Triangulating Pair Signals to Build a Statistical Arbitrage Portfolio
Summary
This article explains how to combine overlapping pair spread signals to infer which individual stocks appear rich or cheap relative to peers. Each spread acts as a relative vote; aggregating votes across a network can help distinguish a likely outlier from a partner that merely moved. Flattening these signals creates a long-short portfolio in which each position targets an apparent mispricing, rather than using one leg mainly as a hedge. The author also describes weighting ticker signals by vote consistency, with stronger agreement receiving more weight.
The method cannot establish why a stock moved. Temporary flows may create a mean-reverting dislocation, while company news may cause a lasting repricing; spread z-scores alone do not distinguish them. News and volume features may help identify informed moves. Regression, lasso, and ridge are discussed as alternative ways to infer ticker-level signals from spread data. The article reports that consistency weighting performed better in the author’s tests and claims a risk-adjusted improvement over traditional pairs, but its plotted comparison is explicitly not a cost-adjusted backtest. Pair selection remains foundational, and the portfolio approach adds complexity, turnover, and concentration trade-offs.
Key ideas
- Overlapping spread signals can be aggregated to identify stocks that appear rich or cheap across several relationships.
- A portfolio built from ticker-level signals can target apparent mispricings on both its long and short sides.
- Signal consistency measures agreement across a ticker’s pair relationships and can be used as a weight.
- Spread signals alone cannot tell temporary flow-driven dislocations from repricing caused by news.
- The comparisons described omit costs and frictions, and the method depends heavily on selecting tradeable pairs.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.