Combining Indicators Without Redundancy: One Tool per Family

You open a chart and stack RSI, Stochastic, CCI and MACD until the screen is full. When every line points the same way, it feels like more confidence. The problem is that most of these indicators are calculated from the same prices and ask the same question. Their agreement is not extra evidence. It is one opinion repeated several times. This article sorts indicators into families, shows in numbers how closely oscillators move together, and builds a small toolkit where each tool has its own job.

The four indicator families

Each indicator answers a question. Group them by the question they answer and you get four main families.

Family	Question it answers	Examples
Trend	Which way is the market going, and how strong is the trend?	Moving Average, MACD, ADX, Ichimoku
Momentum (oscillators)	Is price speeding up, or running out of steam?	RSI, Stochastic, CCI
Volatility	How widely is price swinging?	ATR, Bollinger Band width, Donchian Channel width
Volume	How much is being traded?	Volume, OBV (tick volume in Forex)

Some indicators straddle two families. MACD is the difference between the 12 and 26 EMAs, so it shows trend direction, but its rises and falls are price momentum, and it behaves a lot like an oscillator. Bollinger Bands have a 20 SMA as the middle line, which is about trend, while the band width (±2 standard deviations) is about volatility. ADX only measures trend strength, not direction.

Three oscillators are one opinion repeated

Here are the short formulas for three indicators people like to stack.

• RSI 14 compares the average of up-closes with the average of down-closes over 14 bars (Wilder smoothing) and scales the result to 0–100
• Stochastic 14,3,3 finds where the close sits within the 14-bar High–Low range, averages it over 3 bars for %K, and another 3 bars for %D
• CCI 20 measures how far the typical price (High + Low + Close) ÷ 3 is from its 20-bar average, in units of mean deviation, divided by the constant 0.015

Different formulas, same idea: is the latest price high or low compared with the recent short range? When price runs up hard, all three reach overbought together.

We built an illustrative 160-bar price series and computed all three with the standard formulas, then measured the correlation over the last 120 bars. Correlation runs from −1 to 1. The closer to 1, the more two lines rise and fall together.

Pair	Correlation
Stochastic %K and CCI	0.95
RSI and the MACD line	0.93
RSI and Stochastic %K	0.90
RSI and CCI	0.82
RSI and ATR 14 (different family)	−0.18

These numbers are an illustration, not a constant of the market. We generated four more random series the same way. The correlations between oscillators stayed between 0.79 and 0.95 in every one, while RSI and ATR ranged from −0.14 to 0.29, flipping sign. That matches what the formulas tell you: oscillators move together, and volatility is a different piece of information.

Three lines that agree are not three pieces of evidence. If they are calculated from the same prices and ask the same question, they are one piece of evidence.

A toolkit template: one job per family

Instead of looking for more confirmation, split the work into four jobs, each done by a different kind of tool.

1. Trend filter decides which side you look for today, for example the close above the 200 Moving Average on a higher timeframe, or ADX above 20 to 25, a common threshold for a trending market
2. Entry trigger decides when to enter, for example RSI dipping and turning back up, or price breaking a Donchian Channel. One trigger is enough
3. Volatility-based Stop loss decides how far away the Stop loss goes, for example 1.5 times ATR, so the distance adapts to how widely price is swinging right now
4. Risk-based position size decides how many lots, from the money you accept losing per trade divided by the Stop loss distance. See Position size from your Stop loss

Jobs three and four are the ones people skip, yet they are the only part that reliably limits the damage.

Worked example: ATR Stop loss and position size

This example explains the structure. It is not a tested system. Say you trade EURUSD with a 5,000-dollar account and risk 1% per trade.

1. Money at risk = 5,000 × 1% = 50 dollars
2. ATR 14 on the 4-hour chart at the time = 0.0042, or 42 pips
3. Stop loss distance = 1.5 × 42 = 63 pips
4. On EURUSD, 1 pip on 1 lot = 10 dollars, so position size = 50 ÷ (63 × 10) = 0.079 lot
5. Round down to 0.07 lot (rounding up would risk more than planned)

Check: if the Stop loss is hit, you lose 0.07 × 10 × 63 = 44.1 dollars, or 0.88% of the account, within the 1% plan. On days when the market swings harder, ATR widens, the Stop loss moves further away and the lot size shrinks on its own, so the money at risk stays the same. Try your own numbers with the lot calculator at the end of this article, and read why 1% in Risk 1% per trade.

Real confluence versus fake confluence

Confluence means several reasons pointing the same way at the same place. One question separates the real kind from the fake kind: do those reasons come from different sources of information?

Fake, looks like confirmation but adds nothing:

• RSI, Stochastic and CCI all oversold at once on the same chart
• Divergence on RSI and on MACD at the same point, both measuring momentum from the same closes
• Three moving averages with close periods, such as 20, 21 and 25

More real, because each looks at something different:

• Trend direction from a higher timeframe, timing from a lower one
• An entry at a clear support or resistance level, or where price has turned several times
• Enough volatility to reach the target, for example a target distance that isn’t too large compared with ATR
• Trading at a time without major news. See How to read the economic calendar

Even real confluence still needs testing, because a filter that sounds sensible can make the overall result worse.

A filter that makes the numbers prettier and the profit vanish

In our backtest on gold (XAUUSD), 1-minute chart, 2026 up to September, using real tick data at 0.10 lot per trade, a system that enters on pullbacks to the 50% level of a swing had a Profit factor (total profit ÷ total loss) of 1.48 and a net profit of 10,927 dollars. We added a confluence filter: a smaller swing on a smaller frame also had to stretch to a set Fibonacci level at the entry point.

Version	Trades	Profit factor	Net profit (dollars)
No filter	all	1.48	10,927
Loose filter (stretch to 127.2)	55	1.74	not recorded separately
Strict filter (stretch to 161.8)	27 (51.9% won)	2.14	1,492

The Profit factor improved clearly, but the strict filter kept only about 7% of the trades, and net profit fell by about 86%. The filter really did pick out better-than-average trades, but it threw away a large number of trades that were still profitable. We concluded it was not worth using as a filter. These are backtest results. Past results do not predict future results, and they are not returns you should expect.

A filter that makes you trade less has to beat simply changing the lot size at the same risk. If scaling the original system down to the filtered system’s risk earns more, the filter didn’t help.

Changing the lot size is the cheapest option there is: no extra rules and no change in the number of trades. A filter that removes trades also tends to make drawdown look lower without making the system safer, because fewer trades means fewer chances of a long losing run. A rough rule is that drawdown grows with the square root of the number of trades. If a system has a 3% drawdown over 600 trades and you cut it to 200, fewer trades alone would give about 3 ÷ √3 = 1.7%. If the filtered version shows 2%, it is worse than expected, not better.

How to test an indicator combination

1. Write the rules before you look at results. Every rule must be a number, for example “RSI 14 closes below 40, then the next bar closes above 40”, not “RSI looks low”
2. Count every signal, including losers, the ones you hesitated on, and the ones that fired while you weren’t watching. Counting only the signals you remember is hindsight bias
3. Add one tool at a time. When you add a filter, look at the trades it removes. If they earned about as much on average as the ones left, the filter only reduced the trade count
4. Read profit per trade and total profit alongside the Profit factor, and compare with adjusting the lot size at equal risk
5. Split results by year and by market condition. A combination that only works in strong-trend years can break when the market regime changes
6. Watch out for fitting settings to the past. Good settings sit on a plateau, where nearby values give similar results. Read more in Overfitting

Common mistakes

• Counting several oscillators as several reasons. RSI, Stochastic and CCI agree because they come from the same prices
• Reading Forex volume as real traded volume. What you see is the tick volume of that broker’s feed
• Fading a strong trend with an oscillator. In a strong trend RSI can stay overbought for a long time. Read more in RSI and Stochastic
• Having no tool for the Stop loss and position size, even though that part decides how much a wrong trade costs
• Celebrating a higher Profit factor from a filter without checking how much total profit and how many trades disappeared
• Switching indicators after every losing streak, even though losing streaks are normal for every system