Cointegration vs correlation: why pairs traders care about the difference

Correlation and cointegration both describe two prices that seem to move together, so they are easy to confuse. For pairs trading the difference matters a great deal.

Correlation: moving in the same direction on the same days

Correlation measures whether two stocks tend to rise and fall on the same days. Two bank stocks might both be up on a good day for banks and both down on a bad one. A high correlation says their daily moves line up.

What correlation does not say is whether the distance between the two prices stays put. Two stocks can go up on the same days while one rises a little faster every time. Over a year the gap between them grows steadily, even though their daily moves were highly correlated.

Cointegration: a gap that keeps coming back

Cointegration is about the gap itself. Two prices are cointegrated when a suitable combination of them — for example, one stock minus a fixed multiple of the other — keeps wandering around a stable level instead of drifting away. The gap stretches and shrinks, but it is pulled back toward its usual level.

That pull is what a pairs trade relies on. If the gap has no tendency to come back, a stretched gap is just a gap, and waiting for it to close is a hope rather than a measured tendency.

A dog-walker picture

A common way to picture it: a person walking a dog on a lead. Both wander around the park, sometimes in unpredictable directions. But the lead keeps the dog from getting too far from the person. Their distance apart stretches and shrinks, and keeps returning to roughly the length of the lead. That distance is cointegrated.

How the model uses it

Before a pair can be published, PairsCopilot tests whether its gap has historically behaved like this over several years of data, and re-tests it every night. A pair that stops passing is benched and publishes nothing until it passes again. Past behaviour is evidence, not a guarantee — relationships can and do break.