Correlation and Hidden Risk: Why Several Open Trades Can Be One Bet

Correlation is a number that tells you how often two prices move in the same direction. If you open several trades at once without looking at it, your account can look diversified while really betting on the same thing several times over. This article explains how the number is calculated, why it has to come from returns, and works through the combined risk of a portfolio step by step.

How to read correlation

Correlation (written ρ, “rho”) sits between −1 and +1.

ρ	Meaning
+1	Always move the same way, in a fixed proportion
+0.7 or more	Mostly move the same way
Near 0	No straight-line link. One rising tells you nothing about the other
−0.7 or less	Tend to move in opposite directions
−1	Always move in opposite directions

The formula is ρ = Cov(A, B) ÷ (σ_A × σ_B): the covariance of the two, divided by each one’s standard deviation (σ, its normal size of swing). You don’t need to compute it by hand. Spreadsheets have a CORREL function.

Two common misreadings:

• ρ tells you direction, not size. ρ = 0.8 does not mean B moves 80% as much as A. It can move more or less
• ρ is an average over the period you measured, not a permanent property. Three months and one year can give different answers

Measure returns, not price lines

Take two prices that are both trending up and compute their correlation directly, and the number will always come out high, even if the two are unrelated. What you have measured is “both went up”, not “they went up on the same days”.

We generated two made-up series of 60 bars. Each one is random and independent of the other, with nothing linking them. We only gave both an upward drift.

1. Series A starts at 100 and ends at 135.33 (+35%). Series B starts at 50 and ends at 57.06 (+14%)
2. Correlation of the 60 prices: 0.95. They look tightly linked
3. Convert to per-bar returns = (this bar’s price ÷ previous bar’s price − 1) × 100%, giving 59 values per series
4. Correlation of the returns: 0.04, practically zero
5. Check: count the bars where both moved the same way. It is 26 of 59, less than half, consistent with a value near 0

The profit or loss of an open trade comes from the change in price each day, not from the price level. So the number that tells you “will these lose together?” is the correlation of returns, and only that.

Relationships traders watch

All of these are tendencies seen often in the past, not rules. The actual values drift over time, so measure them yourself on recent data before relying on them.

• EURUSD and USDCHF tend to move opposite, because the dollar sits on opposite sides of the two pairs. When the dollar strengthens, EURUSD tends to fall and USDCHF tends to rise. The euro and the Swiss franc themselves often move alike
• Pairs with USD as the second currency, such as EURUSD, GBPUSD and AUDUSD, tend to move together because the dollar is the common part. One US headline can move the whole group
• Gold and the dollar tend to move inversely. Gold is priced in dollars, so a weaker dollar tends to lift gold. But there are periods when both rise together, for example when money looks for safety in both. For more on rates and gold, read How interest rates move currencies and gold and the XAUUSD gold trading guide

One more point that matters: when markets panic, the correlation between many assets tends to jump toward 1. Things that normally have little to do with each other fall together. Diversification helps least on exactly the day you need it most.

Three trades, maybe one bet

Say you open these at the same time:

• Buy EURUSD = buy euros, sell dollars
• Buy GBPUSD = buy pounds, sell dollars
• Buy XAUUSD = buy gold, paid for in dollars

What the three share: they make money when the dollar weakens and lose when it strengthens. If each one risks 1%, as in Risk 1% per trade, a single US release that sends the dollar sharply higher can hit all three stops, costing 3% in a few minutes.

A simple check is to write down your net position in each currency. In this example USD is minus three units (you sold dollars three times) and each of the other three is plus one. Whenever a currency builds up more than one unit in the same direction, treat those trades as one group.

Direction matters too. Buy EURUSD with Sell USDCHF means selling dollars twice, so the risk stacks. Buy EURUSD with Buy USDCHF means the two positions largely cancel, both the losses and the profits.

Computing the combined risk

The combined risk (σ_p) of several positions is:

σ_p² = Σ w_i²σ_i² + 2Σ ρ_ij w_i w_j σ_i σ_j

w_i is the position size and σ_i the swing of each instrument. To keep it simple we use s_i = w_i × σ_i, the swing of that position’s profit and loss in dollars (the standard deviation of its daily P/L). The formula becomes σ_p² = Σ s_i² + 2Σ ρ_ij s_i s_j.

Two positions, each s = 100 dollars

1. Σ s² = 100² + 100² = 20,000
2. Cross term = 2 × ρ × 100 × 100 = 20,000 × ρ
3. ρ = 1: √40,000 = 200, the same as simply adding them
4. ρ = 0.5: √30,000 = 173.2
5. ρ = 0: √20,000 = 141.4
6. ρ = −1: √0 = 0, they cancel completely

Three positions: buy EURUSD, buy GBPUSD, buy gold, each s = 100 dollars. Assume ρ for EURUSD and GBPUSD = 0.7, EURUSD and gold = 0.4, GBPUSD and gold = 0.3 (made-up figures for the demonstration, not values measured from the market).

1. Σ s² = 3 × 10,000 = 30,000
2. Cross terms = 2 × (0.7 + 0.4 + 0.3) × 10,000 = 28,000
3. σ_p² = 58,000 → σ_p = 240.8 dollars
4. If the three were unrelated (ρ = 0) it would be √30,000 = 173.2 dollars
5. Correlation makes the risk 240.8 ÷ 173.2 = 1.39 times the independent case. To bring it back to that level you would cut every position to 173.2 ÷ 240.8 = 0.72 of its size

Buy EURUSD with buy USDCHF. Assume ρ between these two positions = −0.9. That gives √(20,000 − 18,000) = √2,000 = 44.7 dollars. The swing almost disappears, and so does the profit you were hoping for.

A caution: σ_p is the swing on a normal day, not the worst case. On a day when price jumps, correlated trades can hit their stops together, so the worst case is still the sum of every trade’s risk.

Two layers of risk caps

Set a cap per trade and a cap per group of trades that move together, not just one layer.

Common example rules, to adjust for yourself:

• Per trade: no more than 1% of the account, with the lot size worked out from the stop distance as in Position size from stop distance
• Per group: trades whose return correlation is above roughly 0.7, or that hold the same currency in the same direction, together no more than 2–3%
• Several trades in one group: split the group cap. With a 2% cap and three trades, each risks 2 ÷ 3 = 0.67%
• Total open risk: add one more cap across all groups, sized so you can survive a day when everything falls together

The bands ρ above 0.7 = same bet, 0.3–0.7 = partly overlapping, below 0.3 = mostly separate are a rough rule of thumb for quick decisions, not hard lines. Measure on 60–100 days of daily returns and re-check every month.

If you add to a position along the way (Scaling in and partial closes), a new trade in the same direction on the same instrument has ρ = 1 with the old one. Always count it in the same group.

Diversifying across systems: what our tests showed

Besides spreading across instruments, you can spread across trading systems. Even on the same instrument, systems with different logic can have daily results that are only loosely related.

In our backtests on gold (XAUUSD) on the M1 chart, 2026 uses real tick data (January to August or September, depending on the test), and 2024–2025 use ticks simulated from 1-minute data.

• Five systems combined, each at its tested lot size: the combined profit equals the sum of the parts exactly, but the combined maximum equity drawdown was 2,800 dollars, while adding up each system’s own maximum drawdown gives 5,561 dollars. The worst stretches of each system did not arrive at the same time (on equity drawdown, read Equity vs balance drawdown)
• Two groups of systems with a daily-result correlation of 0.48 lost on the same day on only 24 of the 100 days the first group lost. Combined, their profit-to-drawdown ratio (measured on daily drawdown) rose from 18.3 to 26.8
• A breakout-following system and a pullback system, ρ = 0.64: the breakout system made money only in years when gold trended, lost in 2024, and adding it to the combined set did not improve the set
• Two systems with ρ = 0.94 were practically the same system, so we dropped one
• A five-system set on a 10,000-dollar account, with lots balanced so each system had a similar drawdown, returned +2,287 / +3,637 / +18,755 dollars in 2024 / 2025 / 2026. Positive in all three years, but most of the profit came from 2026 alone

Our takeaway: low correlation between systems really does reduce drawdown, but a system with ρ above about 0.6 mostly adds size, not variety. To read these numbers in a report, see The MT5 backtest report explained, and for why systems work only in some years, Market regime change.

All of these are backtest results. Past results do not predict future results, and they are not returns you should expect. Live results are usually lower than the test (Why live results differ from backtests).

Common mistakes

• Computing correlation from prices. Two trending series always look linked. Use per-bar percentage changes
• Counting trades instead of bets. Five trades that all sell the dollar are one bet, five times bigger
• Ignoring direction. With a negatively correlated pair, opposite trades (buy one, sell the other) stack the risk rather than cancel it
• Using an old correlation. A value measured last year may have changed, and it tends to jump toward 1 on the days markets panic
• Treating a hedge as free. Buying two pairs that move opposite cuts the swing, but the profit goes too, and you still pay spread and swap on every trade
• Adding look-alike systems. Systems with highly correlated daily results add losses in the same bad stretches. They don’t diversify