Part 4: Context, system and reference · Chapter 13 of 14 · 10 min
A system with candles, on trial
A system written on the result of chapter 10 (reversal formations in the direction of the 50-candle trend, one position at a time) gives +11.9 R over 451 trades, but all the gain is in the first half (+14.7 R against −2.7 R) and the rule chosen on the first half gives 0.00 R on the second. Choosing the best variant per type does not beat chance.
The system
The rules are written on the result of chapter 10 and use only information available at the time of entry.
- Formations: the 22 reversal ones of chapters 3-8.
- Filter: the formation must go in the direction of the larger trend (close above that of 50 candles earlier for a rise, below for a fall).
- Entry: at the close of the last candle (the variant with confirmation is tried separately).
- Stop: beyond the extreme of the formation plus 0.25 ATR (risk between 0.5 and 4 ATR).
- Target: equal to the risk (1 R).
- Time exit: after 24 candles.
- One position at a time: signals with a position open are skipped.
- Cost: 0.30 $ per ounce.
Data: H1 gold, December 2025 - October 2026. Attention: the filter comes from chapter 10, so the test period is the same in which it was found: the real test is the second half.
The result
The signals are 824 out of 1,876 reversal formations; with entry at the close and a target of 1 R the valid trades are 775 at +0.03 ± 0.07 R. The random series with the same rules (40 series shuffled in blocks of 6 candles) give −0.06 R (from −0.10 to +0.05): gold sits inside the range of chance. With one position at a time the trades become 451: sum +11.9 R, mean R +0.03, 53% positive, maximum drawdown 15.5 R (the largest fall from the previous peak of the curve), longest losing streak 8. If each trade risked 1% of capital, 15.5 R would be 15.5% of capital.
The random series with one position at a time give a median sum of −26.1 R (from −48.5 to +24.4), 432 trades, maximum drawdown 36.5 R (from 16.3 to 57.4) and a losing streak of 8. The median is far below zero because cost and stop weigh on many trades; gold sits above the median but inside the interval (+11.9 against a 95th percentile of +24.4).
| Variant (one position at a time) | Trades | Sum R | Mean R | Max drawdown | Random: sum (5th/95th) |
|---|---|---|---|---|---|
| Entry at the close, ½ R | 501 | −8.8 | −0.02 | 14.0 | −32.6 (−55.5 / −2.5) |
| Entry at the close, 1 R | 451 | +11.9 | +0.03 | 15.5 | −26.1 (−48.5 / +24.4) |
| Entry at the close, 2 R | 396 | +30.7 | +0.08 | 23.3 | −9.5 (−45.3 / +31.5) |
| With confirmation, ½ R | 190 | −10.8 | −0.06 | 18.2 | −11.9 (−23.5 / +0.5) |
| With confirmation, 1 R | 177 | −24.9 | −0.14 | 32.5 | −10.2 (−23.9 / +8.2) |
| With confirmation, 2 R | 163 | −15.6 | −0.10 | 28.6 | −3.8 (−29.1 / +18.8) |
The three variants with entry at the close are above the median of chance and inside the interval; the three with confirmation are below zero (the confirmation discards two signals out of three and worsens). The variants are not six independent tests: they share the same signals. Nearly half of the base trades are tweezers (194 out of 451, the most frequent type: 101 tops and 93 bottoms), then bullish (40) and bearish (34) engulfings and dojis (27).
The two halves and the test
The curve of the cumulative R (figure 3) rises to +27.4 R after 418 trades (15 September 2026) and ends at +11.9 R. By half of the period: the first gives +14.7 R (224 trades), the second −2.7 R (227 trades). No trade is off the scale: the best is worth 1.00 R, as in the random series; without it +10.9 R.
The rule of the filter can be tested honestly: choose it on the first half and measure it on the second. In the first half the formations with the 50-candle trend give +0.07 R (365 trades) and those against it −0.23 R (486). In the second half, measured with the rule already written, with the trend +0.00 ± 0.09 R (410 trades) and against −0.05 ± 0.09 R (483). The advantage of the first half does not repeat: a difference of 0.05 R remains, inside the error.
Choosing the best variant
With six variants the temptation is to choose the best for each type. For the types with at least 10 trades the best variant chosen with hindsight over the whole period gives +0.03 R (1,603 trades); the same procedure on the random series gives +0.05 R (from +0.01 to +0.08). Choosing on the first half and measuring on the second, the 829 trades give −0.05 R, and the random series −0.05 R (from −0.16 to +0.04). The procedure gives nothing more than chance.
Two trades and the curve
Figure 1 shows a winning trade: a bullish harami of 22 February 2026 (15:00: 5,131.64 → 5,123.00 $, high 5,136.00 $; 16:00: 5,123.00 → 5,127.31 $), in the direction of the larger trend (the close sits above that of 50 candles earlier, the dashed line). Entry at 5,127.31 $ (ATR 5.05 $), stop at 5,116.53 $ (10.78 $, 2.1 ATR), target at 5,138.09 $ reached at 22:00: +0.97 R. Figure 2 shows a losing trade: a dark cloud of 18 January at 12:00 and 13:00 (4,603.01 → 4,605.77 $ and 4,606.06 → 4,603.88 $), entry at 4,603.88 $ downward, stop at 4,607.12 $ (3.24 $, 1.3 ATR), target at 4,600.64 $; stop hit at 17:00: −1.09 R.
Read the chart in words
H1 gold chart from 20 to 23 February 2026 with the two numbered candles, the close of 50 candles earlier and the lines of entry, stop and target.
| When | What | Price |
|---|---|---|
| 20 Feb 2026 14:00 UTC | close 50 candles earlier | 5,057.71 |
| 22 Feb 2026 15:00 UTC | Point 1 | 5,117.79 |
| 22 Feb 2026 16:00 UTC | Point 2 | 5,122.99 |
| 22 Feb 2026 16:00 UTC | Entry | 5,127.31 |
| 22 Feb 2026 16:00 UTC | Stop | 5,116.53 |
| 22 Feb 2026 16:00 UTC | Target | 5,138.09 |
| 22 Feb 2026 22:00 UTC | target reached | 5,148.87 |
Times in UTC.
Read the chart in words
H1 gold chart of 16-19 January 2026 with the two numbered candles, the close of 50 candles earlier and the lines of entry, stop and target.
| When | What | Price |
|---|---|---|
| 16 Jan 2026 11:00 UTC | close 50 candles earlier | 4,614.19 |
| 18 Jan 2026 12:00 UTC | Point 1 | 4,606.50 |
| 18 Jan 2026 13:00 UTC | Point 2 | 4,606.42 |
| 18 Jan 2026 13:00 UTC | Entry | 4,603.88 |
| 18 Jan 2026 13:00 UTC | Stop | 4,607.12 |
| 18 Jan 2026 13:00 UTC | Target | 4,600.64 |
| 18 Jan 2026 17:00 UTC | stop hit | 4,610.83 |
Times in UTC.
Read the chart in words
Daily gold chart from December 2025 to October 2026 with, below, the curve of the cumulative R of the 451 trades.
What the data say
The system cannot be told apart from chance. The result is positive but small (+0.03 R per trade, error ±0.07), the filter was found on the same data (chapter 10), all the gain sits in the first half, the rule chosen on the first half gives zero on the second and the choice of the best variant is level with chance. More data are needed: with a standard deviation of 1 R per trade an error of ±0.10 R requires about 384 trades, and the book has 451 with a single instrument. The right way to go on is a forward test: same rules, no changes, at least 100 trades on a demo account or with minimal risk.
Procedure in five steps
- Write all the rules before looking at the results and use only what is known in real time.
- Measure sum, mean R, maximum drawdown and longest losing streak, with the cost.
- Compare with a random series that has the same rules.
- Choose the rule on one part of the data and test it on the other.
- Go on with a forward test before risking money.
Exercises
Chapter exercises
1. On the chart the letters A-D are on four candles. Which letter is the entry candle of the system (the second candle of the harami, 22 February at 16:00)?
Read the chart in words
| When | What | Price |
|---|---|---|
| 20 Feb 2026 14:00 UTC | close 50 candles earlier | 5,057.71 |
| 22 Feb 2026 09:00 UTC | Point A | 5,112.24 |
| 22 Feb 2026 15:00 UTC | Point B | 5,117.79 |
| 22 Feb 2026 16:00 UTC | Point C | 5,122.99 |
| 22 Feb 2026 22:00 UTC | Point D | 5,148.87 |
Times in UTC.
2. A system has a curve that rises to +27.4 R and ends at +11.9 R. How much is the drop from the peak to the end?
3. The rule chosen on the first half (with +0.07 R, against −0.23 R) gives +0.00 R with and −0.05 R against on the second half. What does it mean?
4. Choosing with hindsight the best variant per type gives +0.03 R and the same procedure on random series +0.05 R. 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.