Gold as a Filter for EURUSD vs a Plain EMA Cross: The Best of 24 Filters Halved the Loss, and None Made Money

“Check gold before you trade EURUSD” sounds like good advice. The two markets move together — in my EURUSD and gold correlation test they moved the same way on two of every three bars. So a EURUSD buy signal while gold is also rising should be a better buy. I tested that on the most common signal there is, the moving average crossover, and compared it with the plain version and with random entries.

What exactly was tested?

• Base rule: EMA 9 crosses EMA 21 on EURUSD M5. Buy on a cross up, sell on a cross down, enter at the next bar’s open, exit after 30 minutes, 1 hour or 2 hours. No overlapping trades.
• Gold filters, all read at the moment of the EURUSD cross:
  • gold’s own EMA 9 above or below its EMA 21
  • gold above or below its 50-period EMA
  • gold above or below its 200-period EMA
  • gold up or down over the last hour
• Agree keeps only the crosses where gold points the same way. Disagree keeps only the opposite ones — if the filter means anything, the two should differ clearly.
• Data: 13,631 EURUSD and gold 5-minute bars that opened at the same minute, June 16 to August 26, 2026. Cost: 1 pip per trade, from my spread test. Every result is split at July 15.
• Random baseline: 200 runs of random buy-or-sell entries on the same bars, with the same holding times and the same cost.

How did the plain EMA cross do?

Net pips per trade after a 1-pip cost.

The plain cross lost more than random entries at every holding time. Random entries lose about the cost — one pip — because they have no direction. The cross lost an extra half pip on top, because on M5 it tends to fire after the move is done. My MACD vs moving average crossover test measured that lateness: the MACD signal line fired a median of 25 minutes before the crossover. So the filter starts with a rule that is worse than a coin flip. The question is whether gold can fix it.

Does a gold filter improve the EMA cross?

Net pips per trade; trades in brackets. Filtered versions can have more trades than the unfiltered one because skipped crosses free up the time for later ones.

At one hour, the slower gold trend filters helped. When gold was above its 50 EMA on a EURUSD buy (or below it on a sell), the loss fell from 1.46 to 0.71 pips per trade, and the opposite group lost 1.83. The gap held in both halves: -0.25 and -0.97 for “agree”, against -1.71 and -1.30 unfiltered. The 200 EMA pointed the same way, with a smaller gap.

The fast filters did nothing useful. Gold’s own EMA 9/21 — the filter most people would reach for, because it is the same indicator — moved the result by 0.07 pips. Gold’s last hour was slightly worse when it agreed.

But even the best filter still lost 0.71 pips per trade after cost. It turned a rule that was worse than random into one slightly better than random. It did not turn it into a strategy.

Does the gold filter hold on other holding times?

This is where the best filter falls apart. Held for two hours, “gold above its 50 EMA” lost 1.87 pips — no better than no filter at all. And at two hours, the fast filters flipped completely: when gold’s EMA 9/21 agreed, the trade lost 2.98 pips; when it disagreed, only 1.00. Waiting for gold to confirm a move that both markets already made meant buying later into a move that was more likely to snap back.

The only filter that helped at every holding time was the slowest one, gold against its 200 EMA — by 0.16 to 0.93 pips. That is the pattern I keep seeing: a slow filter that says “which side is the bigger trend on” helps a little; a fast one that repeats the same signal on another chart helps nothing. My indicator combos vs their own parts test pointed the same way on one chart: the slow 50/200 cross did best, and the added indicators mostly deleted trades.

Why doesn’t gold confirmation work better?

Three numbers from the same bars explain it:

• The crosses rarely line up. Of 670 EURUSD EMA 9/21 crosses, only 173 — 25.8% — had a gold cross in the same direction within 3 bars either side. A correlation of 0.42 bar by bar is not enough to make two sets of moving averages cross together.
• The confirmation arrives at the same time as the signal. When both markets move together, it is usually one dollar move hitting both at once. Gold agreeing at that moment tells you the dollar moved — which EURUSD already told you.
• Gold does not lead. A gold bar’s correlation with the next EURUSD bar was 0.044. There is no advance warning to filter with.

Is a gold filter better than a second indicator?

Not really. In my indicator combinations test, most confirming pairs on one chart measured the same thing twice — CCI and the Bollinger midline agreed on 95.1% of bars — and a coin flip used as the confirming indicator did better than the average real pair. Gold is a different market, so it disagrees more often — which is why the 50 EMA filter separated winners from losers at one hour. But the separation was not stable across holding times, and it never lifted the rule above its cost.

Should you check gold before a EURUSD trade?

• If you already have a EURUSD rule that makes money, a slow gold trend filter may help it a little. Use gold’s 200 EMA or 50 EMA, not a fast gold signal, and test it on your own holding time — the 50 EMA filter helped at one hour and not at two.
• Don’t use gold to rescue a losing rule. The best filter still lost 0.71 pips per trade.
• Don’t wait for gold’s own crossover. It came at the same time or later, and at two hours it made the trade worse.
• Remember what the filter really measures: the dollar. If gold and EURUSD are both up, you are long a dollar move twice, not confirmed once.

The same-rules-on-both-markets comparison is in my gold vs EURUSD test, and if you are tempted to tune the filter until it works, read optimized vs default settings first.

Frequently asked questions

Does gold confirm EURUSD signals?

Weakly. On 13,631 matched 5-minute bars, an EMA 9/21 cross on EURUSD lost 1.46 pips per trade over one hour; taking only the crosses where gold was on the same side of its 50 EMA cut the loss to 0.71. None of 24 gold-filtered versions made money after a 1-pip cost.

What is the best gold filter for EURUSD?

In this test, the slow ones. Gold above or below its 50-period EMA gave the biggest improvement at a 1-hour hold. Gold against its 200 EMA was the only filter that helped at every holding time, by 0.16 to 0.93 pips per trade.

Should both EURUSD and gold give a crossover signal?

No. Only 25.8% of EURUSD crosses had a gold cross in the same direction within 15 minutes, and requiring gold’s EMA 9/21 to agree made the 2-hour trade lose 2.98 pips instead of 1.86.

Is an EMA 9/21 crossover profitable on EURUSD M5?

Not in this data. It lost 1.38 to 1.86 pips per trade after a 1-pip cost depending on the holding time — more than random entries, which lost about the cost.

Why does the EMA cross lose more than random entries?

On a 5-minute chart, the cross usually fires after a short move is already over, so it buys near small tops and sells near small bottoms. Random entries lose only the spread.