42 / 50
Relative value
Does statistical arbitrage differ from one pair in scale, or in the bet?
Statistical arbitrage
What it bets
A price's residual against a basket of peers will revert. The portfolio holds many small residuals, and it does not depend on any one pair staying valid forever.
How the rule is written
Explain each name with the basket, and keep the residual. When a residual is extreme, fade that name and hedge with the basket. The holdings are widely spread out, and any one pair has a small weight. It is still reversion. It is not a risk-free arbitrage. The model, the hedge ratios, and the extreme threshold are all estimated inside the sample.
When it fails
When the peers lose the relationship together, a residual stops being a residual, and the portfolio becomes exposure in one direction. The finer the sample is mined, the more the residuals inside that sample behave as if they must revert.
Do not confuse it with
A pair is one spread. Statistical arbitrage is a portfolio of many residuals. The word arbitrage in the name does not mean a risk-free spread has been locked in.