Part 2: The indicators one by one · Chapter 9 of 14 · 8 min

Divergences: a written rule and few cases

What an RSI divergence is written as a rule (a higher high with a lower RSI), how many there are over ten months (21 bearish and 11 bullish), what happened afterwards and a system with cost: 32 trades and −390 dollars per ounce, with chance doing at least as well in 78% of trials.

The rule

A divergence is a disagreement between price and an indicator. In the book the rule is written like this, for the 14-period RSI on H4.

With hindsight a divergence can always be found: you only have to choose the swings. The written rule fixes which swings count and when the signal exists.

How many there are

In the ten months there are 63 pairs of consecutive highs with the second one higher: 21 (33%) with a lower RSI (bearish divergence) and 42 with a higher RSI (confirmation). For lows, 61 pairs with the second one lower: 11 (18%) with a higher RSI (bullish divergence) and 50 with a lower RSI (confirmation). In ten months, 32 divergences in all: three a month.

Two real cases

Figure 1 shows a bearish divergence that works, on H4, from 20 to 29 August 2026. The first high is $4,635.77 (21 August at 16:00, RSI 77.6) and the second 4,700.63 (25 August at 00:00, RSI 60.5). The signal arrives on 25 August at 12:00 with a close at 4,654.96. Ten candles later price is at 4,603.73: the short earns 50.93 dollars per ounce (the 0.30 cost included).

Figure 1. A bearish divergence that works: the high rises from 4,635.77 (RSI 77.6) to 4,700.63 (RSI 60.5), the sell at 4,654.96 earns 50.93 dollars per ounce after ten candles (4,603.73).Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 20 Aug 2026 00:00 UTC to 29 Aug 2026 04:00 UTC.
Read the chart in words

H4 gold chart from 20 to 29 August 2026 and, below, the 14-period RSI with the 30, 50 and 70 levels. A dashed pink line joins the two highs, with the RSI labels, and two labels mark the signal and the candle ten periods later.

WhenWhatPrice
21 Aug 2026 16:00 UTChigh, RSI 77.64,635.77
25 Aug 2026 00:00 UTChigher high, RSI 60.54,700.63
25 Aug 2026 12:00 UTCsignal: sell4,607.14
27 Aug 2026 04:00 UTC10 candles later4,596.90

Times in UTC.

Figure 2 shows a divergence that does not work, on H4, from 14 to 23 August. The first high is $4,390.00 (15 August at 20:00, RSI 53.5) and the second 4,441.09 (18 August at 00:00, RSI 51.0). The signal arrives on 18 August at 12:00 with a close at 4,370.65. Ten candles later price is at 4,495.95: the short loses 125.60 dollars per ounce. By the rule it is a divergence, even though the RSI falls by only 2.5 points.

Figure 2. A bearish divergence that does not work: the high rises from 4,390.00 (RSI 53.5) to 4,441.09 (RSI 51.0), the sell at 4,370.65 loses 125.60 dollars per ounce after ten candles (4,495.95).Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 14 Aug 2026 00:00 UTC to 23 Aug 2026 04:00 UTC.
Read the chart in words

H4 gold chart from 14 to 23 August 2026 and, below, the 14-period RSI with the 30, 50 and 70 levels. A dashed pink line joins the two highs, with the RSI labels, and two labels mark the signal and the candle ten periods later.

WhenWhatPrice
15 Aug 2026 20:00 UTChigh, RSI 53.54,390.00
18 Aug 2026 00:00 UTChigher high, RSI 51.04,441.09
18 Aug 2026 12:00 UTCsignal: sell4,357.20
20 Aug 2026 04:00 UTC10 candles later4,482.84

Times in UTC.

What the data says

The benchmark with any candle: up carries on 36%, down carries on 40%. Outcomes after the signal, in the direction of the divergence (bearish = down, bullish = up):

SignalSignalsCarries onReversesMean move over 10 candles (ATR)
Bearish divergence, read "down"2152% ±2138%+0.19 ± 0.65
Higher high with higher RSI (confirmation), read "down"4132% ±1439%−0.89 ± 0.61
Bullish divergence, read "up"1127% ±2627%−1.38 ± 0.87
Lower low with lower RSI (confirmation), read "up"5042% ±1432%+0.49 ± 0.49
Strong bearish divergence (RSI lower by 5 points or more), "down"1429% ±2457%−0.13 ± 0.53
Strong bullish divergence, "up"729% ±3329%−0.72 ± 0.91

With 7-50 cases, the errors are between 14 and 33 points. The bearish divergence went down in 52% of cases (benchmark 40%), but with an error of 21 points and a mean move of +0.19 ATR (error 0.65); the bullish one went up in 27% (11 cases). The "strong" divergences did worse than the others. The confirmation, the case with no disagreement, had 32% (down) and 42% (up): neither better nor worse than the benchmark.

With the MACD histogram in place of the RSI: 27 bearish divergences (41% down, ±19) and 21 bullish (52% up, ±21).

A divergence system

You enter at the confirmation (3 candles after the second swing), exit after 10 H4 candles, a cost of 0.30 dollars per ounce: 32 trades, 13 winning (41%), mean −12.19 ± 14.99 dollars per ounce, total −390. Random trades of the same number (random direction, 2,000 trials) have a median total of −4 dollars (5th percentile −833, 95th +836): chance makes at least −390 dollars in 78% of trials.

A four-step routine

  1. Write the rule: which swings (side, distance), which indicator and when the signal exists.
  2. Count all the divergences in the history, including those that did not work.
  3. Compare with the case with no disagreement (the confirmation) and with random trades.
  4. Do not trust a rule that gives three signals a month on one instrument: you need years of data.

Exercises

Chapter exercises

1. On the chart the letters sit on four candles, with the RSI below. Which letter is the second high, higher than the first but with a lower RSI?

Chart exercise. Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 20 Aug 2026 00:00 UTC to 29 Aug 2026 04:00 UTC.
Read the chart in words
WhenWhatPrice
25 Aug 2026 00:00 UTCPoint A4,700.63
25 Aug 2026 12:00 UTCPoint B4,661.84
21 Aug 2026 16:00 UTCPoint C4,635.77
27 Aug 2026 04:00 UTCPoint D4,596.90

Times in UTC.

2. What is a bearish divergence under the book's rule?

3. After 21 bearish divergences price went down 52% of the time (±21), against 40% for any candle. What do you conclude?

4. The divergence system gave −390 dollars per ounce over 32 trades, and random trades do at least as well in 78% of trials. What do you conclude?

Have a question or want to share your exercise?

Post in the community, or join the free signals room on Telegram.

Educational content, not financial advice. Trading involves risk.