Why you still can't win even with overlapping indicators — what remained after the verification was the "price structure"

“This indicator on its own won’t win, but if I combine it with that indicator, I should be able to win,” I believed for a long time as I kept piling on filters.

In short, this didn’t work out well.Even stacking several indicators that have no inherent advantage on their own didn’t make them win. Rather, the more conditions I added, the more it tended to become a “visible upward trend” that only fit past data, and it collapsed when moved to a different period. This article honestly explains what I discarded and what I kept after that detour.

Pardon the delay,I am Lulu.fx. I am a current discretionary trader focusing on GOLD (XAUUSD), repeatedly building and testing EA (expert advisor) programs myself. I am now challenging the prop firm (Company A).

In this “Strengths and Weaknesses Found in Indicator Verification” series, I have checked ADX, moving averages, oscillators, Ichimoku, volume, and even the differences in implementation by tool—doing each by hand. This time, as a closing piece, I will answer, “So what should have been the core approach?” This is the entry I wish most to reach those who are tired of piling on filters.Closing as the answer to “In the end, what should I have based my approach on?” I will write. This is the post I want to reach the most for people who are exhausted from stacking filters.

The more filters you add, the weaker you become for some reason

First, I’ll start with the pain I struggled with the longest.

I want to improve win rate. I want to reduce losses. When you think that way, the most natural thought is to “add conditions.” “Only when the moving average is rising,” “Only when ADX is above a certain level,” “Only when the oscillator isn’t oversold,”— you stack filters like this, andthe performance on past data becomes astonishingly clean.

The graph trends upward, losing trades disappear, and the win rate rises. That moment feels truly good. I thought, “I’ve finally found a winning logic.” I tasted this pleasure many times.

However, that “perfectly complete logic” when applied to adifferent period not used for testingfell apart quickly. The nice upward trend in one period became jagged in another. In the worst case, the simple logic before adding filters performed better.

This wasn’t a one-time thing. Every time I added a condition, past performance looked better, but the future (in another period) was weaker.“The more you stack, the weaker you are in the real run”This reverse phenomenon kept occurring. I didn’t understand why at the time.

---

“Weak + Weak” does not become “Strong”

As I tested more, I realized one obvious truth.

Indisputably, indicators that have no edge on their own often don’t gain an edge when combined. Of course it’s obvious in hindsight, but when you’re in the thick of it you can’t see it. The idea that “you can win by combining indicators that each cannot win alone” often has little to no basis.

Why so? One reason is thatcombining indicators with similar weaknesses doesn’t diversify losses.

For example, both moving averages and ADX are calculations that “smooth the price by averaging.” So in clearly trending markets, both are strong, and in narrow ranges they are weak. They behave similarly. Stacking these two together only leads tothe same loss scenarios (false breakouts in ranges) piling up, and one doesn’t cover the other’s losses.

When diversifying risk in investment, you combine assets with different price movement tendencies. Collecting ones that move the same won’t diversify. The same goes for indicators: stacking similar ones doesn’t remove their weaknesses; it amplifies losses in the same situations.

What started as a “combination to raise win rate” turned into a “combination that reinforces the same losses.” This was one of the major reasons it never won despite stacking.

---

Adding more conditions brings you closer to overfitting

There’s another, deeper reason.Overfitting.

If this term is unfamiliar, I’ll put it simply. Overfitting meansoverfitting the logic to the past by fitting to a randomPast sequence. In test prep terms, it’s like memorizing only past questions so you cannot handle new problems in the actual test.

This is the scary part, butthe more filters you add, the more overfitting progresses. Each additional condition makes the logic tailor to the details of past sequences. Past performance becomes cleaner and cleaner. But that cleanliness is not about future robustness; it is about fitting to the past.

To give a concrete example. At one point I stacked several indicator filters and crafted a logic whose testing period showed a flawless upward trend. Win rate was high, and losses were truly few. I thought, “This is real.” When I applied it to a different, unused testing period, it hardly worked. It only fit the past market structure, and as soon as the market structure changed, it stopped dead.

I fell for that trap many times. And only then did I learn.“Cleanliness in the verification period does not prove strength”. In fact, the cleaner it is, the more one should suspect overfitting. Complex logic that adds many conditions usually carries this trap.

---