FoxAlgo → Research → Academic Audit → Pairs Trading with Stocks

Academic audit

Failedequity stat-arb

Pairs Trading with Stocks

The three-gate gauntlet · genuine only if it clears all three and survives adversarial refutation
Gate 1
Survivorship-free
free
clean panel
Eliminated here
Gate 2
Placebo ≥ P95
P51.5
outranked ~103 of 200 baskets
Gate 3
Cost-aware net
RF -1.00
net-negative after costs
Failed
Worst 12-month leg (RF)-1.00
−1.00 floor0
Every strategy here — winners included — loses in its worst 12 months. Depth is honest context, not the verdict.
Rejected at the luck gate — its net ranked no better than random baskets (below the P95 skill line).

Pairs trading pairs up two historically co-moving stocks, then bets on convergence when their prices diverge by a set threshold and closes when the spread reverts. It is a classic relative-value, market-neutral rule from the Gatev, Goetzmann & Rouwenhorst study.

What we found

On our survivorship-free test the distance-based rule is net negative and does not clear the screen. Its placebo percentile is essentially a coin flip, and the four-fill round-trip needed to open and close each pair eats the spread on modern data. This is consistent with the well-documented post-2002 decay of the effect: the classic edge appears to be arbitraged away, and what remains is cost-fatal after realistic trading costs.

How we tested it
2005–2026 test windowmodelled liquidity-aware costssurvivorship free
  • Data: survivorship-free 1077-name US common-stock panel, 2005-2026. Realistic modelled costs.
  • Placebo / robustness test: real result vs random baskets or shuffled signals (real vs the 95th percentile of random)
Source: Gatev, Goetzmann & Rouwenhorst (2006), "Pairs Trading: Performance of a Relative-Value Arbitrage Rule", RFS
Read the paper ↗
← The Academic Audit — all 54 studies

Research, not investment advice. “Validated” factor-legs are market-neutral diversifying building blocks with a losing worst year — none is a standalone tradeable strategy. Metrics are cost-aware and modelled (not live fills); the 2005–2026 test window is out-of-sample versus the source paper. Dollar figures are not returns and are omitted by design.