Market-neutral strategy using hundreds of small positions. Exploits temporary mispricings between related securities using cointegration and factor models.
History
Statistical arbitrage evolved from pairs trading at Morgan Stanley in the 1980s under Nunzio Tartaglia. The approach scaled from simple pairs to large portfolios of hundreds or thousands of positions, using factor models to remain neutral to market, sector, and style exposures. D.E. Shaw, Renaissance Technologies, and Citadel Securities became dominant practitioners in the 1990s and 2000s. The strategy experienced a severe crisis in August 2007 when crowded quant positions unwound simultaneously, causing losses of 10-30% in a single week for many stat-arb funds. This event, documented by Khandani and Lo (2007), demonstrated the systemic risks of crowded quantitative strategies.
How It Works
Build a multi-factor model explaining expected stock returns (value, momentum, quality, size, volatility factors)
Identify residual mispricings: stocks trading above or below their factor-model-implied fair value
Construct a portfolio long undervalued stocks and short overvalued stocks, neutralized across sectors, industries, and factor exposures
Maintain hundreds to thousands of small positions to diversify idiosyncratic risk
Use principal component analysis (PCA) or eigenvector decomposition to identify latent return drivers
Lever the portfolio 3-8x to generate meaningful returns from small per-position alpha
Rebalance intraday or daily to maintain neutrality constraints
Example Trades
Factor model identifies 200 stocks cheap vs peers on residual basis after controlling for sector, size, momentum
entry Long 200 undervalued names, short 200 overvalued names, each ~0.1% of portfolio
exit Rebalance daily as residuals converge; average holding 3-5 days
result +0.02-0.05% per day, compounding to ~8-15% annually before leverage
ETF/basket arbitrage: SPY trades at 0.15% premium to its fair value based on constituent stocks
entry Short SPY, long the constituent basket
exit Premium converges within hours
result +0.15% gross, scaled by leverage
Related Charts
Who Runs This
When It Works vs. Fails
works
Normal volatility environments with stable cross-sectional dispersion. When fundamental drivers dominate returns rather than macro/sentiment.
fails
Systemic deleveraging events (2007, 2020 March). Macro-driven markets where all stocks move together. Extremely low dispersion environments.
Risks
01 August 2007 quant crisis: crowded stat-arb positions unwound simultaneously, causing 10-30% losses in a week
02 Requires significant leverage (3-8x) to generate meaningful returns, amplifying tail risks
03 Alpha decay: signals lose profitability as more funds adopt similar factor models
04 Model risk: factor models may fail to capture regime changes or structural breaks
05 Execution risk: with thousands of positions, implementation costs and market impact are critical
Research
Khandani, Lo, 2007
Statistical Arbitrage in the US Equities Market
Avellaneda, Lee, 2010
Guijarro-Ordonez, Pelger, Zanotti, 2025