Statistical Arbitrage

Statistical arbitrage (stat arb) is a quantitative trading strategy that uses statistical models to identify and exploit price discrepancies between related financial instruments. Unlike pure arbitrage which is risk-free, statistical arbitrage involves residual statistical risk: the model assumes that historical relationships will continue to hold, which is not always the case.

What Is Statistical Arbitrage?

Stat arb is based on the assumption that certain securities have a stable, quantifiable statistical relationship (mean reversion, cointegration, or correlation). When the relationship temporarily deviates from historical norms, stat arb strategies take positions to profit from the expected reversion.

It generalises the pair trading concept to portfolios of many securities, using mathematical models to define the expected relationship and measure divergence.

Common Stat Arb Approaches

**Cointegration-based models:**
Two or more securities are cointegrated if their price ratio is stationary (returns to a stable long-term mean). The trader models this relationship and trades when the spread exceeds a threshold.

**Factor neutralisation:**
Build a portfolio that is neutral to broad market factors (beta, sector, size) while having specific bets on individual securities based on statistical signals.

**Mean reversion models:**
Securities that have deviated significantly from their historical mean return to it. Traders buy underperformers and short outperformers with this expectation.

The Role of Data

Stat arb heavily depends on large datasets:
– Price and volume data
– Corporate fundamentals
– Alternative data (satellite images, web traffic, credit card transactions)
– Economic and sentiment data

Machine learning models (neural networks, gradient boosting) are increasingly used to find non-linear statistical relationships.

Risks

**Model risk**: historical statistical relationships can break down. During the 2007-08 financial crisis, many stat arb funds suffered simultaneous blow-ups as their models failed to handle correlated stress.

**Crowding risk**: when many funds use similar stat arb signals, the trade becomes crowded and no longer profitable, or worse, unwinds suddenly.

**Short-selling constraints**: in some markets, shorting stocks is expensive or restricted.

Practical Example

A quant fund finds that IT sector stocks in India exhibit strong cointegration. They model the price spread between a basket of IT exporters and the INR/USD exchange rate. When the spread deviates by 2 standard deviations (rupee depreciated but IT stocks didn’t rally as much as the model predicts), they go long the IT basket. When the relationship reverts over 2 to 3 weeks, they exit for a profit.

Key Takeaways

– Statistical arbitrage exploits quantifiable statistical relationships between related securities
– Based on mean reversion, cointegration, or factor-based pricing models
– Involves residual model risk unlike pure (no-risk) arbitrage
– Widely used by quantitative hedge funds with large datasets and sophisticated models
– Crowding and regime change are the two biggest risks that can destroy stat arb strategies