Cointegration

Traders often confuse cointegration with correlation, but they measure fundamentally different things. Correlation measures how two assets move together at a single point in time — two stocks can be highly correlated for a period, then decouple completely. Cointegration, by contrast, tests whether a linear combination of two non-stationary price series produces a stationary series over the long run. Two cointegrated assets may diverge substantially over days or weeks, but their spread (or price ratio) reliably reverts to a mean. This long-run equilibrium relationship is the mathematical prerequisite for pairs trading. Without cointegration, a mean-reversion trade on a diverging pair has no statistical reason to close — making it speculation rather than a quantified edge.

The two primary tests for cointegration are the Engle-Granger two-step procedure and the Johansen test. Engle-Granger is simpler: regress one price series on the other, then test the residuals for stationarity using an ADF (Augmented Dickey-Fuller) test. If residuals are stationary (ADF p-value below 0.05), the pair is cointegrated. The Johansen test is more powerful and can detect multiple cointegrating relationships among several assets simultaneously, making it suitable for basket trades. Both tests require a sufficient data history — typically two to five years of daily prices — to achieve reliable statistical power. Results should be validated on out-of-sample data to avoid overfitting to historical noise.

Knowing that a spread is cointegrated is not enough to trade it profitably — you also need to know how fast it mean-reverts. Half-life, derived from the Ornstein-Uhlenbeck process, estimates how many days it takes for a deviation to close by 50%. A half-life of 3 days requires frequent monitoring and rapid entry/exit; a half-life of 60 days ties up capital for extended periods. Most active pairs traders target half-lives of 5 to 30 days as a practical sweet spot. When cointegration is strong but half-life is too slow, the opportunity cost of capital outweighs the statistical edge. Educational note: these calculations are for informational purposes — actual trade profitability depends on execution costs, regime changes, and risk management.

Cointegration is not permanent. Corporate events such as mergers, spinoffs, major capital structure changes, or regulatory shifts can destroy the long-run equilibrium that once linked two assets. Sector rotation and macroeconomic regime changes can also weaken formerly strong cointegration relationships. Robust pairs trading systems continuously re-test cointegration on rolling windows and remove pairs whose p-values have degraded. When the spread widens beyond three or four standard deviations and does not show signs of reversion within the expected half-life, a prudent risk management rule is to exit the position, treating the breakdown as a stop-loss event rather than a deeper opportunity.

While pairs trading is the most common application, cointegration has broader uses in quantitative finance. Index arbitrageurs test cointegration between ETFs and their underlying baskets to identify temporary NAV dislocations. Fixed-income traders apply cointegration to yield-spread strategies across bonds of similar credit quality but different maturities. Crypto quant funds use cointegration between spot and perpetual futures to run funding-rate-neutral strategies. Factor investors test whether factor spreads (value vs. growth, for example) are cointegrated to determine if factor tilts will eventually revert to long-run equilibria. In all these contexts, the underlying logic is the same: find a statistically reliable long-run relationship and trade the short-term deviations.