Selecting Pairs for Mean Reversion with Distance and Return Metrics

Summary

This article focuses on selecting stock pairs for statistical arbitrage. It argues that finding pairs whose prices reliably diverge and reconverge matters more than the details of hedge-ratio estimation or other implementation models. Historical correlation and rolling cointegration estimates may be unstable and do not directly establish that a spread will be tradable. The relevant question is whether the pair stays related and whether deviations tend to close.

The proposed screening process begins with structurally related companies, such as firms in the same industry. It then measures how close their normalized price series have remained and how much a frictionless mean-reversion simulation would have earned. Combining these measures can distinguish persistent co-movement from pairs whose apparent profitability came from a single convergence. The author says the measures should be checked for persistence in later periods rather than assumed to generalize. No numerical results or detailed formulas are provided in the text, and the historical metrics are only selection signals, not proof of future returns. Capital limits, costs, and the small number of pairs a trader can hold also constrain implementation.

Key ideas

• Pair selection should focus on tradeable divergence and convergence behavior rather than statistical significance alone.
• Historical correlation and rolling cointegration coefficients may be unstable and weak guides to future mean reversion.
• Distance between normalized prices and simulated mean-reversion returns capture complementary pair characteristics.
• Starting with companies that share an industry can reduce the risk of pairing unrelated assets.
• Selection metrics need to demonstrate persistence in later periods before they are trusted.
• Capital, transaction costs, and limited trading capacity constrain how many selected pairs can be used.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.