🏠
Guest Not signed in

Cointegration: when two non-stationary series share a stationary spread

Two stock prices each wander randomly. Their *combination* $X - \beta Y$ can be stationary, giving you a mean-reverting spread to trade. Engle-Granger's 1987 procedure makes the test mechanical.

Method · Cointegration Pairs Trading
Intro

Two stock prices look like random walks (unit-root, non-stationary), so neither is mean-reverting and you cannot fade them individually. But a linear combination $Z_t = X_t - \beta Y_t$ can be stationary for some $\beta$ — the cointegrating coefficient. When this happens the spread itself is the tradeable object: it wanders away from its mean, then snaps back, repeatedly. This is the workhorse of statistical-arbitrage pairs trading. The Engle-Granger (1987) procedure tests it in two steps: estimate $\beta$ via OLS on the levels, then test the residuals for stationarity via Dickey-Fuller. Pass: tradeable. Fail: the relationship was spurious, and you almost just lost money.

βœ“ Intro Β· expand
Independent · Legal