Pairs Trading

The Market-Neutral Structure

Pairs trading holds a long position in one instrument and a short position in a related one simultaneously, sized so the combined position is close to market-neutral: a broad market move that lifts or drops both legs together washes out, leaving only the change in the spread between them as the source of profit or loss. The bet is not a directional view on either instrument — it is a view that two related instruments have drifted apart from their normal relationship and will converge again. Because the market-wide component is largely hedged away, a pairs trade isolates a narrower and more specific risk than an outright long or short position, which is both its appeal and its limitation: it can only make money on the relationship reverting, not on the broader market moving in either direction.

Correlation Is Not Cointegration

Two instruments can be highly correlated — moving in the same direction most of the time — without being cointegrated, and that distinction is the entire basis for whether a pairs trade is statistically sound. Correlation describes how two return series co-move day to day; it says nothing about whether the price levels themselves share a stable long-run relationship. Cointegration is the stronger, more useful property: the spread between the two instruments' price levels is itself stationary, wandering around a fixed mean rather than drifting indefinitely. A correlated-but-not-cointegrated pair can still see its spread widen without limit, because nothing anchors the two prices to a common level over time — which is why the spread, not the correlation coefficient, is what a pairs trade actually bets on.

Half-Life: Sizing the Holding Period

Once a pair is confirmed cointegrated, the half-life of its spread — the expected time for a deviation to close halfway back to its mean — gives a practical estimate of how long a position should take to work, covered in full in Closelook's Half-Life and Hurst Exponent entries. A short half-life supports a tighter holding period and tighter risk controls; a long half-life means the same statistical edge requires far more patience and capital efficiency to trade. Closelook's cointegration monitor tracks a matrix of pairs alongside their current half-life and Hurst readings, so a stretched spread can be read together with an estimate of how long convergence might reasonably take, rather than in isolation. A spread that is two standard deviations wide with a short half-life is read very differently from the same stretch on a pair whose half-life has been drifting longer — the first supports a defined-horizon position, the second is closer to a research note than a tradeable setup.