Indicator Combos vs Their Own Parts: All Eight Finished Below Their Best Half

My test of all 56 indicator pairs answered the general question: confirmation between two indicators lowered accuracy by 1.1 points and a coin flip filtered better. But “all 56 pairs” is an abstraction, and nobody trades an abstraction. People trade specific, named combinations they were taught by name.

So this is the specific version. Eight combinations that exist in real trading education, each measured against the indicators inside it, on the same bars, with the same rule. The question is narrow and fair: did the combination beat its own parts?

How the comparison is set up

A combination trades when all of its members hold the same state and that joint state has just changed. A single indicator trades when its own state changes. Both enter at the close of the signal bar, both exit 24 bars — two hours — later, and both pay 1 pip, the cost I measured in my spread and swap test.

The eight components, each reduced to a single bullish-or-bearish state, are RSI 14 against 50, Stochastic 14/3/3 %K against %D, MACD 12/26/9 line against signal, the 50/200 EMA cross, CCI 20 against zero, close against the Bollinger 20 midline, +DI against -DI, and Parabolic SAR 0.02/0.2. Alone, they perform like this:

One profitable row out of eight, and it is the slowest and rarest signal on the list — the same 50/200 cross I measured on its own in the moving average crossover test. Keep that row in mind, because every combination below is being judged against the better of its own halves, not against it.

Did any combination beat its own components?

Eight combinations, eight “below.” Every single one finished worse per trade than the better indicator inside it, and six of the eight finished worse than both of their halves.

The table also reads top to bottom as a count. Two-indicator combos occupy the top; the three-indicator stack is fourth from the bottom; the four-indicator stack is last, at 42.1% and -1.45 pips. The more indicators in the combination, the worse it did. That is not a subtle gradient — from the best two-member combo to the four-member one is 6.9 points of win rate.

What about the one combination that made money?

MACD + MA cross is the only positive row, at +0.44 pips over 98 trades, and it deserves a careful look because it is the combination most likely to be presented as proof that confluence works.

The combination kept 98 of the moving average’s 101 signals and produced 40% less profit. MACD removed three trades and those three trades were worth 29 pips.

Notice what is actually happening here. MACD alone lost 1,772 pips. Attached to the 50/200 cross it looks like part of a winning system, but it contributed nothing except the removal of three signals. The system is the moving average. MACD is a passenger with a 0.5-point win-rate alibi and a 0.27-pip fare.

This is the trap in every combination that contains one slow component: the slow component carries the result, and whatever is bolted onto it inherits the credit. My MACD versus moving average crossover comparison found the same relationship from the opposite direction — 677 of 677 signals landed on the same bar.

Does a longer hold change the ranking?

A fair objection: two hours is short for a trend combination. So here is every combination held for 48 bars — four hours — instead of 24.

Win rates improved across the board — every combination gained two to three points simply by being given more time — and the pips per trade barely moved, because a longer hold pays the same 1-pip cost over more distance travelled in both directions. The ordering is unchanged except for cosmetic swaps in the loss column.

And the headline comparison survives intact: MACD + MA cross at 4 hours makes +1.92 pips per trade. The 50/200 cross alone at 4 hours makes +2.54. The combination is still paying MACD a fee to do nothing.

Why do combinations lose to their own parts?

Because a combination is a filter, and a filter can only do one thing: remove trades. It cannot improve the trades it keeps. So the only way a combination beats its components is if the trades it removes were worse than average — and across 16,926 bars, they were not.

Two mechanisms show up in the numbers. The first is delay. A combination signals when the last member agrees, which by definition is later than the earliest member signalled. On a 5-minute chart, later means worse entry: the RSI + Stochastic + MACD stack fires on average well after the first of the three has already moved, and its win rate is 1.9 points below the Stochastic alone.

The second is sample destruction. The four-indicator stack kept 832 of the Stochastic’s 4,197 signals — 80% of the data thrown away — and finished 3.1 points worse. If the filter carried information, cutting the sample by four fifths should have concentrated the good trades. It concentrated nothing.

This is the same structural finding as my RSI divergence versus overbought comparison: the more conditions you stack onto an entry, the rarer and more special it feels, and the less it is supported by anything you can count.

The comparison that actually matters

Forget combos against components for a moment. The spread between the best and worst row in this entire study is between 101 trades at +0.71 pips and 4,197 trades at -1.08 pips. That gap dwarfs every combination effect measured here, and it is a gap in signal frequency, not in indicator count.

Everything slow did comparatively well. Everything fast did badly, alone or combined, filtered or unfiltered. If you are searching for the lever, it is there — the same conclusion my session versus all-day trading test reached by restricting when you trade rather than what confirms it.

What I would actually do with this

• Test any combination against its own halves before trusting it. Eight out of eight failed that test here, including the one that made money.
• Suspect the slow member. If a combination contains one slow indicator and performs well, measure the slow indicator alone. In this study it was better alone every time.
• Stop adding a third. Two-member combos averaged 43.9% here, the three-member stack 43.3%, the four-member stack 42.1%. Each addition cost accuracy.
• Give trades more time before adding more conditions. Doubling the hold added two to three win-rate points to every combination. No confirming indicator anywhere in my 56-pair study added more than 0.9.
• Count what the filter removes. If a filter cuts 80% of your signals and moves your result by a point, it is not a filter, it is a delay.

Frequently asked questions

Is RSI + Stochastic + MACD a good combination?

On 16,926 EURUSD 5-minute bars it produced 1,066 trades at 43.3% and lost 1,316 pips after costs. Each of its three components did better on its own — the Stochastic at 45.2%, RSI at 43.6%, MACD at 43.3% — so the stack finished below all three in pips per trade at -1.23.

Are two indicators better than one in forex?

Not in this test. All eight named combinations finished below their best single component in pips per trade, and six of eight finished below both components in win rate. The aggregate version of the same question across all 56 possible pairs gave 44.2% unfiltered against 43.1% confirmed.

Which indicator combination was profitable?

Only MACD + the 50/200 EMA cross, at +0.44 pips over 98 trades and +43 pips total. The 50/200 cross traded alone made +0.71 pips over 101 trades and +72 pips total, so the profitable combination was less profitable than one of its own halves.

Does adding more indicators reduce false signals?

It reduces signals, not false ones. The four-indicator stack kept 832 of the Stochastic’s 4,197 signals — an 80% cut — and its win rate fell 3.1 points. If the removed signals had been disproportionately bad, the remaining ones would have improved. They did not.

Should I use a longer timeframe instead of more indicators?

The data points that way. Doubling the hold from 2 to 4 hours added two to three win-rate points to every combination tested, while the best confirming indicator in my 56-pair study added 0.9. And the only profitable base, the 50/200 cross, is simply the slowest signal in the set at 101 trades in three months.

How do I build these combinations in MetaTrader?

All eight components ship with MT4 and MT5 as standard indicators, so no purchase or download is involved — it is a matter of which inputs you set and what you require of them. I walk through the whole grid, including which settings survived costs, in my multi-indicator MT4 template guide, and the installation basics are in setting up your first MT4 chart.