Can "Uncorrelated EAs" Slash Drawdown? Why Our Dream Failed!

A beginner-friendly summary of the verification: “Can “Uncorrelated EAs” Slash Drawdown? Why Our Dream Failed!”.

Breakout entry example (XAUUSD daily, real data): buy when price breaks above the recent high.

What’s the idea?

We’re always looking for ways to make our algorithmic trading systems (EAs) more robust and profitable. One of the biggest enemies of consistent profits is “drawdown” (DD) – that’s when your trading capital shrinks from its peak before it recovers. A common belief is that high drawdown often comes from having too many trades that move in lockstep, or are “correlated.” So, we had a hypothesis: if we could build the core of our trading system using currency pairs that have very low correlation with each other, we could significantly reduce system drawdown. Lower drawdown means we can safely use more leverage, which in turn could lead to higher profits. It’s like building a diversified investment portfolio; you don’t want all your assets to tank at the same time. If some go down, others might go up or stay stable, smoothing out the overall ride.

How I tested it

To test this idea, I focused on a “Trend Long” strategy – essentially, an EA designed to buy into strong upward trends, often after a breakout, confirmed by signals on higher timeframes. This is a common and often profitable strategy. I ran this strategy across a wide range of assets: all 19 major FX currency pairs plus Gold (XAU). For each of these, I calculated two key things:

1. Trend Edge (Profit Factor - PF): This tells us how profitable a strategy is. A Profit Factor (PF) is simply your gross profit divided by your gross loss. If your PF is greater than 1, it means you’re making more money than you’re losing – a good sign!
2. Daily Correlation: This measures how much the daily movements of these pairs relate to each other. Do they tend to move together, or in opposite directions, or are they completely independent? The goal was to find pairs with a good trend edge and low correlation, then combine them into a “decorrelated core.”

What happened?

Sometimes, research throws up a surprise that completely changes your perspective. This was one of those times!

The Big Discovery

My testing revealed something absolutely critical:

• Only 8 out of the 19 currency pairs showed a positive trend edge (PF > 1). In other words, only these 8 pairs were consistently profitable with our Trend Long strategy.
• And here’s the kicker: 7 of these 8 profitable pairs were JPY crosses! That’s USDJPY (PF 1.56), GBPJPY (PF 1.25), EURJPY (PF 1.21), AUDJPY (PF 1.19), CHFJPY (PF 1.15), NZDJPY (PF 1.08), and CADJPY (PF 1.01). The 8th profitable asset was Gold (XAU) with a PF of 1.34.
• Crucially, none of the other major non-JPY pairs (like EURUSD, GBPUSD, EURGBP, GBPNZD, etc.) showed any trend edge. They simply weren’t profitable with this trending strategy. What does this mean? It means that the very assets that do exhibit strong, profitable trends in FX – the ones you’d want in your “decorrelated core” – are overwhelmingly the JPY crosses and Gold. And these assets are inherently highly correlated with each other! When the Yen strengthens or weakens significantly, it affects all JPY crosses. Similarly, Gold often moves in relation to the USD, which impacts USDJPY. In essence, we discovered that “trending currencies” are the correlated cluster. There simply aren’t any decorrelated trending currency pairs in the FX market. It’s like looking for a unicorn: a consistently trending FX pair that moves independently of all the other trending FX pairs. They just don’t exist.

The Correlation-Minimizing Experiment

Despite this revelation, I tried to build a basket of strategies that did minimize internal correlation as much as possible, just to see what would happen. We managed to slightly reduce the internal correlation of our chosen basket from 0.21 to 0.16. That’s a small improvement, right? Unfortunately, the results were counterproductive:

• Drawdown worsened! It jumped from -21.7% to -24.8%.
• Efficiency (r/DD) also decreased! This metric, which measures your return per unit of drawdown (how much “bang for your buck” you get considering the risk), fell from 8.85 to 7.63. Why did this happen? To reduce the overall correlation, I had to swap out one of the more stable trending pairs (like EURJPY) for another (like AUDJPY) that, while less correlated, had a higher individual drawdown. The tiny benefit we got from lower correlation was completely overwhelmed by the increased risk from the individual asset. It seems our existing “robust5” basket, which was selected purely based on its strong individual trend edge, was already performing optimally, even with its inherent correlation of 0.21.

What I learned

This research led to a profound, structural conclusion about FX trending strategies: FX trend edges are found almost exclusively within the highly correlated cluster of JPY crosses and Gold. This means that trying to reduce system drawdown by diversifying within the FX market using trending strategies is fundamentally flawed. You simply cannot build a truly “decorrelated core” of profitable trending FX pairs because the building blocks – uncorrelated trending pairs – don’t exist. It’s like trying to build a house with only red bricks, but needing blue ones for stability; if blue bricks don’t exist, you’re stuck. So, where do we go from here? The path to genuinely reducing system drawdown isn’t through intra-FX diversification for trending strategies. The real solution lies in diversification across different asset classes. We’ve already started down this path! Our system (v1.1.0+) has incorporated other asset classes, like stock indices, which have a much lower correlation (around 0.16) with our FX core. These provide true “decorrelated sleeves” to our overall system. The only other remaining avenues for significant system DD reduction would be new, uncorrelated asset classes (like cryptocurrencies, but data availability is a current challenge). This study, while closing the door on the idea of a decorrelated FX core, isn’t a failure. Instead, it’s a crucial clarification. It confirms that the frontier for system improvement, particularly in drawdown reduction, lies outside the traditional FX market. It validates the direction we’re already heading in with our multi-asset approach (v1.4.1), pushing the boundaries of what’s possible in algorithmic trading.