For pairs trading and stat-arb, cointegration is often used in addition to correlation.
While correlation measures short-term return co-movement (which can drift apart indefinitely), cointegration tests whether a linear combination of two asset prices forms a stationary, mean-reverting spread. Cointegrated assets move together long-term, so when their price spread widens, it creates an entry point for mean-reversion trades.
For anyone testing this, Portfolio Visualizer has a free cointegration tool: Portfolio Visualizer Cointegration. It's also straightforward to set up in Python using statsmodels (or with AI assistance). Claude or Google's Antigravity can get members set up if they want to try this.
It really is essential for pairs trading at least. It is always used for pairs-trading, professionally. Also as part of a workflow, I have seen a dendrogram (hierarchical clustering) used as a screen to find candidates for pairs with both correlation and cointegration (in an online Coursera course).
From Portfolio Visualizer:
"This tool allows assets to be tested for cointegration using the Augmented Dickey-Fuller Test. Cointegrated assets typically have a financial or economic relationship that prevents them from diverging and the price difference tends to be mean reverting."