Baskets and Synthetic Spreads
When no single asset is a clean hedge for another, build a weighted basket of several that replicates it instead, and trade the spread between the target and its synthetic twin.
Prerequisites: Spread Construction and Hedge Ratios, Cointegration
A pairs trade needs a good single hedge for the asset you're trading. Often none exists — a stock's closest peer is only 60% correlated, or a bond has no perfect futures match. The fix is to stop looking for one asset and build several into a weighted basket instead, chosen so the basket, as a whole, tracks the target closely enough to trade the residual gap between them.
A synthetic spread is a target asset minus a weighted basket of other assets, with weights chosen (usually by regression) so the basket replicates as much of the target's typical movement as possible. What's left over — the spread between the real asset and its synthetic twin — is what actually gets traded.
Building the basket
Given a target asset and a set of candidate assets , regress on the basket to get weights that minimize the leftover variance:
In words: the target is expressed as a weighted sum of the basket components, plus a residual spread . The weights define how much of each component to hold to replicate the target; the residual is the synthetic spread — the piece of the target's price that the basket, however well-chosen, still can't explain. If that residual is stationary and mean-reverting, it's tradable exactly like a two-asset pairs spread, just built from more pieces.
Worked example
A desk wants to trade a mid-cap airline stock but its closest single peer only explains about 45% of its daily variance (). Regressing the stock instead against a basket of two peer airlines and a jet-fuel-cost proxy lifts to 0.81 — the three-asset basket, weighted peer A, peer B, and fuel proxy, tracks the target far more tightly than any single asset could. The residual spread from that basket regression, historically ranging 1.5 points and reverting within about 8 days, becomes the tradable signal: long the stock and short the basket in those weights when the spread is unusually wide, unwound as it reverts.
What this means in practice
Baskets trade off tracking quality against complexity: more components generally improve the fit, but each added leg adds transaction costs, financing, and operational risk (more positions to manage, more things that can each individually break the relationship). A basket is worth it only when it meaningfully tightens the fit over the best available single-asset hedge.
A basket's weights are fit on historical data and, like any hedge ratio, can go stale — see time-varying hedge ratios. A basket with many components has more moving parts that can each drift, so it needs re-estimation at least as often as, and usually more often than, a simple two-asset pair.
Related concepts
Practice in interviews
Further reading
- Avellaneda & Lee, 'Statistical Arbitrage in the U.S. Equities Market', Quantitative Finance (2010)