Statistical Arbitrage

Statistical arbitrage, usually called stat arb, is a family of quantitative strategies that try to profit from small, temporary mispricings among large numbers of related securities. Unlike true arbitrage, which locks in a riskless profit, stat arb relies on probabilities: each trade has only a slight edge, but by holding hundreds or thousands of positions and keeping market exposure near zero, the edge can add up. It grew out of pairs trading at Morgan Stanley in the 1980s and became a core strategy for quantitative hedge funds.

From pairs to portfolios#

A single pair is exposed to the risk that one relationship breaks. Stat arb spreads that risk by trading many relationships at once.

How a stat arb model works#

1. Define a universe, such as the 1,000 most liquid US stocks.
2. Model expected returns: for each stock, estimate how it "should" move based on factors such as its sector, market beta and related stocks. See Factor Models.
3. Compute residuals: the part of each stock's move not explained by the model.
4. Trade the residuals: buy stocks that have fallen more than the model predicts and sell those that have risen more, expecting the residuals to revert. See Mean Reversion.
5. Neutralise exposures: keep the portfolio close to zero net market, sector and factor exposure.
6. Execute efficiently to keep costs low. See Execution Algorithms vs Alpha Algorithms.

What drives stat arb profits#

• Liquidity provision: when large investors push prices to trade quickly, stat arb takes the other side and earns a reward for supplying liquidity.
• Overreaction to news or flows that reverses over days.
• Lead and lag relationships between related securities.

Profits per trade are small, so results depend heavily on costs, execution and the number of independent bets. See Transaction Costs and Market Impact.

Risks#

• Model risk: factor models can miss important exposures. See Operational and Model Risk.
• Crowding and deleveraging: in early August 2007, many quant funds holding similar positions lost heavily as one large fund's forced selling pushed prices against everyone. Amir Khandani and Andrew Lo studied this "quant quake". See Factor Timing, Crowding and Crashes.
• Regime changes when old relationships stop holding.
• Short selling constraints and borrow costs.
• Leverage: market neutral portfolios are often leveraged to reach meaningful returns, which magnifies losses.

Stat arb and high frequency trading#

At very short horizons, stat arb overlaps with High-Frequency Trading and Market Making, trading tiny price differences between related instruments such as an ETF and its components, or futures and the underlying index. These versions require low latency infrastructure.

Can individuals do stat arb?#

Simplified versions, such as a portfolio of pairs or sector neutral mean reversion on daily data, are possible with Python and a broker that supports shorting. Competing with large firms on speed or breadth is unrealistic, and costs must be modelled carefully. See Portfolio and Multi-Asset Backtesting.

Frequently asked questions#

What is statistical arbitrage?#

A quantitative strategy that trades many small mispricings among related securities, keeping overall market exposure near zero.

Is statistical arbitrage really arbitrage?#

No. It is not riskless. Each trade has only a probable edge, and the strategy can lose money, sometimes sharply.

What happened in the August 2007 quant crisis?#

Many quantitative funds with similar positions suffered large losses within days as forced selling by some funds moved prices against all of them.

Next, learn how traders profit from the price difference between related contracts in Spread Trading.

Sources#

• Khandani, A. and Lo, A., What Happened to the Quants in August 2007?, 2007. Summary: Wikipedia, Statistical arbitrage