Statistical arbitrage

Statistical arbitrage (stat-arb) is a quantitative investment strategy of identifying statistical mispricings among multiple stocks or assets — using regression models, correlation analysis, or other statistical techniques — then establishing offsetting long and short positions to exploit those mispricings while hedging market risk.

How statistical arbitrage works

A stat-arb model:

1. Identifies factors. Constructs a regression or factor model predicting stock returns based on characteristics (valuation, momentum, quality, etc.).
2. Ranks stocks. Scores all stocks in the universe (e.g., S&P 500) based on the model.
3. Creates baskets. Long the highest-scoring stocks (predicted to outperform) and short the lowest-scoring stocks (predicted to underperform).
4. Sizes positions. Constructs the long and short baskets with equal gross exposure, ensuring market risk is neutral.
5. Monitors and rebalances. Updates scores regularly, trimming losers, adding winners.

Example: A model predicts that stocks with low debt-to-equity ratios and high profitability will outperform for the next month. The strategy:

• Long: 20 high-quality, low-leverage stocks, $100 million combined.
• Short: 20 low-quality, high-leverage stocks, $100 million combined.
• Expected return: If the model is right, the long basket outperforms the short basket, capturing the 50–200 basis point spread.
• Market exposure: Zero, since longs and shorts are equal.

Advantages

• Market-neutral returns. Immune to bull or bear market moves; returns depend only on factor performance.
• Diversification. Many baskets (50+ stocks) dilute single-stock risk.
• Systematic and scalable. Rules-based; can be applied to thousands of stocks mechanically.
• Multiple factor bets. A sophisticated stat-arb model can have exposure to value, momentum, quality, and other factors simultaneously.

Challenges

1. Model risk. If the model breaks (factors stop working, relationships change), losses can accelerate.
2. Crowding. As more hedge funds use the same factors, mispricings narrow, reducing returns.
3. Execution costs. Long 20 stocks and short 20 stocks = 40 transactions. Costs compound.
4. Correlation breakdowns. Assumed correlations among stocks may break, especially in crises.
5. Tail risk. Models trained on normal data often underestimate crash risk. A sharp market reversal can cause model-wide losses.

Historical returns

Stat-arb strategies historically delivered 3–8% annualized returns with 5–10% volatility (lower than the broader market). However, as of the 2010s, returns have compressed due to crowding — more capital chasing the same factors.

See also