Part 3: Using them together · Chapter 13 of 14 · 10 min
Indicators on top of POI: do they really help?
Seven indicators used as filters on the 291 points of interest of book 4 (216 trades filled, mean R −0.10 ± 0.18): the "bounce" ones switch on in 4-18 cases and do worse, the "trend" ones are almost always on, and the only visible difference (MACD histogram, +0.34 R) is inside what chance produces.
The question
Book 4 gathered 291 points of interest (POI): 130 order blocks and 161 FVGs of gold on H4. A frequent question is whether an indicator on top of the POI improves the entry: "I buy the demand zone only if the RSI is below 40", "only if the MACD turns", "only if the ADX confirms". Here it gets tested.
What is measured
For each POI, the limit order at the near edge of the zone, stop from the far side plus 0.25 ATR, target 2R, cost 0.30 dollars per ounce (the same rules as book 4). The indicator is read at the candle in which the POI is complete, when it is already known. Of the 291 POI, 216 have the order filled (price comes back to touch the zone within 40 candles, without closing beyond the far side). With no filters the mean R is −0.10 ± 0.18 (winners 67, 31%), where ± is the 95% interval (book 4 writes the same result as −0.10 ± 0.09, with one standard error). Differences between groups are measured in standard errors. Some POI overlap (an order block and an FVG close together): the trades are not independent and the real error is larger.
Seven filters, in the direction of the POI (demand = up, supply = down):
| Filter | Trades with the filter on |
|---|---|
| RSI favoring a bounce (demand with RSI below 40, supply above 60) | 12 |
| Stochastic %K favoring a bounce (below 30 / above 70) | 18 |
| Band %B favoring a bounce (below 0.25 / above 0.75) | 4 |
| RSI in the direction of the POI (above 50 for demand, below 50 for supply) | 178 |
| Close beyond the EMA 50 in the direction | 160 |
| ADX above 20 | 178 |
| MACD histogram in the direction | 152 |
The first result is in the counts. Bounce filters switch on in very few cases (4 to 18 out of 216), because the POI is nearly always born inside a strong move, not in an oversold state. Trend filters are already on in 152-178 cases out of 216 (70% to 82%): a filter that is almost always on does not filter.
Bounce: few cases, worse result
| Bounce filter | With filter | Without filter |
|---|---|---|
| RSI (below 40 / above 60) | 12 tr., R −0.51 ± 0.66 (winners 17%) | 204 tr., R −0.08 ± 0.19 |
| Stochastic (below 30 / above 70) | 18 tr., R −0.51 ± 0.53 (winners 17%) | 198 tr., R −0.06 ± 0.19 |
| Band %B (below 0.25 / above 0.75) | 4 tr., all stopped (R −1.01) | 212 tr., R −0.08 ± 0.18 |
With 12 or 18 trades the errors are larger than the differences, and with 4 nothing can be measured. It points the same way as chapters 3 and 6: in this period, overbought and oversold readings did not anticipate a bounce. Figure 1 shows one: the demand FVG of 30 December 2025 at 04:00 (zone 4,355.35-4,368.52 $) is born with the RSI at 30. The limit order at 4,368.52 fills at 08:00, and the stop at 4,346.01 is hit at 20:00: −1.01 R.
Read the chart in words
H4 gold chart from 27 to 31 December 2025 with the demand FVG zone, the entry line, the stop line and the target line; below the 14-period RSI with levels 40 and 60.
| When | What | Price |
|---|---|---|
| 30 Dec 2025 00:00 UTC | Demand FVG | 4,355.35 – 4,368.52 |
| 30 Dec 2025 08:00 UTC | Entry | 4,368.52 |
| 30 Dec 2025 08:00 UTC | Stop | 4,346.01 |
| 30 Dec 2025 08:00 UTC | Target 2R | 4,413.53 |
| 30 Dec 2025 20:00 UTC | stop hit | 4,330.64 |
Times in UTC.
Trend: on almost always
| Trend filter | With filter | Without filter | Difference |
|---|---|---|---|
| RSI in the direction of the POI | 178 tr., R −0.12 ± 0.20 | 38 tr., R −0.04 ± 0.45 | −0.08 R (−0.3 errors) |
| Close beyond the EMA 50 | 160 tr., R −0.07 ± 0.21 | 56 tr., R −0.21 ± 0.34 | +0.14 R (0.7 errors) |
| ADX above 20 | 178 tr., R −0.11 ± 0.20 | 38 tr., R −0.05 ± 0.45 | −0.06 R (−0.3 errors) |
| MACD histogram in the direction | 152 tr., R 0.00 ± 0.22 (winners 34%) | 64 tr., R −0.34 ± 0.31 | +0.34 R (1.7 errors) |
Three filters out of four change nothing: the differences are a fraction of the error. The MACD histogram does better (0.00 against −0.34 R, winners 34% against the 31% baseline) and is the largest difference: 1.7 times its error, so inside the interval (the usual requirement is at least 2). Figure 2 shows a POI with the histogram in favor that reaches the target: the demand FVG of 22 January at 20:00 (zone 4,871.65-4,910.90 $), order at 4,910.90, filled on the 23rd at 08:00, target at 5,015.78 hit on the 24th at 16:00 (+1.99 R). Figure 3 shows the other side: the supply FVG of 5 March at 16:00 (zone 5,102.94-5,135.00 $) with a negative histogram, so in favor: the order at 5,102.94 fills at midnight on 6 March and the stop at 5,153.11 is hit at 12:00 (−1.01 R).
Read the chart in words
H4 gold chart from 20 to 25 January 2026 with the FVG zone, the entry line, the stop and target lines; below the MACD with histogram, line and signal.
| When | What | Price |
|---|---|---|
| 22 Jan 2026 16:00 UTC | Demand FVG | 4,871.65 – 4,910.90 |
| 23 Jan 2026 08:00 UTC | Entry | 4,910.90 |
| 23 Jan 2026 08:00 UTC | Stop | 4,858.46 |
| 23 Jan 2026 08:00 UTC | Target 2R | 5,015.78 |
| 24 Jan 2026 16:00 UTC | target hit | 5,032.90 |
Times in UTC.
Read the chart in words
H4 gold chart from 3 to 7 March 2026 with the supply FVG zone, the entry line, the stop and target lines; below the MACD with histogram, line and signal.
| When | What | Price |
|---|---|---|
| 5 Mar 2026 12:00 UTC | Supply FVG | 5,102.94 – 5,135.00 |
| 6 Mar 2026 00:00 UTC | Entry | 5,102.94 |
| 6 Mar 2026 00:00 UTC | Stop | 5,153.11 |
| 6 Mar 2026 00:00 UTC | Target 2R | 5,002.61 |
| 6 Mar 2026 12:00 UTC | stop hit | 5,167.24 |
Times in UTC.
The benchmark: seven filters tried
Seven filters are seven tries. The largest gap among the four measurable filters (at least 20 trades with and without) is 1.7 errors, the MACD one. The test is repeated assigning the indicator readings to the POI at random (1,000 times) and each time the largest gap among the four filters is taken: the median is 1.3 errors, the 95th percentile 2.6. A result like 1.7 errors comes out of chance in 27% of the trials, and to say a filter works you would need something beyond 2.6.
On top of the book 4 SMC filters
Book 4 had its own filters (aligned D1 trend and right half of the range): 25 trades, R +0.35 ± 0.59. Adding an indicator:
| SMC filter + indicator | Trades | Mean R | Without the indicator (same 25) |
|---|---|---|---|
| + MACD histogram in the direction | 13 | +0.82 ± 0.83 | 12 tr., −0.16 |
| + ADX above 20 | 18 | +0.55 ± 0.69 | 7 tr., −0.17 |
| + RSI or EMA 50 in the direction | 16 | +0.29 ± 0.75 | 9 tr., +0.44 |
With 13 or 18 trades the intervals are as wide as the results and overlap. Two indicators out of three seem to improve and the third gets worse: with this sample you are reading chance.
How to test a filter, in four steps
- Write the filter beforehand, with standard thresholds, and count how often it switches on: a filter almost always on or almost always off says nothing.
- Compare all the trades with and without the filter, with the error.
- If you try several filters, count the tries and compare the largest difference with the one chance produces.
- Add it to the other filters only if the sample stays large enough: under 30 trades, do not conclude.
Exercises
Chapter exercises
1. On the chart (demand FVG of 22 January 2026, H4) the letters sit on four candles. Which letter is the candle in which the limit order fills (price returns to the edge of the zone)?
Read the chart in words
| When | What | Price |
|---|---|---|
| 22 Jan 2026 16:00 UTC | Point A | 4,865.00 |
| 23 Jan 2026 08:00 UTC | Point B | 4,900.39 |
| 24 Jan 2026 08:00 UTC | Point C | 4,988.10 |
| 24 Jan 2026 16:00 UTC | Point D | 4,996.44 |
Times in UTC.
2. The filter "RSI below 40 for demand, above 60 for supply" switched on in 12 trades out of 216, with R −0.51 ± 0.66. What can be said?
3. The filter ADX above 20 is on in 178 trades out of 216. Why does it not filter?
4. The MACD histogram gives a +0.34 R difference, equal to 1.7 errors, among seven filters tried. With indicators assigned at random the largest gap among the four measurable filters has a 95th percentile of 2.6 errors. What follows?
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.