Correlation

### What the number actually is Correlation between two pairs is the Pearson correlation coefficient computed over a window of closing prices. It runs from +1 (they move in lockstep) through 0 (no linear relationship) to -1 (they mirror each other). [OANDA's correlation tool](https://www.oanda.com/bvi-en/lab-education/tools/correlation-tool/) does exactly this, using closing mid prices, and offers windows from 1 hour, 4 hours and 24 hours up to 1 week, 1 month, 3 months, 6 months and 1 year, while its correlation matrix is fixed to a 1 year lookback. ### Why USD-quoted pairs run together EURUSD and GBPUSD are not mysteriously linked. They share a leg. The dollar is the quote currency in both, so when the dollar moves, both prices move, in the same direction. The stronger the day's driver is a pure dollar story, the tighter that relationship gets. Flip the dollar to the other side of the fraction and the sign flips with it, which is why EURUSD and USDCHF have historically run strongly *negative*: same dollar, opposite side. So a strong positive reading between EURUSD and GBPUSD is not a coincidence waiting to be discovered. It is arithmetic, and it holds for exactly as long as the dollar stays the dominant driver. ### The lookback window is the whole argument Correlation is not a property of a pair. It is a property of a pair **and** a time window. The same two pairs can read strongly positive on a 1 hour window and much weaker on a 3 month one. That is precisely why a serious tool makes you pick the window before it gives you a number. So when someone says "those two are correlated" as though it were a fixed constant of the market, they have skipped the only question that matters: over what period? If you size your risk off a one-year coefficient while executing on a 15M chart, you have measured something that has almost nothing to do with the trade in front of you. ### Correlations break, and they break at the worst possible time A correlation is a description of the past. It holds while the thing driving it holds, and it stops the moment that thing stops. The cleanest illustration is the Swiss franc. For years the Swiss National Bank enforced a minimum exchange rate of CHF 1.20 per euro, and every CHF relationship was quietly shaped by that floor. On 15 January 2015 the SNB [announced it was discontinuing the minimum rate](https://www.snb.ch/en/publications/communication/press-releases/2015/pre_20150115) and cutting its sight deposit rate to -0.75%. The regime that had been generating those stable relationships was gone in a sentence, and the relationships went with it. Central banks moving in different directions does the same thing more slowly: the legs pull apart, and the historic coefficient keeps reassuring you right up until it doesn't. ### How it bites in practice - Two positions, each risking 1%, on two heavily positive-correlated pairs is not two independent 1% bets. It is much closer to a single 2% bet on one theme. Your risk report says diversified. Your equity curve knows better. - Correlation tells you about **direction**, not **magnitude**. Two pairs can move together almost perfectly and one still travels twice as far. Check ATR before you assume equal damage. - A correlated pair is not confirmation. If the second chart is effectively the same trade, it cannot independently confirm the first one. That is circular, and it feels like conviction, which is what makes it dangerous. If you want one clean reference for the dollar leg that sits behind most of this, that is what DXY is for.