 regular ones, with the price change from the confirmation close. For a bearish divergence, a fall is the expected direction; a short sold at the confirmation close would have returned the same percentage with the sign reversed.
| Type | First swing | Second swing | Confirmed, close | After 5 days | After 10 days | After 20 days |
|---|---|---|---|---|---|---|
| Bearish | May 2, 2025: high 97,895.68, RSI 70.10 | May 12, 2025: high 105,819.45, RSI 69.40 | May 17, 2025: 103,126.65 | +8.31% | +5.64% | +1.13% |
| Bearish | July 14, 2025: high 123,218.00, RSI 75.95 | August 14, 2025: high 124,474.00, RSI 53.25 | August 19, 2025: 112,872.94 | +0.55% | −3.98% | −0.72% |
| Bullish | September 26, 2025: low 108,620.07, RSI 38.39 | October 10, 2025: low 102,000.00, RSI 41.78 | October 15, 2025: 110,763.28 | −0.21% | +0.80% | −8.37% |
| Bullish | February 12, 2026: low 65,118.00, RSI 29.33 | February 24, 2026: low 62,510.28, RSI 30.70 | March 1, 2026: 65,776.47 | +3.55% | +6.71% | +4.78% |
| Bullish | June 5, 2026: low 59,130.91, RSI 15.62 | June 25, 2026: low 58,115.01, RSI 30.54 | June 30, 2026: 58,624.71 | +8.57% | +9.44% | +11.31% |
| Bullish | June 25, 2026: low 58,115.01, RSI 30.54 | July 1, 2026: low 57,800.19, RSI 37.21 | July 6, 2026: 64,042.02 | −0.35% | −0.33% | +2.12% |
| Bullish | October 8, 2026: low 80,393.56, RSI 47.05 | October 15, 2026: low 79,800.00, RSI 48.88 | October 20, 2026: 83,028.15 | −3.46% | not yet known | not yet known |
Ten days after each regular divergence on Bitcoin, in the expected direction

Of the 6 regular divergences with ten days of history after them, price moved in the expected direction 4 times after ten days and 4 times after twenty. The first bearish one was wrong at every horizon: Bitcoin was 8.31% higher five days after it. The October 2025 bullish one was right after ten days by 0.80%, then wrong by 8.37% after twenty. For comparison, on any day of the same period Bitcoin closed higher ten days later 275 times out of 547, about half the time.
Divergences against the base rate of the same period

The overbought and oversold filter is one of the few things the sample can test, barely. Four of the regular divergences had their first swing at 70 or above (bearish) or at 30 or below (bullish): May 2025, July 2025, February 2026 and June 2026. Three of the four went the expected way after ten and twenty days. The two others with ten days of history, in October 2025 and July 2026, were each right at one horizon and wrong at the other. Four cases cannot separate a filter from luck.
A strong trend: the S&P 500 ETF
RSI(14) divergences on the daily S&P 500 ETF, marked by rule

- Price, daily candles (hollow up, filled down)
- RSI(14) (solid line)
- Regular bearish: higher high in price, lower high in the oscillator
- Regular bullish: lower low in price, higher low in the oscillator
- Hidden bearish: lower high in price, higher high in the oscillator (dashed)
The S&P 500 ETF shows the second warning. From August 9, 2024, the first usable bar, to October 28, 2026, it rose from 519.37 to 772.31 in our dividend-adjusted data. Prices are rounded to four decimals; at that precision the highs of January 12 and 13, 2026 are equal (690.70), so neither is a swing high. Over that period the rule found 10 regular bearish divergences against 1 bullish one. Twenty sessions after confirmation, the ETF was lower after only 2 of the 10 bearish ones: February 2025 (−4.08%) and February 2026 (−0.71%). On any session of the same period it was lower twenty sessions later 157 times out of 537, about 29%. On 10 cases, 2 lower closes against a base rate of about 29% is no evidence either way; what the table does show is the warning StockCharts gives: in a strong uptrend, bearish divergences keep appearing while the price keeps rising. All 11, with the price change from the confirmation close:
| Type | Second swing | Confirmed, close | After 5 sessions | After 10 | After 20 |
|---|---|---|---|---|---|
| Bearish | October 17, 2024 | October 24, 2024: 566.17 | −1.83% | +2.83% | +2.49% |
| Bearish | February 19, 2025 | February 26, 2025: 583.08 | −1.93% | −6.00% | −4.08% |
| Bearish | June 11, 2025 | June 18, 2025: 587.69 | +2.72% | +4.98% | +5.36% |
| Bearish | July 31, 2025 | August 7, 2025: 623.77 | +2.01% | +0.52% | +2.37% |
| Bearish | September 23, 2025 | September 30, 2025: 659.07 | +0.44% | −0.59% | +3.13% |
| Bearish | October 10, 2025 | October 17, 2025: 657.30 | +1.94% | +2.66% | +1.13% |
| Bearish | December 26, 2025 | January 5, 2026: 682.39 | +1.08% | −1.47% | +0.26% |
| Bearish | January 28, 2026 | February 4, 2026: 680.87 | +0.84% | −0.25% | −0.71% |
| Bullish | February 17, 2026 | February 24, 2026: 682.02 | −1.02% | −1.48% | −4.71% |
| Bearish | June 2, 2026 | June 9, 2026: 733.34 | +1.80% | −0.26% | +2.25% |
| Bearish | July 15, 2026 | July 22, 2026: 745.56 | −2.40% | +2.99% | +2.90% |
The only bullish divergence failed as well: the ETF was 4.71% lower twenty sessions after it.
RSI 9, 14 or 21?
The same rule with three RSI lengths, regular divergences only. A longer RSI needs a longer warm-up, so its first usable bar comes a few days later: April 5, April 20 and May 11, 2025 on Bitcoin; July 19, August 9 and September 10, 2024 on the ETF. The rule found 8, 7 and 7 regular divergences on Bitcoin, and 11, 11 and 8 on the ETF; the chart counts those with ten or twenty bars of history after confirmation.
Share of regular divergences that went the expected way

- After 10 bars
- After 20 bars
On Bitcoin the RSI(21) looks best, on 7 cases; on the ETF no length escapes the trend. Picking the length that worked best on such samples is how a backtest overfits.
Does it work?
We have no long-run measurement of RSI divergences: our bench tests chart patterns detected by rule against random entries, not indicators (see our methodology). What the counts above show is narrow: on Bitcoin, regular divergences went the expected way 4 times out of 6, too few to tell from chance; on a rising ETF, bearish divergences mostly failed. Treat a divergence as a reason to look at the chart, then wait for price to confirm, for example a break of a trendline or of the neckline of a double top.
Related pages: the Relative Strength Index itself, divergences on other oscillators, such as the MACD and the stochastic, the stochastic RSI, and the list of technical indicators.
Frequently asked questions
What is RSI divergence?
A disagreement between price and the RSI: price makes a higher high while the RSI makes a lower high (bearish), or price makes a lower low while the RSI makes a higher low (bullish). Wilder read it as momentum no longer confirming the move.
Is RSI divergence profitable?
Not shown. On Bitcoin from April 2025 to October 2026, 4 of 6 regular divergences went the expected way after ten days, a sample too small to separate from chance; on the rising S&P 500 ETF, only 2 of 10 bearish divergences were followed by a lower close twenty sessions later.
What is the difference between regular and hidden divergence?
A regular divergence points to a reversal: price extends the trend, the RSI does not. A hidden divergence points to a continuation: price holds a higher low (or a lower high) while the RSI makes a lower low (or a higher high). StockCharts calls these Cardwell’s positive and negative reversals.
How do you confirm an RSI divergence?
First the swing itself must be confirmed: with our rule, 5 bars must pass after the second swing. Then wait for price to confirm, for example a break of a trendline. Our counts are measured from the first of these two steps only.
Which RSI setting is best for divergence?
None has been shown to be. On our data the RSI(21) did best on Bitcoin, on 7 cases, and no length did well on the S&P 500 ETF; choosing the best length on such small samples is overfitting.
