Statistical arbitrage is a family of strategies that buy and sell related securities when their prices drift apart by more than usual, expecting them to move back together. The positions are usually balanced so that overall market exposure is small.
How a simple version works
- Find related securities, such as two stocks in the same industry that have moved together, using correlation or cointegration.
- Measure the gap, or spread, between their prices and convert it to a z-score against its recent average.
- Trade extremes. If the spread is unusually wide, for example a z-score above +2, sell the expensive stock and buy the cheaper one.
- Exit when the spread returns toward its average, or stop out if it keeps widening.
An illustrative example
Stocks A and B usually trade so that A is priced at about 1.2 times B. One day A trades at 1.35 times B, and the spread's z-score is +2.1. The strategy sells A and buys B in matching amounts. If the gap closes, the pair earns the difference whatever the market does. If it keeps widening, the trade loses. The figures are made up.
Variations
| Type | Idea |
|---|---|
| Pairs trading | Two closely related stocks |
| Basket or sector-neutral | Many stocks traded against a basket, often after removing sector and market effects. See factor neutralization |
| Factor-based long-short | Long high-scoring stocks and short low-scoring ones |
Risks
- Relationships break. A merger, a change in a business or a sector shift can end a relationship, and the spread may never come back. See mean reversion.
- Crowding. When many funds trade the same pairs, forced selling can push spreads wider together.
- Costs. Frequent trading means slippage and fees can absorb the small gains.
- Shorting limits. In India, shorting single stocks is generally done through futures, which limits the tradable universe.
Frequently asked questions
What is statistical arbitrage?
It is a group of strategies that trade related securities against each other based on statistical relationships, expecting temporary price gaps to close, usually with little net market exposure.
Is statistical arbitrage risk-free?
No. The word arbitrage is misleading. The relationships are statistical and can break, and losses can be large when they do.
What is pairs trading?
It is the simplest form of statistical arbitrage: buying one stock and selling another closely related stock when the gap between their prices becomes unusually wide.
Why is cointegration used in statistical arbitrage?
Cointegration tests whether a combination of two price series stays stable over time, which is the property a pairs trade needs for the gap to close.