Part 3: Elliott waves · Chapter 7 of 14 · 8 min
Elliott waves: the rules and how to test them
Five waves in the direction of the trend and three of correction, with three rules that cannot be broken. On H4 gold, 14 windows of six swings out of 203 (6.9%) meet them all: fewer than a random path with the same sizes would give (median 10.6%).
The structure
Elliott's theory describes price as a sequence of waves. An impulse has five waves in the direction of the trend: three of thrust (1, 3, 5) and two of correction (2, 4). After the impulse comes a correction of three waves (A, B, C) in the opposite direction. The waves repeat at different degrees: each thrust wave divides in turn into five smaller waves, and each correction into three.
The three rules a count cannot break:
- Wave 2 does not retrace beyond the start of wave 1.
- Wave 3 is not the shortest among waves 1, 3 and 5.
- Wave 4 does not enter the territory of wave 1 (it does not fall below the end of wave 1 in an upward impulse).
Alongside the rules there are guidelines: wave 3 is often the longest, waves 2 and 4 alternate (one sharp, one flat), the lengths have Fibonacci ratios (chapter 10). The guidelines do not bind, and chapter 8 counts how many more counts you get by admitting diagonals, the exception the theory provides for rule 3.
How the book makes the count testable
A count with no rules on the starting points fits any chart. Here a wave is a stretch between two consecutive swings (the highs and lows of book 1, with legs of at least 2 ATR on H4 and confirmation two candles later). A possible impulse is a window of six consecutive swings, P0 ... P5: wave 1 goes from P0 to P1, wave 2 from P1 to P2 and so on. The window is "valid" if the three rules are met.
On the data (December 2025 - October 2026, gold, XAUUSDT perpetual future) the swings are 213 and the windows 203 (five of the 208 possible sequences do not truly alternate up and down). The windows overlap: each swing sits in six windows. Only those in which the five legs truly alternate up and down are kept.
How many windows meet the rules
| Rule | Windows that meet it | Random path (median, 5th-95th percentile) |
|---|---|---|
| 1: wave 2 not beyond the start | 100 out of 203 (49.3%) | 50.0% (46.6%-53.4%) |
| 2: wave 3 not the shortest | 141 out of 203 (69.5%) | 66.8% (64.4%-69.2%) |
| 3: wave 4 outside the territory of wave 1 | 23 out of 203 (11.3%) | 16.8% (13.5%-20.2%) |
| All three | 14 out of 203 (6.9%) | 10.6% (7.2%-13.9%) |
The "random path" is the same number of swings with the same leg sizes, in shuffled order (2,000 trials), with the up/down alternation kept. Rules 1 and 2 are met in the data as in chance. Rule 3 is the strictest: on gold only 23 windows out of 203 meet it, and fewer than the random path. The valid sequences are 14 (6 up, 8 down, all distinct): 97% of random paths produce at least as many. On the data, a complete Elliott impulse is no more frequent than it would be in a succession of moves of the same order of magnitude thrown together.
With other size filters: legs of at least 1.5 ATR, 250 windows and 19 valid (7.6%); 3 ATR, 110 windows and 11 valid (10.0%); 4 ATR, 63 windows and 3 valid (4.8%). On the daily chart (legs of at least 1 ATR) there are 47 windows and 4 valid.
Figure 1 shows a valid upward impulse: from 5,101.20 $ (P0, 24 February 2026 at 12:00) to 5,420.00 $ (P5, 2 March at 04:00). The waves measure 118, 86, 263, 121 and 145 $: wave 3 is the longest, wave 2 stays above the start (5,133.05 against 5,101.20) and wave 4 (5,275.08) stays above the end of wave 1 (5,218.80). Figure 2 shows a downward one: from 4,373.53 $ (22 September at 16:00) to 4,139.17 $ (2 October at 00:00), waves of 124, 70, 202, 110 and 88 $. Figure 3 shows a window that fails: from 4,601.50 $ (6 April at 00:00) to 4,794.24 $ (10 April at 12:00), with the end of wave 1 at 4,703.87 $ and the low of wave 4 at 4,692.37 $, 11.50 $ inside the territory of wave 1. For a manual count it would "almost" be an impulse. For the book's rules it is not: it is the rules that make the count something that can be refuted.
Read the chart in words
H4 gold chart from 24 February to 2 March 2026 with the broken line of the six swings numbered 0 to 5 and the length of each wave in dollars.
| When | What | Price |
|---|---|---|
| 24 Feb 2026 12:00 UTC | Point 0 | 5,101.20 |
| 25 Feb 2026 16:00 UTC | Point 1 | 5,218.80 |
| 26 Feb 2026 12:00 UTC | Point 2 | 5,133.05 |
| 28 Feb 2026 08:00 UTC | Point 3 | 5,396.44 |
| 1 Mar 2026 00:00 UTC | Point 4 | 5,275.08 |
| 2 Mar 2026 04:00 UTC | Point 5 | 5,420.00 |
| 25 Feb 2026 04:00 UTC | 1: 118 $ | 5,160.00 |
| 26 Feb 2026 04:00 UTC | 2: 86 $ | 5,175.93 |
| 27 Feb 2026 12:00 UTC | 3: 263 $ | 5,264.75 |
| 28 Feb 2026 16:00 UTC | 4: 121 $ | 5,335.76 |
| 1 Mar 2026 16:00 UTC | 5: 145 $ | 5,347.54 |
Times in UTC.
Read the chart in words
H4 gold chart from 22 September to 2 October 2026 with the broken line of the six swings numbered 0 to 5 and the length of each wave in dollars.
| When | What | Price |
|---|---|---|
| 22 Sept 2026 16:00 UTC | Point 0 | 4,373.53 |
| 24 Sept 2026 08:00 UTC | Point 1 | 4,249.11 |
| 25 Sept 2026 08:00 UTC | Point 2 | 4,318.68 |
| 28 Sept 2026 12:00 UTC | Point 3 | 4,116.60 |
| 30 Sept 2026 12:00 UTC | Point 4 | 4,226.82 |
| 2 Oct 2026 00:00 UTC | Point 5 | 4,139.17 |
| 23 Sept 2026 12:00 UTC | 1: 124 $ | 4,311.32 |
| 24 Sept 2026 20:00 UTC | 2: 70 $ | 4,283.90 |
| 27 Sept 2026 00:00 UTC | 3: 202 $ | 4,217.64 |
| 29 Sept 2026 12:00 UTC | 4: 110 $ | 4,171.71 |
| 1 Oct 2026 08:00 UTC | 5: 88 $ | 4,183.00 |
Times in UTC.
Read the chart in words
H4 gold chart from 5 to 11 April 2026 with the broken line of the six swings numbered 0 to 5, the line of the end of wave 1 and a label on wave 4.
| When | What | Price |
|---|---|---|
| 6 Apr 2026 00:00 UTC | Point 0 | 4,601.50 |
| 6 Apr 2026 08:00 UTC | Point 1 | 4,703.87 |
| 7 Apr 2026 12:00 UTC | Point 2 | 4,612.13 |
| 8 Apr 2026 00:00 UTC | Point 3 | 4,854.98 |
| 9 Apr 2026 00:00 UTC | Point 4 | 4,692.37 |
| 10 Apr 2026 12:00 UTC | Point 5 | 4,794.24 |
| 6 Apr 2026 08:00 UTC | end of wave 1 | 4,703.87 |
| 9 Apr 2026 00:00 UTC | wave 4 enters wave 1 territory | 4,692.37 |
Times in UTC.
Procedure in four steps
- Write the swing rule (here: five-candle fractals, legs of at least 2 ATR).
- Take all the windows of six consecutive swings and check the three rules one by one.
- Count the valid ones, those that break only one rule and the distinct ones (not overlapping).
- Compare with a random path that has the same leg sizes.
Exercises
Chapter exercises
1. On the chart the six swings of an upward impulse are marked with letters A-F (A = start of wave 1). Which letter is the end of wave 3?
Read the chart in words
| When | What | Price |
|---|---|---|
| 24 Feb 2026 12:00 UTC | Point A | 5,101.20 |
| 25 Feb 2026 16:00 UTC | Point B | 5,218.80 |
| 26 Feb 2026 12:00 UTC | Point C | 5,133.05 |
| 28 Feb 2026 08:00 UTC | Point D | 5,396.44 |
| 1 Mar 2026 00:00 UTC | Point E | 5,275.08 |
| 2 Mar 2026 04:00 UTC | Point F | 5,420.00 |
Times in UTC.
2. An upward impulse starts at 100, wave 1 reaches 150, wave 2 falls to 90 (hypothetical numbers). What does rule 1 say?
3. Out of 203 windows of six swings, 14 (6.9%) meet the three rules, and a random path with the same sizes produces a median of 10.6%. What does it mean?
4. In an upward impulse the end of wave 1 is at 150 and the low of wave 4 at 145 (hypothetical numbers). Which rule is broken?
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.