Building Statistical Arbitrage Portfolios Around Return, Risk, and Costs
Summary
The document presents statistical arbitrage as a broader portfolio problem than trading matched pairs. It ranks assets by expected cheapness or expensiveness, then builds long and short positions intended to capture relative value convergence while offsetting shared market risks. Predictive inputs may include momentum, mean reversion, microstructure effects, lead-lag relationships, and carry. The author recommends assessing features with information coefficients, decay rates, and relationships to other signals before combining them into expected returns.
Portfolio construction must balance expected return against variance, constraints, and turnover. A risk model or simpler exposure rules can reduce common risks, even if that means shorting an asset expected to rise when it is less attractive than correlated alternatives. Rebalancing should account for current holdings: a small improvement in forecast may not justify certain trading costs when estimates are uncertain. The discussion is a conceptual framework rather than a worked implementation; detailed feature research, optimization examples, and empirical performance evidence are deferred.
Key ideas
- Statistical arbitrage can rank a broad universe of assets and trade relative value without relying on fixed pairs.
- Forecast quality depends on predictive strength, signal decay, and how signals relate to one another.
- Risk-aware sizing can offset shared exposures and improve risk-adjusted portfolio behavior.
- Rebalancing should consider current positions because expected gains must justify known transaction costs.
- Simple heuristics or constrained optimization can express the trade-off among return, risk, costs, and portfolio limits.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.