Part 4: Context, system and reference · Chapter 10 of 14 · 9 min
Context
Context is the part the theory says is decisive: trend, level, volatility, time. Over 1,749 trades of 22 reversal formations, eight contexts out of nine cannot be told from chance; the only one that can is the direction of the larger trend: +0.03 ± 0.07 R with it against −0.14 ± 0.06 R against it (p = 0.001), but no context lifts the R above zero.
The method
The chapter puts together the 22 reversal formations of chapters 3-8 (doji, dragonfly, gravestone, spinning top, hammer and the like, engulfings, harami, piercing, dark cloud, tweezers, stars, three inside, three outside): 1,876 formations on H1 gold, 1,749 valid trades, mean R −0.06 ± 0.05 (806 targets, 852 stops). Price goes in the expected direction in 39% of cases (control 37%).
For each context the trades are split in two groups and the mean R is compared. The permutation test asks: if I shuffle at random the labels "context yes" and "context no" (2,000 times), how often do I get a difference as large as the true one? That frequency is the p-value. Nine contexts are tried: with a threshold of 0.05, 0.5 are expected by chance alone.
The nine contexts
| Context | Trades | Mean R | The others | Mean R | p |
|---|---|---|---|---|---|
| At a level (extreme within 0.5 ATR of a swing on the same side) | 741 | −0.02 ± 0.07 | 1,008 | −0.09 ± 0.06 | 0.12 |
| After an opposite trend (≥ 1.5 ATR) | 620 | −0.12 ± 0.08 | 1,129 | −0.03 ± 0.06 | 0.07 |
| High volatility (ATR > 1.25 times the mean) | 706 | −0.02 ± 0.07 | 1,043 | −0.09 ± 0.06 | 0.16 |
| Low volatility (ATR < 0.8 times the mean) | 579 | −0.13 ± 0.08 | 1,170 | −0.03 ± 0.06 | 0.05 |
| London band (08-12 UTC) | 382 | −0.05 ± 0.10 | 1,367 | −0.07 ± 0.05 | 0.78 |
| New York band (13-20 UTC) | 648 | −0.02 ± 0.07 | 1,101 | −0.08 ± 0.06 | 0.21 |
| Asia band (00-07 UTC) | 579 | −0.07 ± 0.08 | 1,170 | −0.06 ± 0.06 | 0.84 |
| Weekend (Saturday and Sunday UTC) | 495 | −0.13 ± 0.09 | 1,254 | −0.03 ± 0.05 | 0.05 |
| In the direction of the 50-candle trend | 775 | +0.03 ± 0.07 | 974 | −0.14 ± 0.06 | 0.001 |
Eight contexts out of nine have a p above 0.05 or just at the limit (0.05-0.07): for example the opposite trend right before, which the theory wants as a condition, gives the worst result (−0.12 R). Low volatility and the weekend (thinner hours and days) do worse, but at the limit. The only one with a very low p is the direction of the larger trend.
The larger trend
A bullish formation is "with" if the close is above that of 50 candles earlier (and symmetrically for bearish ones). With it are 775 trades at +0.03 ± 0.07 R; against it, 974 trades at −0.14 ± 0.06 R. The difference is 0.17 R with p = 0.001: it survives the correction for nine tries (0.001 × 9 = 0.009). The two halves of the period agree in sign: with +0.07 and 0.00, against −0.23 and −0.05. With other windows the difference stays but changes: 20 candles +0.00 against −0.11; 30 candles −0.01 against −0.10; 100 candles −0.03 against −0.09.
Three cautions. First: this is not about profit. The group "with" makes +0.03 ± 0.07 R, that is zero; what changes is that reversal formations against the larger trend are the worst. Second: the control, any candle with a random direction, a stop at 1.5 ATR and a target of 1 R, gives −0.02 R with the 50-candle trend and −0.07 R against: part of the difference (0.05 R) is the drift of the period, not the formation. Third: for continuation formations (marubozu, soldiers, crows, three methods, inside bar) there is no effect: −0.08 R with against −0.06 R against.
The level and the timeframe
43% of reversal formations (802 out of 1,876) have the extreme within 0.5 ATR of a previous swing on the same side; shifting the swings by 3 ATR the share falls to 34% (643): formations form a little more often where price had already stopped. At a level the R is −0.02 ± 0.07 against −0.09 ± 0.06 (p = 0.12): better, but inside the margin, and the expected direction comes in 36% against 41%.
On H4 the reversal formations are 407 (375 trades, −0.02 ± 0.10 R); on D1 77 (72 trades, −0.13 ± 0.23 R). Changing timeframe does not change the picture.
Three examples
Figure 1 shows a bullish engulfing with the larger trend: on 20 February 2026 at 00:00 (4,990.72 → 5,008.61 $, after 5,003.95 → 4,990.75 $), with the close above that of 50 candles earlier (the dashed line). Entry at 5,008.61 $, stop at 4,984.41 $ (24.20 $, 1.3 ATR), target at 5,032.81 $ reached at 08:00: +0.99 R. Figure 2 shows the same formation against the trend: on 19 April at 19:00 (4,804.18 → 4,813.60 $), with the close below that of 50 candles earlier; entry at 4,813.60 $, stop at 4,800.82 $ (12.78 $, 2.2 ATR), target at 4,826.38 $; stop hit at 20:00: −1.02 R.
Figure 3 shows an engulfing at a level: on 18 May 2026 at 19:00 (4,539.38 → 4,561.70 $), with the low (4,539.01 $) a short distance from the swing low of 16 May at 21:00 (4,532.58 $). Entry at 4,561.70 $, stop at 4,532.73 $ (28.97 $, 1.4 ATR), target at 4,590.67 $ reached at 23:00: +0.99 R. Two win and one loses: they say nothing on their own.
Read the chart in words
H1 gold chart from 17 to 20 February 2026 with the two numbered candles, the close of 50 candles earlier and the lines of entry, stop and target.
| When | What | Price |
|---|---|---|
| 19 Feb 2026 23:00 UTC | Point 1 | 4,989.24 |
| 20 Feb 2026 00:00 UTC | Point 2 | 4,989.51 |
| 20 Feb 2026 00:00 UTC | Entry | 5,008.61 |
| 20 Feb 2026 00:00 UTC | Stop | 4,984.41 |
| 20 Feb 2026 00:00 UTC | Target | 5,032.81 |
| 20 Feb 2026 08:00 UTC | target reached | 5,033.49 |
| 17 Feb 2026 22:00 UTC | close 50 candles earlier | 4,889.23 |
Times in UTC.
Read the chart in words
H1 gold chart from 17 to 20 April 2026 with the two numbered candles, the close of 50 candles earlier and the lines of entry, stop and target.
| When | What | Price |
|---|---|---|
| 19 Apr 2026 18:00 UTC | Point 1 | 4,804.18 |
| 19 Apr 2026 19:00 UTC | Point 2 | 4,802.30 |
| 19 Apr 2026 19:00 UTC | Entry | 4,813.60 |
| 19 Apr 2026 19:00 UTC | Stop | 4,800.82 |
| 19 Apr 2026 19:00 UTC | Target | 4,826.38 |
| 19 Apr 2026 20:00 UTC | stop hit | 4,785.00 |
| 17 Apr 2026 17:00 UTC | close 50 candles earlier | 4,861.56 |
Times in UTC.
Read the chart in words
H1 gold chart from 16 to 19 May 2026 with the two numbered candles, the previous swing and the lines of entry, stop and target.
| When | What | Price |
|---|---|---|
| 18 May 2026 18:00 UTC | Point 1 | 4,537.91 |
| 18 May 2026 19:00 UTC | Point 2 | 4,539.01 |
| 18 May 2026 19:00 UTC | Entry | 4,561.70 |
| 18 May 2026 19:00 UTC | Stop | 4,532.73 |
| 18 May 2026 19:00 UTC | Target | 4,590.67 |
| 18 May 2026 23:00 UTC | target reached | 4,590.94 |
| 16 May 2026 21:00 UTC | previous swing | 4,532.58 |
Times in UTC.
Procedure in four steps
- List the contexts to try before looking at the results.
- Compare each context with a permutation test and correct for the number of tries.
- Always look at the R in absolute value, not only at the difference between two groups.
- Compare with a control (any candle with a random direction) to separate the drift from the formation.
Exercises
Chapter exercises
1. On the chart the dashed line is the swing low of 16 May 2026 and the letters A-D are on four candles. Which letter is the swing low that acts as the level?
Read the chart in words
| When | What | Price |
|---|---|---|
| 16 May 2026 21:00 UTC | Point A | 4,532.58 |
| 17 May 2026 13:00 UTC | Point B | 4,542.71 |
| 18 May 2026 19:00 UTC | Point C | 4,567.90 |
| 18 May 2026 23:00 UTC | Point D | 4,590.94 |
Times in UTC.
2. Nine contexts are tried with a threshold of p = 0.05. How many are expected below the threshold by chance alone?
3. The formations with the 50-candle trend give +0.03 ± 0.07 R and those against it −0.14 ± 0.06 R. What do you conclude?
4. The control (any candle with a random direction) gives −0.02 R with the trend and −0.07 R against. What is it for?
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.