Divergence in Trading: The Four Types, RSI vs MACD vs Stochastic, and Why They Fail in Trends

A divergence is a disagreement between price and an oscillator. Price makes a higher high and the oscillator a lower high, or price makes a lower low and the oscillator a higher low: the move goes on, the momentum behind it does not. The idea is the same whatever the oscillator, RSI, MACD or stochastic. Below: the four types, the rule we use to find them, and a count of the divergences each of the three oscillators found on the S&P 500 ETF and on Bitcoin, with what followed.

The four types of divergence

The two regular divergences are the textbook ones. StockCharts gives them for the MACD: “A bullish divergence forms when a security records a lower low and the MACD forms a higher low”, and “a bearish divergence forms when a security records a higher high and the MACD line forms a lower high.” Its stochastic and RSI pages give the same definitions for those oscillators. A regular divergence is read as a possible reversal.

The two hidden divergences are read the other way, as a continuation. StockCharts describes them on its RSI page as Andrew Cardwell’s positive and negative reversals: a “positive reversal forms when RSI forges a lower low, and the security forms a higher low,” and a negative reversal is the opposite.

How we find a divergence

A divergence drawn by hand after the fact is easy to find and impossible to test. The MACD chart under the introduction, and every count on this page, uses one fixed rule:

A swing point is only known once the bars after it are in, so a divergence can be acted on 5 bars after its second swing at the earliest. Every return below starts from the close of that bar. The rule reads the oscillator on the swing bars of price; it compares each swing with the one just before it, never with an older one.

RSI, MACD or stochastic: the same rule on three oscillators

The rule was run on the full daily history of the site's test bench, from January 29, 1993 for the S&P 500 ETF (dividend-adjusted prices) and from August 17, 2017 for Bitcoin, both to October 28, 2026, with the RSI(14), the MACD line (12, 26) and the slow stochastic %K (14, 3). Each oscillator needs its own warm-up, so its first usable bar differs. “Expected direction” means a lower close 20 bars after confirmation for a bearish divergence and a higher one for a bullish divergence. The last column is what chance gives for the same mix of cases: the share of all days, from the first usable bar, with a lower close 20 bars later, applied to the bearish cases, and with a higher close, applied to the bullish ones.

The last Bitcoin RSI divergence is too recent to have 20 bars after it, so that line counts 51 cases out of 52.

On the ETF every line is below chance. The ETF closed higher 20 sessions later on 65% of all days since 1993, so most bearish divergences were followed by a rise: with the RSI, 33 of 107 bearish cases were followed by a lower close (31%), against 34.5% of all days. On Bitcoin, the RSI and the stochastic sit at chance. The MACD is the only line above it, 30 of 45 against about 50%: one line out of six, on 45 cases, is not a result to rely on.

When oscillators agree

In the 600-bar window ending October 28, 2026, from June 7, 2024 onward, and from November 2024, when all three oscillators are usable, five ETF highs were flagged as bearish divergences by the RSI, the MACD and the stochastic at once, each against the same earlier high. Twenty sessions after confirmation, the ETF was higher after four of them.

Agreement between oscillators is not independent evidence: all three are built from the same prices.

Why divergences fail in a strong trend

StockCharts says it plainly on its MACD page: “Bearish divergences are commonplace in a strong uptrend, while bullish divergences occur often in a strong downtrend.” The ETF shows it. Over its full history since 1993, the MACD’s 108 bearish divergences were followed by a lower close twenty sessions later 28 times (26%), against 34.5% of all days. In 2025, the MACD found four bearish divergences confirmed between June 18 and October 17, 2025, and the ETF closed higher twenty sessions after each of them; over those four months it went from 587.69 to 657.30, an 11.8% rise.

The MACD chart under the introduction covers the last 250 sessions and shows the last five of these divergences, from December 2025 to September 2026. Twenty sessions after confirmation, the ETF was lower once, after the January 28, 2026 high (−0.71%), and higher after the four others (+0.26%, +2.25%, +2.90%, +0.71%). Hidden bullish divergences, which point the other way, went with the trend: with the MACD, the ETF was higher twenty sessions later after 5 of 6, against 319 of 475 sessions (67%) for any day of the period from November 6, 2024, too few cases to call it better than the base rate.

Does it work?

Divergences are not part of our bench, which tests chart patterns detected by rule against random entries, not indicators (see our methodology), so we ran the rule separately. On the full daily history of our two markets, regular divergences did worse than chance on the ETF, which mostly rose, and about as well as chance on Bitcoin, with one exception, the MACD on Bitcoin at 30 of 45 against about 50%, which a single line out of six does not establish. A divergence says momentum has slowed; it does not say when or whether price will turn. Price confirmation, such as the break of a trendline, a lower high in Dow theory terms or the neckline of a head and shoulders, is a separate signal.

Related pages: RSI divergence with every Bitcoin signal, the MACD, the stochastic oscillator, the RSI, and the list of technical indicators.