Correlation Confound (Part 3)

Correlation confound #1 dealt with portfolio diversification.  Today I will discuss correlation confound #2, which has to do with pairs trading.

To design a pairs trade, the first step is said to be finding stocks that are highly correlated. This could involve businesses in the same industry or sub-sector, index ETFs like QQQ and SPY, highly-correlated futures pairs, etc.

The next step is to monitor a chart showing price ratio between the two markets. When the price ratio diverges far enough, I want to go long (e.g. sell an OTM put spread) the relatively cheap market and go short (e.g. sell an OTM call spread) the relatively expensive market. Since the two markets generally move together (i.e. highly correlated), when they move apart I should expect mean reversion where they come together again.

Correlation confound #1 also applies to pairs trading.  Just because markets have been correlated in the past does not mean they will be correlated in the future.  That could cause problems for pairs trades.

Correlation confound #2 is something called cointegration.  Correlation measures the tendency of two variables to move together but it’s not guaranteed that they’ll stay close to each other. Look at the following charts of the gold and silver markets:

Believe it or not, in both charts these markets have the same statistical correlation of +0.75!  If you took on a traditional pairs trade when these markets were far apart with the expectation that they would get closer together, you might be sorely disappointed with the graph on the right but happy with the graph on the left.

Cointegration is a measure of how well assets are tied together. Two correlated variables are allowed to wander arbitrarily far apart but a cointegrated variable is not.  Cointegrated variables are expected to stay parallel to one another since the difference between the two will be corrected over time.