Pairs trading

Definition

Pairs trading holds one instrument long against a related instrument short, trading the SPREAD between them rather than either price: enter when the spread is unusually wide, exit when it normalizes. It is mean reversion (strategy-mean-reversion) applied to a relative price, with market direction largely hedged out — the thesis is about the relationship, not the tape.

How it works / structure

• Pair selection: economic relation first (same industry, share classes, index vs constituents), statistical confirmation second — distance methods (normalized price gap; the Gatev-Goetzmann-Rouwenhorst formation rule) or cointegration tests; correlation alone is insufficient.
• Parameters (engine-executable): formation window, entry threshold (spread z-score, e.g. 2.0), exit at mean vs opposite band, stop as spread z (e.g. 3.5) or thesis-break event, maximum holding time, dollar-neutral vs beta-neutral sizing (port-exposure-netting), borrow-cost ceiling on the short leg (ms-short-locate-borrow).
• Book form: many pairs concurrently — single-pair outcomes are noisy; the strategy is statistical at the portfolio level.

When it applies

Related-instrument universes with stable structural links (sector peers, dual listings, futures inter-market versions — strategy-inter-market-spread), calm-to-normal regimes, and accounts with clean shorting capacity. The hedged construction makes it a common first systematic strategy — its risks are subtler than they appear (below).

Risk profile & failure modes

• Divergence for a reason: the spread widening can be information (one company deteriorating) — the classic loss is shorting the winner and buying the loser of a real repricing; event filters and thesis-break stops are structural.
• Cointegration decay: relationships estimated in-sample break silently (mergers, business-model drift); stale pairs are the strategy’s rot.
• Two-legged costs: every position pays two spreads plus borrow; Do-Faff (2010) attribute much of the strategy’s decline to costs and competition.
• Correlated unwinds: many managers hold similar pairs; stress liquidations widen all spreads together — hedged is not the same as safe (risk-correlation-exposure).

Evidence & limits

Gatev-Goetzmann-Rouwenhorst (2006) documented ~11% annualized excess returns to a simple distance rule over 1962-2002 — the canonical study; Do-Faff (2010) found the returns declined substantially in later samples, largely cost- and crowding-driven. Current-period profitability of simple public rules is doubtful; the platform treats pair theses as individual falsifiable claims graded on replay, not a presumed edge class.

Falsifiable-thesis examples

Illustrations only, not signals:

• “The X/Y spread, 2.5 standard deviations wide today, will return to its 60-day mean within 20 sessions” — falsified by the spread series.
• “A 20-pair distance-rule book on universe U will be net profitable after two-leg friction this quarter in replay” — falsified by replay P&L.

Cross-references

• Underlying logic: strategy-mean-reversion; futures cousin: strategy-inter-market-spread
• Short-leg mechanics: strategy-short-selling, ms-short-locate-borrow
• Exposure accounting: port-exposure-netting, risk-correlation-exposure
• Method: lens-quantitative (cointegration vs correlation, decay monitoring)