Correlation table I check before stacking positions

[FXS London Scalper]

I sometimes have signals on two pairs at once. If they're highly correlated, taking both doubles the risk on one idea.

Every Monday I update a small table: the correlation between 15-minute returns across the last 20 days for EURUSD, GBPUSD, USDCHF, AUDUSD and EURGBP.

Typical values:

• EURUSD and GBPUSD: around 0.7.
• EURUSD and USDCHF: around minus 0.8.
• EURGBP with either: much lower, and it changes week to week.

Rule: if two signals have correlation above 0.6 in the same direction (adjusted for sign), I take one at full size or both at half.

It's calculated in a simple Python script from the broker's history. Has anyone found correlation on short timeframes too unstable to use?

[Amanhasnoname]

I kept a table like this for most of last year and found the 20 day window fine for the majors but close to useless for anything involving the pound on its own news. EURUSD and GBPUSD sat between 0.65 and 0.8 almost every week, so treating them as one idea was always the right call. What caught me out were the weeks around Bank of England meetings, when the reading dropped to 0.3 or so and then snapped back the week after.

My fix was to add a second, shorter column using only the last five days. If the two numbers disagree by more than about 0.25, I treat the pair as unstable and assume the higher figure for sizing. That way a temporary decoupling never tricks me into running double risk.

The other thing I learned is that correlation tells you nothing about the size of the move. AUDUSD and EURUSD can be well correlated while AUD moves twice as far, so I also compare the average 15 minute range before splitting size evenly.