Correlation caps for gold specialists: beginner mistakes

[LondonScalper]

When people start scalping gold seriously, they often add silver, then a gold miner index, then AUDUSD because "it's a gold currency". Within a month they're running what is really one big position spread across four tickets. I did exactly this around 2012.

The mistakes I see most often, and made myself:

Counting gold and silver as separate risks. On my data they move together far more than they move apart during London hours. If I'm long both, I'm long one thing at double the size.

Ignoring the dollar. A gold long and a EURUSD long are both, at least partly, dollar shorts. On a strong US data print both can hit their stops in the same minute.

Assuming the correlation is stable. It isn't. During some risk-off episodes gold and the dollar have both rallied, which breaks the usual relationship. A cap based on last month's correlation can be wrong this month.

Sizing each ticket as if it were the only one. If my normal risk per trade is 0.5%, two correlated tickets at 0.5% each is 1% at risk on one idea.

What I use now is a simple cap: no more than 0.75% of the account at risk across gold, silver and any dollar pair in the same direction. If I want a second position, the first gets reduced.

It's crude, but crude and followed beats precise and ignored.

[Shadow Trader]

Your cap is close to mine, and the one ticket that keeps slipping past people is USDJPY. A short there is a dollar short and often a risk off trade, so pairing it with a gold long during a tense London morning is the same bet twice, yet it rarely gets counted.

I weight tickets by a rolling correlation instead of a flat yes or no. Using M15 returns during London hours over the last 20 sessions, anything above 0.7 with gold counts fully toward the cap, 0.4 to 0.7 counts at half, and below that it stands alone.

The weakness is exactly what you describe about stability. Twenty sessions can look calm right up to a data print that drags everything into one move, so on CPI, payrolls and FOMC days I override the weights and count every dollar ticket at full weight.

Silver adds its own trap. Its moves on news are roughly twice as violent as gold, so a 0.5% silver ticket behaves like a bigger position the moment stops start to slip.

Do you count opposite direction tickets as reducing the total, or ignore them? I ignore them, because hedges inside one idea tend to fail together on news.

[Shadow Trader]

LondonScalper, the cap held for me on paper but leaked on data days through slippage. On three US data mornings last quarter my book was capped at 0.75%, yet the realised losses came to 1.05%, 1.10% and 0.95% because gold stops filled well past the level.

Since then the planned cap drops to 0.6% on those mornings. That leaves room for about 25% slippage on top before the real loss passes 0.75%.

The reduce the first ticket rule has a cost of its own. Cutting an open gold position in half pays the spread again, so when I already expect a second idea I open both at half size from the start.

AUDUSD turned out to be the least gold like of the four in my numbers. Over the last 20 London sessions its correlation with gold sat mostly under 0.4, so the gold currency story was weaker than the dollar link.

Miners are the one I would cap hardest. They open later and gap on equity sentiment, so the risk on them is hard to measure with the same stop logic.

Does your cap reset at the end of London, or carry into New York for anything still open? Mine carries, which keeps afternoon tickets small.