Two Indicators Are Not Better Than One: What Confluence Testing Shows

The Promise of Confluence

Every trading course eventually arrives at the same slide: never trade on one signal alone. Wait for two indicators to agree. Three is better. This idea — called confluence — is sold as the cure for false signals, the thing that separates disciplined traders from impulsive ones.

The logic sounds reasonable. If RSI is oversold and price is at a support level and MACD is curling up, surely that's stronger evidence than any one of those conditions alone. More filters, cleaner trades. The pitch is almost universal, and it's almost never tested.

What 660,005 Backtests Actually Show

We ran 660,005 out-of-sample backtests across 903 assets — stocks, forex, crypto, ETFs, commodities, and indices — across the 1-Hour, 4-Hour, Daily, and Weekly timeframes, with realistic costs applied throughout.

For single indicators tested individually, 63% of assets had at least one indicator that beat buy-and-hold on a risk-adjusted basis. That's a reasonable hit rate. When we tested indicator combinations — pairing a second indicator as a filter on top of the first — the number dropped to 26%.

Adding a second indicator to filter signals reduced the probability of outperformance by more than half. That's not noise. Across hundreds of assets and thousands of combinations, the result was consistent: stacking indicators tends to hurt more than it helps.

Why Stacking Usually Backfires

There are a few mechanical reasons the data comes out this way.

First, signal reduction compounds costs. Every filter you add eliminates trades. Fewer trades means the trades you do take carry more weight individually — but they still have to overcome the same fixed friction: spreads, slippage, overnight exposure. When a second indicator cuts your signal count in half, your edge needs to roughly double just to stay even. It rarely does.

Second, correlated indicators aren't independent confirmation. RSI and Stochastic both measure recent momentum relative to recent range. MACD and an EMA crossover both track moving-average divergence. When you stack two indicators from the same family, you're asking the same question twice with slightly different wording. The second filter feels like additional evidence, but it shares most of its information content with the first.

Third, and most insidious, is intuitive curve-fitting. When a trader adds a second indicator because it "confirmed the last few setups," they've discovered a pattern in a small sample. Small samples lie. Our results are tested on data the system never saw during development — which is why the 26% figure is lower than most traders expect.

What the Data Points Toward Instead

The more useful question isn't which two indicators to combine. It's which single indicator best fits a specific asset and timeframe — and then using it cleanly, without layering noise on top.

The top performer varied widely by asset class. Fisher Transform led forex, topping 17 currency pairs. Fibonacci Pivots led stocks with 22 assets. Keltner Mean-Reversion topped commodities. The winning indicator for your asset is likely not the same one that wins elsewhere — which is itself evidence against universal confluence rules. A rule that says 'always confirm with a second momentum oscillator' assumes that oscillator adds value everywhere. It mostly doesn't.

The trap is clearest in the win-rate data. Holy Grail Confluence — an indicator built entirely around the confluence concept — showed a median win rate of 73.3% across the assets where it was tested. That sounds excellent. But only 8% of those assets actually beat buy-and-hold. High win rate, poor risk-adjusted return. Filtering out losing trades feels productive. What it actually does, more often than not, is filter out the larger winners that drive overall performance — leaving you with a lot of small wins and no room to compound.

A Note on These Results

Everything cited here comes from hypothetical backtests with realistic transaction costs applied. These are not live trading results, and past backtest performance does not guarantee future results in live markets. Nothing here is financial advice — use it to inform your own research, not as a signal to buy or sell anything.

The purpose of running 660,005 tests isn't to hand you a ready-made system. It's to give you a less biased starting point than a handful of forum posts and your most recent trades. The data says most combination strategies underperform their best single component. That's worth knowing before you build your next confluence setup.