ADX vs Supertrend as a Trend Filter: Same Data, Four Versions

Supertrend beat ADX as a trend filter on EURUSD 5-minute bars, and the reason has nothing to do with which indicator is smarter. Across 16,926 real bars from June 3 to August 26, 2026, Supertrend left a losing EMA 12/26 system less bad and ADX left it more bad. Both were tested with identical rules on identical data, so the comparison is clean.

What exactly was compared?

One base system, four versions of it. Nothing else changed between the runs.

• Base signal: EMA 12/26 crossover, long when fast closes above slow, short when below.
• ADX filter: take the signal only if ADX(14) is above 25 on that bar.
• Supertrend filter: take the signal only if Supertrend (ATR 10, multiplier 3) already points the same way.
• Both: the signal must satisfy the ADX and the Supertrend condition at once.
• Measurement: move in pips two hours later, no stop, no target, overlapping trades skipped.

Which filter did better?

Notice that Supertrend has a lower win rate than no filter and still lost less money. It removed a small number of trades that were losing badly rather than a large number of trades at random. That is what a filter is supposed to do.

Why did stacking both make it worse?

Because the two filters are not independent. Supertrend flips direction when price moves decisively, and ADX rises after price moves decisively. Requiring both means requiring the move to be already well advanced twice over. What survives is the late part of trends, which is exactly the part that reverts.

This is the same failure mode I documented when adding an EMA 200 filter to crossovers in the crossover test: two lagging conditions do not add up to confirmation, they add up to arriving late.

How do the two indicators actually differ?

Once you put it that way, the test result stops being surprising. Asking ADX whether to take a long is asking a thermometer which way the wind blows. The correct pairing for ADX is its own +DI and -DI lines, which did improve with an ADX filter, and I show those numbers in the ADX above 25 test.

Does Supertrend work as a system on its own?

Not from this evidence. Every version in the table above lost money, including the best one. What Supertrend did here was reduce damage, and that makes it a reasonable gate on top of a signal that already has an edge, not a signal generator. The same distinction applies to the Keltner and Bollinger comparison in my bands article, where the wider envelope was the better filter and the worse trigger.

What I would actually do with this

• Pick a filter by what it removes, not by its reputation. Count the signals it deletes before you count the win rate it produces.
• Do not stack two lagging filters. The combined version was the worst of the four here and the worst in the crossover test too.
• Give ADX its own direction lines instead of borrowing someone else’s entry.
• Match the exit to the filter. Trend filters paid at an eight hour hold and lost at two hours, shown in the MT5 guide.

Frequently asked questions

Which Supertrend settings did you use?

ATR period 10 with multiplier 3, the standard defaults. I did not sweep the parameters here because the point was a like for like filter comparison, not an optimisation. Parameter sweeps on other indicators are in my Stochastic settings guide and my Bollinger settings guide.

Would a different ADX threshold beat Supertrend?

ADX above 20 came closest, turning the average trade slightly positive at +0.11 pips over 150 signals, which is better than Supertrend on average pips and worse on sample size. Neither is a business. The threshold table is in the ADX test.

Can I use Supertrend as a trailing stop instead of a filter?

That is its most defensible use, since the line is an ATR distance from price and moves only in the trade’s favour. I have not measured it as an exit on this dataset, so I will not quote a number for it.

Does this hold outside EURUSD?

Unknown from this test. One instrument, one timeframe, one three-month window. What generalises is the method: count what a filter deletes, and compare filters on identical data rather than on separate charts.

Why do both filters still lose money?

Because the underlying EMA 12/26 crossover has no edge on this data to begin with, which I established in the EMA against SMA comparison. A filter can only reduce the damage of a losing signal. It cannot invert it.