Part 1: The Fibonacci ratios · Chapter 1 of 14 · 9 min

What Fibonacci is and how to test it

The sequence, the ratios 0.382, 0.5, 0.618 and 0.786, how to draw a retracement and the method the book uses to test it: out of 211 gold retracements, 41 end within ±0.03 of the four levels (19.4%), slightly more than ratios scattered at random would give.

The sequence and its ratios

The Fibonacci sequence starts with 1, 1 and adds the last two numbers each time: 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89. The ratio between a number and the next one approaches a fixed value: 0.500, 0.667, 0.600, 0.625, 0.615, 0.619, 0.618 and then 0.618 at every step. It is the golden ratio, about 0.618 (and its inverse 1.618).

The levels used on charts come from there:

LevelWhere it comes from
61.8%21 ÷ 34 = 0.6176, the ratio between two consecutive numbers
38.2%13 ÷ 34 = 0.3824, a number divided by the one two places later
23.6%8 ÷ 34 = 0.2353, a number divided by the one three places later
78.6%square root of 0.618 = 0.7861
50%not from the sequence: it is the half
127.2% and 161.8%extensions: the root of 1.618 and the ratio 34 ÷ 21 = 1.619

Nobody derived these numbers from markets: they were chosen for the properties of the sequence. That price respects them is a hypothesis, and the book treats it that way: it is turned into a rule and measured.

How to draw a retracement

You take a price leg, a stretch from a swing A to a swing B. The retracement level r is the price at distance r × (B − A) from B, in the direction of A. A down leg from 4,731.55 $ (A) to 4,515.12 $ (B) is 216.43 $ long, and the 61.8% level sits at 4,515.12 + 0.618 × 216.43 = 4,648.87 $.

The retracement ratio of a correction is computed at the end: if after B price stops at C, r = |B − C| ÷ |A − B|. With r = 1 the correction has taken back the whole leg; with r above 1 it went beyond A.

The book's method

The swings are those of book 1: highs and lows with the five-candle rule (a swing exists only two candles after its extreme), alternating, with a size filter: a leg counts only if it is at least 2 ATR long. On H4, from 15 December 2025 to 6 October 2026 (1,772 gold candles, XAUUSDT perpetual future) this gives 213 swings, so 211 retracements A → B → C.

The benchmark matters as much. A level is "hit" if the retracement ratio falls within ±0.03 of one of the four levels (0.382, 0.5, 0.618, 0.786). The number of hits is compared with what fake levels would get: random sets of four levels, and the same four levels all shifted together.

The first result

Half of the retracements are not retracements: 107 out of 211 go beyond 100%, that is, they take back the whole leg and continue (the median is 1.01). Up to 100% there are 104 (49%). By ratio band:

Ratio band rRetracements
0 - 0.2369
0.236 - 0.3827
0.382 - 0.513
0.5 - 0.61816
0.618 - 0.78636
0.786 - 123
1 - 1.27236
1.272 - 1.61825
above 1.61846

Within ±0.03 of the four levels fall 41 retracements out of 211: 19.4%, with a margin of ±5.3 points. With ratios spread uniformly from 0 to 1.2 the four bands 0.06 wide would collect 20%. By level: 3 near 0.382, 6 near 0.5, 18 near 0.618, 14 near 0.786. The result comes almost entirely from 0.618 and 0.786, which sit in the band where ratios are densest.

Figure 1 shows a case in favor of Fibonacci: the down leg from the high of 27 April 2026 at 04:00 (4,731.55 $) to the low of 29 April at 12:00 (4,515.12 $), then the retracement stops on 30 April at 12:00 at 4,648.62 $: ratio 0.617, 25 cents from the 61.8% level (4,648.87 $). Figure 2 shows one of the 107: from the high of 10 April at 12:00 (4,794.24 $) to the low of 12 April at 20:00 (4,647.18 $), then price climbs to 4,873.03 $ on 15 April: ratio 1.536, beyond the starting high and beyond every level.

Figure 1. A down leg (A 4,731.55, B 4,515.12) and the retracement that stops at 4,648.62, ratio 0.617, 25 cents from the 61.8% level.Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 26 Apr 2026 08:00 UTC to 2 May 2026 20:00 UTC.
Read the chart in words

H4 gold chart from 26 April to 2 May 2026 with the 38.2%, 50%, 61.8% and 78.6% levels drawn on the leg, points A and B marked with a letter and a label at C.

WhenWhatPrice
27 Apr 2026 04:00 UTCPoint A4,731.55
29 Apr 2026 12:00 UTCPoint B4,515.12
30 Apr 2026 12:00 UTCC: retracement 61.7%4,648.62

Times in UTC.

Figure 2. A down leg (A 4,794.24, B 4,647.18) followed by a climb to 4,873.03, ratio 1.536: beyond the starting high and beyond every level.Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 9 Apr 2026 20:00 UTC to 17 Apr 2026 08:00 UTC.
Read the chart in words

H4 gold chart from 9 to 17 April 2026 with the 38.2%, 50%, 61.8% and 78.6% levels drawn on the leg, points A and B marked with a letter and a label at C.

WhenWhatPrice
10 Apr 2026 12:00 UTCPoint A4,794.24
12 Apr 2026 20:00 UTCPoint B4,647.18
15 Apr 2026 00:00 UTCC: retracement 153.6%4,873.03

Times in UTC.

The comparison with fake levels

The number 41 has to be read against what random levels would get.

Comparison (band ±0.03)Hits
Fibonacci (0.382, 0.5, 0.618, 0.786)41 out of 211
5,000 sets of 4 random levels between 0.2 and 0.9 (minimum distance 0.08)median 30; 5th percentile 22, 95th percentile 40
The same four levels shifted together by ±0.06 ... ±0.12from 24 to 42, median 30.5
Small shifts (+0.01, +0.02, +0.03, +0.04, +0.05)38, 39, 40, 35, 39

Fibonacci sits above the 95th percentile of random levels (5% of the sets get at least 41) and above 13 of the 14 shifted sets. But levels shifted by +0.03 get 40, one fewer, and those shifted by +0.05 get 39: the number of hits has no peak on the right levels, it stays high over a wider area, the one between 0.6 and 0.8 where price retraces more often. Changing the band changes the verdict: with ±0.02 random levels get at least 27 hits in 10% of the sets, with ±0.05 at least 56 in 29%; with legs of at least 3 ATR (113 retracements) Fibonacci gets 22 hits (19.5%) and 22% of random sets do as well.

An honest reading: there is a weak indication that price stops more often between 0.6 and 0.8 than elsewhere, but the Fibonacci levels cannot be told apart from nearby levels, and the result does not survive every choice of band and size.

Procedure in four steps

  1. Write the swing rule before looking at ratios (here: five-candle fractals, legs of at least 2 ATR).
  2. Count all the retracements, not only those that end on a level.
  3. Compute the ratio r = |B − C| ÷ |A − B| and count the hits within a declared band.
  4. Compare with fake levels (random sets and shifted levels) and try other bands and sizes.

Exercises

Chapter exercises

1. On the chart the letters sit on four points: the leg starts at A, ends at B, then price retraces. Which letter marks the end of the retracement?

Chart exercise. Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 12 Apr 2026 08:00 UTC to 19 Apr 2026 12:00 UTC.
Read the chart in words
WhenWhatPrice
12 Apr 2026 20:00 UTCPoint A4,647.18
15 Apr 2026 00:00 UTCPoint B4,873.03
17 Apr 2026 00:00 UTCPoint C4,769.86
18 Apr 2026 08:00 UTCPoint D4,831.08

Times in UTC.

2. An up leg goes from 1,000 to 1,200 $ (hypothetical numbers). Where is the 61.8% level?

3. Out of 211 retracements, 41 fall within ±0.03 of the four levels (19.4%), while with ratios spread uniformly you would expect 20%. What does it mean?

4. Shifting all four levels by +0.03 gives 40 hits against Fibonacci's 41. What does it say?

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.