In the early days of paired trading, traders would group stocks into broad industry classifications, and every pair within that group was a candidate to trade. But you couldn’t trade just any pair, there had to be a set of filters applied to induce greater control over [...]]]> Which pairs to trade? That is the question.

In the early days of paired trading, traders would group stocks into broad industry classifications, and every pair within that group was a candidate to trade. But you couldn’t trade just any pair, there had to be a set of filters applied to induce greater control over return variability.

So the first filter was to determine how well correlated two stocks were. The correlation coefficient measures this on a scale of -1 to 1. Typically a pair with a correlation of at least 0.70 might be deemed an acceptable pair. The theory that highly correlated pairs will continue to correlate and provide a degree of predictability.

The next filter is how well, or to what degree of regularity the pairs move apart and then back together. We call these event moments. Event moments are times when the correlation becomes unwound to an extreme, usually due to an event; a market disruption, a political event, economic report, or major catastrophe. The most desirable outcome is a pair that diverges and then comes together again in a reasonable amount of time.

We measure a pairs ability to respond to events by examining its true range, a measure of volatility. With the true range we can see how often a pair is subjected to disruptive event moments and how well and how quickly it settles down. Further more, we can compare true range on the same timeline with the running correlation to see the relationship between volatility and correlation.