Statistical Arbitrage (Pairs)
Trade the spread between cointegrated pairs when it deviates from the equilibrium; exit at mean reversion or stop.
Thesis (edge)
Two cointegrated assets have a stationary spread. When the spread moves to an extreme (e.g. z > 2), short the outperformer and long the underperformer; exit when spread reverts.
Where it works (regimes)
Works in mean-reverting regimes. Fails when cointegration breaks (structural break, M&A, sector shift). Requires ongoing monitoring.
Signals
- Spread = ( P_1 - \beta P_2 ); ( \beta ) from cointegration.
- Z-score of spread; enter when |z| > 2, exit when |z| < 0.5 or stop.
Portfolio construction
Dollar-neutral or beta-neutral. Size by spread vol. Max exposure per pair.
Risk model
Tail: one leg gaps (earnings, news). Stress: both legs fall but spread widens. Use position limits and stop per pair.
Costs & implementation
High turnover; two legs. Slippage and spread critical. Use liquid names only.
Failure modes
Overfitting pair selection; cointegration breakdown; execution cost eats edge.
Our Notes & Suggestions
Walk-forward cointegration; avoid look-ahead in hedge ratio. Use robust spread estimation (e.g. rolling). Limit number of pairs to control complexity.
Our Notes & Suggestions
See the "Our Notes" subsection in the body above for practical guidance, gotchas, and best practices. Always validate regime assumptions and transaction cost assumptions before scaling.
Implementation Checklist
- Identify candidate pairs (cointegration test)
- Estimate hedge ratio (OLS or rolling)
- Trade spread z-score; define entry/exit
- Monitor cointegration; exit if broken