Cointegration means that two or more price series, each of which wanders without a fixed average, move together in the long run so that a particular combination of them stays stable. It is the statistical basis of pairs trading.
The idea in plain terms
Imagine two stocks whose prices both drift up and down over the years, but the gap between them keeps returning to a typical level, like a dog on a lead walking beside its owner. Each stock alone is unpredictable, but their combination is not. For example, the combination price of A - 1.2 x price of B may keep oscillating around zero. That stable combination is called the spread.
Cointegration compared with correlation
| Correlation | Cointegration | |
|---|---|---|
| What it measures | Whether returns move in the same direction | Whether price levels stay tied together |
| Based on | Changes (returns) | Levels (prices) |
| Can the gap grow without limit? | Yes, even with high correlation | No, the gap tends to come back |
Two stocks can have a high correlation of daily returns and still drift apart in price over time, which is why correlation alone is not enough for a pairs trade.
How it is tested
- Check that each price series is non-stationary, meaning it has no fixed average.
- Estimate the combination, for example by regressing one price on the other.
- Test whether the resulting spread is stationary, for instance with the Engle-Granger two-step method or the augmented Dickey-Fuller test.
Cautions
- Relationships break. Cointegration found in the past may not continue, for example after a merger or a change in a business.
- Estimation leaks the future. Estimating the hedge ratio on the whole data set and applying it to the past is look-ahead bias.
- Many pairs, some false. Testing thousands of pairs will produce some that look cointegrated by chance. See overfitting and correlation vs causation.
Cointegration is the basis for statistical arbitrage and for strategies built on mean reversion.
Frequently asked questions
What is cointegration?
It is a statistical property of two or more non-stationary series whose combination is stationary, meaning they may wander individually but stay tied together over the long run.
What is the difference between correlation and cointegration?
Correlation looks at how returns move together. Cointegration looks at whether price levels stay linked, so the gap between them tends to return to a typical value.
How do you test for cointegration?
Common tests are the Engle-Granger two-step test and the Johansen test. They check whether the spread between the series is stationary.
Why is cointegration important in pairs trading?
It suggests the spread between two stocks will revert to its average, which is the condition a pairs trade needs to profit from the gap closing.