How We Select Currency Pairs for an Automated Forex Strategy
When developing an automated Forex strategy, one of the first questions is usually: Which currency pairs should the EA trade?
At first glance, this looks like a relatively simple problem. In the previous article, I discussed why I believe that an automated trading system should be viewed as more than just an Expert Advisor. That article looked at the role of execution, infrastructure, costs and risk management. Here I would like to focus on one specific part of that equation: currency-pair selection.
Run the strategy across EURUSD, GBPUSD, EURAUD, AUDCAD, USDJPY and a few other major and cross pairs. Compare the results. Select the best performers. But after many years of developing and running automated systems, we have learned that this approach can be surprisingly misleading. The pair with the highest historical return is not necessarily the pair that is best suited to the strategy. And adding more currency pairs does not automatically make a system more diversified.
The real question is different: Does this particular currency pair provide the market conditions that the strategy needs?
That question has become an important part of our development process.
Starting with the strategy, not the currency pair
One mistake we made early in our research was to think about currency pairs first and strategy second.
The logic was simple: "This is a Forex EA, so it should probably trade as many liquid currency pairs as possible."
It sounds reasonable. But different strategies exploit different characteristics of the market. A trend-following system may benefit from instruments that regularly produce extended directional movements. A mean-reversion strategy may prefer a completely different environment. A short-term strategy may be particularly sensitive to spread, execution and intraday volatility. Therefore, there is no universal list of "best Forex pairs" for automated trading. There are only pairs that are more or less suitable for a particular methodology. This changed our approach considerably.
Instead of asking: "Which pairs produce the highest backtest?" we started asking: "What characteristics does our strategy need in order to operate efficiently?"
Only after answering that question does the process of selecting instruments become meaningful.
What do we actually look for?
There is no single metric that determines whether a currency pair is suitable. We normally look at several characteristics simultaneously.
1. Trading costs
This is probably the most obvious factor, but it is often underestimated.
If an automated strategy opens and closes positions relatively frequently, the cost of each transaction becomes an important part of its mathematical expectation.
The relevant costs may include:
- spread;
- commission;
- swap;
- slippage;
- and other execution-related costs.
A pair may look attractive before transaction costs and significantly less attractive after them. This is especially important for short-term systems. A difference of a fraction of a pip may appear insignificant on an individual trade. Across hundreds or thousands of trades, it is not insignificant anymore. For this reason, we do not evaluate a currency pair only by its gross trading performance. We want to understand the net trading environment.
2. Volatility
Volatility is another important characteristic. A strategy needs a certain amount of price movement to generate opportunities. But "more volatility" does not automatically mean "better." Excessive volatility can also increase:
- slippage;
- spread expansion;
- stop-out probability;
- execution uncertainty;
- and the size of individual losses.
What we are looking for is not simply the most volatile instrument. We are looking for a volatility profile that is compatible with the strategy's holding time, entry logic and risk management. This distinction is important. An EA does not trade volatility in isolation. It trades a specific pattern of price movement within a specific market environment.
3. Liquidity
Liquidity is closely connected with execution quality. A strategy may identify an attractive entry at a particular price, but the actual execution can be different. This matters much more for short-term strategies than for systems that hold positions for days or weeks. The shorter the expected trading opportunity, the less room there may be for execution costs and slippage. This is one of the reasons why we pay attention not only to the historical behavior of a currency pair but also to the practical trading conditions available for that instrument.
In other words:
A theoretically attractive market is not necessarily a practically attractive market.
4. Trading session
Currency pairs do not behave identically throughout the day. The market environment changes as different financial centers become active. Liquidity, volatility and spread can all change depending on the time of day. For an automated strategy, this means that "EURUSD" is not really one single trading environment. EURUSD during one session can behave differently from EURUSD several hours later. This is why trading hours are part of our instrument research.
Sometimes the question is not: "Can the strategy trade this pair?" but: "Can the strategy trade this pair during the particular hours when its statistical edge exists?"
That is a much narrower — and more useful — question.
5. Correlation
This is where the idea of diversification becomes particularly interesting. Suppose an EA trades six currency pairs. It is tempting to assume that six pairs automatically mean six different sources of opportunity. But that is not necessarily true. Currency markets are highly interconnected. Several pairs may respond to the same underlying factors. For example, adding several instruments with strong exposure to the same currency can increase concentration rather than reduce it.
So when we evaluate a potential new pair, we do not only ask: "Does it make money?". We also ask: "What does this pair add to the existing portfolio?"
If two instruments produce very similar behavior, the second one may add much less diversification than its name suggests.
The EURAUD lesson
Our own experience with EURAUD is a good example of why this process matters. The original SCR_EURAUD approach became relatively focused on EURAUD and AUDCAD. These instruments were not selected simply because they happened to produce the highest backtest numbers. Over time, we found that their market behavior was compatible with the characteristics of our trading methodology and the execution environment we were targeting. That became the foundation of our focused approach.
But eventually we wanted to investigate another question: Could the same general methodology benefit from a broader selection of instruments?
That research eventually led to the development of SCR_EURAUD_advanced. The purpose was not simply to add more pairs and increase the number of trades. Instead, we wanted to investigate whether a broader portfolio could provide complementary trading opportunities and potentially change the overall risk profile. This distinction is important.
More trades are not the objective. The objective is better portfolio behavior.
Why we don't simply select the top five backtested pairs
Imagine that we run an EA across 30 currency pairs. Suppose five of them show the highest historical profit. It would be tempting to simply select those five.
But this introduces several problems.
Problem #1 - Overfitting
The best historical performers may simply be the instruments that happened to match the particular market conditions of the tested period. A strategy can be unintentionally optimized to the past.
Problem #2 - Different execution costs
A pair with excellent gross performance may have significantly higher trading costs. After spread, commission and slippage, the ranking can change.
Problem #3 - Correlation
The five best performers may have highly correlated behavior. Instead of diversification, we may simply create five versions of the same exposure.
Problem #4 - Market regime
Currency pairs can behave differently during different market regimes. A pair that was excellent during one period may become much less attractive when volatility, monetary policy expectations or market correlations change. For these reasons, we try to avoid treating historical ranking as the final decision.
Backtesting is a filter, not a final answer
This is perhaps the most important distinction. We use backtesting extensively. It is an excellent tool for eliminating obviously unsuitable instruments and identifying areas that deserve further research. But we do not consider a backtest sufficient evidence for deploying a pair in live trading.
Our general process is closer to: Backtest → robustness testing → forward testing → live observation → evaluation
Each stage answers a different question.
Backtest
Does the strategy have any statistical potential on this instrument?
Robustness testing
Does the result survive reasonable changes in parameters and assumptions?
Forward testing
Does the behavior continue outside the optimization sample?
Live observation
How does the strategy interact with actual spreads, execution and market liquidity?
Long-term evaluation
Does the instrument continue to make sense as part of the overall portfolio? This process is slower than simply choosing the five best backtests. But automated trading is not a competition to produce the fastest backtest. The objective is to build something that has a reasonable chance of surviving the transition from historical data to the real market.
What about adding new currency pairs?
We continue to test new instruments. Some are rejected almost immediately. Some look promising in historical data but fail during forward testing. Others remain interesting for a long time before we have enough evidence to make a decision. And sometimes a pair that initially looks mediocre becomes interesting after we change the trading hours, execution conditions or portfolio composition. This is another reason why we don't consider the selection of currency pairs to be a one-time development task. It is an ongoing research process. Markets change. Transaction costs change. Liquidity changes. Correlations change. And the behavior of an instrument relative to a particular strategy can change as well.
The difference between "more pairs" and "better diversification"
This is a distinction I think is worth emphasizing. Consider two portfolios.
Portfolio A
Six currency pairs that all have strong exposure to similar market factors.
Portfolio B
Three currency pairs with more independent behavior.
Portfolio A contains twice as many instruments. But Portfolio B may actually provide better diversification. The number of symbols is therefore a poor measure of diversification by itself.
For automated trading, I think it is much more useful to ask: How does each additional instrument change the behavior of the entire portfolio?
- Does it reduce concentration?
- Does it introduce a new source of opportunity?
- Does it improve the distribution of returns?
- Does it increase drawdown?
- Does it increase execution costs?
- Does it create additional exposure to the same underlying currency?
These questions are more important than simply counting the number of pairs.
What ten years of development changed for us
When we started developing automated strategies, we were naturally focused on finding profitable entry conditions. Today, our thinking is much more portfolio-oriented. We still care about the entry logic. But we also care about everything around it. The instrument is part of the strategy. The trading session is part of the strategy. Execution costs are part of the strategy. Liquidity is part of the strategy. And the relationship between instruments is part of the strategy.
This is one of the biggest lessons we have learned from working with SCR_EURAUD, SCR_EURAUD_advanced and the SCR_NightScalper technology behind them.
The EA may contain the trading logic.
But the market environment determines how that logic is expressed in practice.
Final thoughts
There is no universally "best" Forex currency pair. There is only a currency pair that may be more or less suitable for a particular strategy, during particular market conditions, with particular execution characteristics.
This is why we have become increasingly skeptical of simple statements such as: "This EA works best on EURUSD."
or: "These are the five most profitable pairs."
The more useful question is: Why does this strategy work on this instrument, and under what conditions?
Once you start asking that question, currency-pair selection becomes much more than a backtest ranking exercise. It becomes part of strategy design itself. And that is the approach we continue to use today.
The goal is not to trade more markets. The goal is to find the markets where the strategy makes the most sense.
