This paper provides a comprehensive review of existing pair-selection methods used in statistical arbitrage.

The analysis highlights how pair selection remains the key determinant of strategy performance, with approaches ranging from simple correlation and distance measures to cointegration and more advanced machine-learning techniques.

A central conclusion is that pairs trading profitability has weakened over time and become increasingly regime-dependent.

Performance is now far more sensitive to transaction costs, execution quality, and changing market structure, suggesting that implementation efficiency is as important as signal generation.

Want more research breakdowns like this? Subscribe for weekly insights