Pairs trading: long one asset, short its statistical partner

Pairs trading is a strategy almost as old as systematic trading itself. The premise is simple: identify two assets whose prices have historically moved together, wait for them to diverge, then go long the cheaper of the pair and short the more expensive one. When the relationship reverts to its historical norm, both legs profit. The strategy is the original implementation of statistical arbitrage and remains a useful entry point into long-short construction.

What pairs trading is

Pairs trading is a market-neutral strategy that pairs a long position in one asset with a short position in another. The two assets are typically chosen because of a historical relationship—same industry, same factor exposures, same supply chain—that suggests their prices should move in tandem most of the time. When the relationship breaks down temporarily, the strategy bets on its reversion.

The most-cited canonical pairs example is Coke and Pepsi: two companies in the same industry with similar product lines, similar cost structures, and broadly similar exposures to consumer-spending dynamics. Their share prices have historically moved together over multi-year horizons, and short-term divergences have tended to revert. The strategy goes long whichever is below its historical relative position and short the other; profits accrue as the divergence narrows.

The technique was developed at Morgan Stanley by Bamberger and others in the 1980s, where it was the foundation for one of the earliest dedicated quantitative trading desks. It generalises into the broader category of statistical arbitrage, in which the same logic is applied across thousands of pairs simultaneously to produce a diversified market-neutral strategy.

How it works

The standard pairs-trading workflow has three stages. First, identify candidate pairs: assets whose price relationship has been stable historically, typically measured by cointegration tests or by the standard deviation of the price ratio over a defined look-back window. Second, define the entry signal: the spread (or ratio) between the two assets has moved beyond a threshold (typically 2 standard deviations from its long-run mean). Third, monitor for the exit: when the spread reverts to its mean, both legs are unwound and the trade is closed.

For a pair like Coke and Pepsi, the spread might be the log-ratio of their share prices. If the long-run average of this log-ratio is, say, +0.05 with a standard deviation of 0.10, an entry signal triggers when the current log-ratio exceeds +0.25 (two standard deviations above the mean)—go long Pepsi, short Coke. The position is held until the log-ratio reverts to +0.05, at which point both legs are closed.

Position sizing matters. The two legs should be sized so that they have approximately equal dollar exposure (or equal beta-adjusted exposure, depending on the implementation), so that the trade's P&L depends on the relative rather than the directional movement of the pair. Imperfect sizing introduces directional risk that the strategy is supposed to eliminate.

What the evidence shows

Gatev, Goetzmann, and Rouwenhorst (2006), in Pairs Trading: Performance of a Relative-Value Arbitrage Rule, document the strategy's historical profitability using a simple distance-based pair selection on US equities over 1962–2002. The study found average annualised excess returns of approximately 11% with relatively low volatility, producing Sharpe ratios above 1.0 in the early decades of the sample.

The same study and subsequent work documents the strategy's decay over time. Pairs trading produced strong returns through the 1980s and into the 1990s, weakened materially in the 2000s as quantitative strategies became more crowded, and has been only marginally profitable since 2010 in standard implementations. The pattern is the McLean-Pontiff (2016) finding applied to pairs trading: the strategy was real, it became known, capital deployed against it, and the apparent edge largely disappeared.

Modern pairs and statistical-arbitrage implementations have evolved beyond the simple distance-based selection of the original Gatev paper. Cointegration tests, Bayesian regime-switching models, and machine-learning approaches to pair selection are all in use; the more sophisticated implementations have continued to produce profitability in some periods, but the simple textbook version has been arbitraged away.

Limitations and trade-offs

The strategy's main vulnerability is structural break in the historical relationship. The whole framework depends on the assumption that the two paired assets will revert to their historical relationship; if the relationship has fundamentally changed (a merger, a regulatory shift, the emergence of new competition), the divergence is not a temporary aberration but a new equilibrium, and the strategy will lose money waiting for a reversion that never comes.

The cost of the short leg is non-trivial. Short positions require borrowing the underlying asset, paying borrow fees, and posting margin. The borrow cost on hard-to-borrow names can erase the trade's expected return entirely; on easy-to-borrow names the cost is small but non-zero, and it accumulates over the holding period. Pairs trades that take longer than expected to resolve face a structural cost the original setup did not account for.

The strategy is also crowded. Many quantitative funds run pairs and statistical arbitrage strategies simultaneously, and the resulting positioning can produce correlated drawdowns when the trades unwind together—the August 2007 quant equity crisis was partly driven by simultaneous pairs-strategy unwinds across major funds. The systemic risk is small relative to broader market events but real for the strategy class itself.

Finally, pair selection is the biggest source of overfitting risk in the entire methodology. The temptation to optimise selection criteria on the historical sample produces strategies that perform impressively in-sample and disappointingly out-of-sample. Robust cross-validation and out-of-sample testing are essential.

Pairs trading in pfolio

Pairs trading—long one asset, short a related one—can be expressed in pfolio's Portfolio Builder by combining a positive allocation to one asset with a negative allocation to another. The Asset Builder supports synthetic short positions where direct short instruments are not available. Risk metrics for the resulting paired position are visible in pfolio Insights.

Related articles

• Market neutral strategies: generating returns independent of broad market direction
• Mean reversion in systematic investing: the counterpart to trend following
• Long-short equity strategies: how taking both sides of the market changes the return profile
• Correlation in portfolio management: why diversification depends on it
• Statistical arbitrage: applying pairs-trading logic across many simultaneous positions