Statistical arbitrage finds groups of assets that should trade together and bets on convergence when they don't. The classic form is pairs trading — long one asset, short a related one, profit when the spread reverts.
Modern stat arb generalizes to many-asset models. Construct a basket via PCA, regression on factors, or learned embeddings; trade the residual mean-reverting component. Half-life is a critical parameter: the time scale on which the spread reverts. Short half-lives mean fast turnover and high transaction costs; long half-lives mean slow capital turnover and high regime-change risk.
Key risks:
- Regime breaks: cointegration relationships fail, often suddenly.
- Crowding: when everyone runs the same factor model, exits become correlated.
- Transaction costs: turnover-heavy strategies are eaten alive without low-cost execution.
Successful stat arb requires clean signal construction, robust risk control, and realistic cost modeling. The quant equity factor model business is essentially industrialized stat arb.