Part 3: Using them together · Chapter 12 of 14 · 10 min

An indicator system under test

A complete system written before looking at results (100-period average, ADX, RSI, stop and target): in ten months it gives 6 trades, and with the rules widened 22 trades with +0.29 R on average, an error of ±0.54 R and a result that random candles in the same trend match in 45% of trials.

The rules, written beforehand

A system is a set of rules that leaves no choices: when to enter, where to put the stop, where to take profit, when to leave. The rules below were fixed before looking at results, with the standard values of the earlier chapters: no optimization.

  • Trend: close above the EMA 100 with a 14-period ADX above 20 (long only); close below the EMA 100 with ADX above 20 (short only).
  • Signal (pullback): the 14-period RSI was below 40 and the candle closes with RSI at least 40 (long); it was above 60 and closes with RSI at most 60 (short).
  • Entry: at the close of the signal candle, on H4.
  • Stop: 1.5 ATR from entry. Target: 3 ATR (2R). Exit on time: at the close of the 24th candle after entry.
  • Bookkeeping rules: if stop and target fall in the same candle the stop counts; one position at a time; cost 0.30 dollars per ounce, converted to R by dividing by the risk.

The period is December 2025 - October 2026, 1,772 H4 gold candles (XAUUSDT perpetual future).

First result: six trades

With these rules, in ten months, the system makes 6 trades (2 long and 4 short): 2 reach the target, 2 the stop, 2 exit on time. Winners are 3 out of 6 (50%), the mean is +0.40 R with an error of ±1.09 R (95% interval, that is 1.96 standard errors), the total +2.4 R.

Six trades say nothing. The standard deviation of a single trade is 1.36 R: to bring the error to ±0.10 R you would need about 700 trades, roughly a century of these signals. That +0.40 R is neither proof nor refutation.

Widening the rules is already optimizing

The step many take at this point is to make the system more "active": less extreme RSI thresholds, tighter or wider stops. Every tweak is a variant tried (chapter 11). Here are nine, with long/short RSI threshold and stop in ATR, target always 2R:

RSI thresholdsStop 1 ATRStop 1.5 ATRStop 2 ATR
35 / 654 tr., −1.01 R4 tr., −0.76 R3 tr., −0.68 R
40 / 60 (the declared rules)6 tr., +0.27 R6 tr., +0.40 R5 tr., +0.66 R
45 / 5523 tr., +0.40 R22 tr., +0.29 R21 tr., +0.33 R

Six variants out of nine are positive. The 35/65 thresholds have 3-4 trades and no winner; the 40/60 thresholds have 5-6 trades; only with 45/55 does the sample reach 21-23. The highest and lowest means (from −1.01 to +0.66 R) almost all come from samples under seven trades, and a sample that small swings like this without the system changing.

From here on the version with 45/55 thresholds and a 1.5 ATR stop is used. It was chosen after seeing that the first one gave six trades: keep that in mind when reading the numbers.

Second result: twenty-two trades

The 45/55 version makes 22 trades (10 long and 12 short): 6 at the target, 8 at the stop, 8 on time. Winners are 11 (50%), the mean is +0.29 R ± 0.54 R, the total +6.4 R. The maximum drawdown from the peak is −3.4 R and the longest losing streak is 4 trades. The error is almost twice the mean: zero sits inside the interval.

By half of the period, split on 11 May:

PeriodTradesMean per trade
First half8+0.05 ± 0.82 R
Second half14+0.43 ± 0.72 R

Figure 1 shows three winning trades from early June: three shorts with entries at 4,518.52, 4,529.69 and 4,479.68 $, all reaching the target (+1.98, +1.99 and +1.99 R) below the EMA 100, after the RSI had climbed above 55 and started falling again. Figure 2 shows the other page, at the end of August: two longs (entries at 4,626.26 and 4,614.41 $) hit by the stop a few hours apart (−1.01 R each), then a short on 4 September at 4,439.40 $ closed on time with +0.71 R. Three trades of the same system: the result depends on the sample, and the sample is small.

Figure 1. Three shorts in early June 2026, all at the target (+1.98, +1.99 and +1.99 R): entries at 4,518.52, 4,529.69 and 4,479.68.Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 29 May 2026 00:00 UTC to 8 Jun 2026 20:00 UTC.
Read the chart in words

H4 gold chart from 29 May to 8 June 2026 with the EMA 100 (blue), for each trade the dashed line of the stop and of the target and a label with the result in R; below the 14-period RSI with levels 45 and 55.

WhenWhatPrice
29 May 2026 00:00 UTCEMA 1004,544.20 → 4,450.87
1 Jun 2026 00:00 UTCshort: target +1.98 R4,518.52
2 Jun 2026 08:00 UTCshort: target +1.99 R4,529.69
4 Jun 2026 12:00 UTCshort: target +1.99 R4,479.68

Times in UTC.

Figure 2. Two longs hit by the stop (−1.01 R each: entries at 4,626.26 and 4,614.41) and one short closed on time (+0.71 R, entry at 4,439.40, closed on 8 September).Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 24 Aug 2026 00:00 UTC to 4 Sept 2026 20:00 UTC.
Read the chart in words

H4 gold chart from 24 August to 4 September 2026 with the EMA 100 (blue), the dashed lines of the stop and target of each trade and labels with the result in R; below the 14-period RSI with levels 45 and 55.

WhenWhatPrice
24 Aug 2026 00:00 UTCEMA 1004,432.46 → 4,466.35
26 Aug 2026 20:00 UTClong: stop −1.01 R4,626.26
28 Aug 2026 04:00 UTClong: stop −1.01 R4,614.41
4 Sept 2026 12:00 UTCshort: time +0.71 R4,439.40

Times in UTC.

Figure 3 shows the curve of results in R over the whole period: at mid-period, 11 May, the sum of closed trades is +1.43 R; at the end of the period +6.43 R. That is 22 trades spread over ten months.

Figure 3. Cumulative result of the 22 trades: +1.43 R on 11 May, +6.43 R at the end of the period (daily candles).Data: gold, XAUUSDT perpetual future (real candles) · daily (D1) · from 15 Dec 2025 to 5 Oct 2026.
Read the chart in words

Daily gold chart from December 2025 to October 2026, split in two halves, with below the cumulative result in R of the system's 22 trades (black line, zero marked).

WhenWhatPrice
15 Dec 2025 → 12 May 2026first half
12 May 2026 → 5 Oct 2026second half

Times in UTC.

The benchmark: what chance does with the same exits

A positive result has to be compared. Same stop, target and time rules, but random entries: 10 longs and 12 shorts picked at random among the candles that have the trend (close beyond the EMA 100, ADX above 20), without looking at the RSI. Repeated 2,000 times.

Random entries in the same trend (2,000 trials)Mean
Median+0.26 R
5th to 95th percentilefrom −0.21 to +0.76 R
Trials with a mean at least equal to the observed one (+0.29 R)45.5%

Random candles in the same trend make +0.26 R in the median, and almost half of the trials (45.5%) do at least as well as the system. Even taking all candles with a trend and leaving with the same rules the mean is positive: +0.19 R for the 598 candles with an uptrend (winners 43%) and +0.32 R for the 699 with a downtrend (49%). The RSI signal adds nothing visible in these data: the positive mean exists without it, and the period had long moves that the trend filter and the exits took advantage of.

There is also a detail by half of the period: in candles with an uptrend the mean is +0.46 R (351 candles) in the first half and −0.19 R (247 candles) in the second. The short side holds in both (+0.34 R with 237 candles, +0.32 R with 462). The trend filter has no fixed edge: it depends on the market.

How to test a system, in five steps

  1. Write the rules first, with standard values or values decided in advance: entry, stop, target, exit on time, cost.
  2. Count the trades. With a few dozen, the error is as large as the mean: say so.
  3. Record every variant tried and decide beforehand which one counts.
  4. Compare with random entries that use the same exits, in the same trend.
  5. Split the period and see whether the two pieces say the same thing.

Exercises

Chapter exercises

1. On the chart (H4, from 24 August to 4 September 2026) the letters sit on the system's entries. Which letter is the trade closed on time with a gain of +0.71 R?

Chart exercise. Data: gold, XAUUSDT perpetual future (real candles) · 4 hours (H4) · from 24 Aug 2026 00:00 UTC to 4 Sept 2026 20:00 UTC.
Read the chart in words
WhenWhatPrice
26 Aug 2026 20:00 UTCPoint A4,588.66
28 Aug 2026 04:00 UTCPoint B4,574.03
4 Sept 2026 12:00 UTCPoint C4,482.23

Times in UTC.

2. The system with the declared rules made 6 trades with a mean of +0.40 R ± 1.09 R. What can be said?

3. Widening the RSI thresholds from 40/60 to 45/55 takes the trades from 6 to 22. What is the problem to declare?

4. Random entries in the same trend, with the same exits, make +0.26 R in the median, and 45.5% of trials do at least as well as the system (+0.29 R). What does it mean?

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.