Pair Trading Statistical Arbitrage: Bet on Relative Spreads

Key Takeaways

• Pair trading statistical arbitrage goes long the cheap leg and short the rich leg of two related securities when their spread exceeds a z-score threshold.
• Gatev, Goetzmann, and Rouwenhorst found the strategy earned about 11% per year in excess returns from 1962 to 2002 before costs.
• The most common mistake is assuming a spread will revert when one company has fundamentally changed, breaking the historical relationship.
• Pair trading adds market-neutral return to a portfolio because profit comes from relative moves, not from the direction of stocks overall.

What It Is

A pair trade is built around two assets that tend to move together: two oil majors, two regional banks, an ETF and its largest holding. When the spread between their prices widens beyond its normal range, you buy the cheap one and short the rich one. When the spread snaps back toward its historical mean, you close both legs and pocket the convergence.

Because the long and short legs offset each other on a dollar or beta basis, the position has very little exposure to the broad market. The profit source is the relative move between the two names, not the direction of stocks as a whole.

The Intuition

Two companies in the same business face most of the same shocks. If oil prices jump, both oil majors should rally. If rates fall, both regional banks should benefit. When one of them moves sharply without its partner, something idiosyncratic has happened: a rating change, a rumor, a liquidity event. Often that dislocation fades within days or weeks as the market digests the news.

Stat arb formalizes this intuition. Instead of guessing which stock is mispriced in absolute terms, you only need to judge which of two linked names has wandered away from the other. That is a much easier question than "is AAPL fairly valued?"

How It Works

The canonical approach has three steps.

1. Find pairs. Screen a universe for security pairs whose normalized prices have tracked each other closely over a formation period, typically 12 months. Gatev, Goetzmann, and Rouwenhorst measured similarity using the sum of squared deviations between two normalized price series and selected the 20 closest matches.
2. Set a trading rule. Over a trading period, usually the 6 months that follow, open a position whenever the spread diverges by more than a threshold such as two historical standard deviations. Close it when the spread returns to zero or at the end of the window.
3. Size the legs. Allocate equal dollars to the long and short legs, or equal beta, so the combined position has near-zero market exposure.

More sophisticated variants replace the distance metric with a formal cointegration test, which checks that a linear combination of the two price series is stationary. Stationarity is the statistical property that justifies a mean-reversion bet.

spread_t = price_A_t - beta * price_B_t
open trade when |spread_t - mean| > k * stdev
close trade when spread_t crosses the mean

Worked Example

Coca-Cola (KO) and PepsiCo (PEP) have historically moved within a narrow relative band. Suppose over the last year the ratio PEP / KO has averaged 2.80 with a standard deviation of 0.12. One morning PEP gaps up on a strong earnings surprise while KO is unchanged, and the ratio jumps to 3.10.

That is 2.5 standard deviations above the mean. A pair trader shorts PEP and buys KO in equal dollar amounts. Over the next three weeks KO drifts up and PEP consolidates, pulling the ratio back toward 2.85. The trader closes both legs. The short made money, the long made a little, and the total return is the spread compression, independent of whether the S&P 500 rose or fell during those three weeks.

Common Mistakes

1. Assuming the spread will always mean-revert. Sometimes one of the two companies has genuinely changed. A merger, a lawsuit, or a shift in business mix can break a relationship that held for a decade. Pair traders need hard stop-loss rules for positions that keep diverging.

2. Underestimating crowding. Gatev et al. documented excess returns around 11 percent per year from 1962 through 2002. Returns collapsed after 2003 as hedge funds piled in and high-frequency systems arbitraged the simplest distance-method opportunities away in minutes.

3. Ignoring borrow costs and short rebates. The short leg is not free. Hard-to-borrow names can charge borrow fees of 5 percent or more annualized, which eats the spread's expected return. Always check borrow availability before opening the trade.

4. Using price, not total return. Dividend payments on one leg shift the raw price spread without signaling any dislocation. Work with total-return series, or at minimum adjust for ex-dividend dates.

5. Backtesting without transaction costs. A pair trade round trip involves four fills (open long, open short, close long, close short). At retail spreads plus commissions plus market impact, the expected edge on many pairs disappears entirely.

Frequently Asked Questions

Q: What is pair trading statistical arbitrage in simple terms? Pair trading means simultaneously buying one stock and shorting a closely related one when their prices have diverged unusually. You profit not from either stock going up or down, but from the gap between them closing back to a normal range.

Q: How does pair trading statistical arbitrage affect investment decisions? It shifts research focus from absolute valuation to relative valuation. You ask which of two linked companies is mispriced relative to the other, which is a narrower and more tractable question than asking whether a stock is cheap in absolute terms.

Q: What is a real-world example of pair trading statistical arbitrage? The article's PEP/KO example shows the ratio jumping to 3.10 (2.5 standard deviations above the average of 2.80) on a PepsiCo earnings surprise. A trader shorts PEP and buys KO in equal dollar amounts, then closes when the ratio drifts back toward 2.85.

Q: How can investors use pair trading statistical arbitrage in their portfolio? Screen pairs by cointegration, not just correlation. Use a 2-standard-deviation entry threshold and a hard stop at 3. Check borrow availability and cost before opening the short leg, and work in total-return series to avoid being misled by ex-dividend price gaps.

Q: How is pair trading statistical arbitrage different from long/short equity? Pair trading bets on the relative move between two specifically selected securities with a verified historical relationship. Long/short equity builds a broader book of long and short positions based on overall valuation views without necessarily pairing individual names.

Sources

1. Gatev, E., Goetzmann, W.N., Rouwenhorst, K.G. (2006). "Pairs Trading: Performance of a Relative-Value Arbitrage Rule." Review of Financial Studies, 19(3), 797-827. https://academic.oup.com/rfs/article-abstract/19/3/797/1646694
2. NBER Working Paper 7032 (earlier version of the Gatev et al. study). https://www.nber.org/papers/w7032
3. Hudson and Thames. "The Comprehensive Introduction to Pairs Trading." https://hudsonthames.org/definitive-guide-to-pairs-trading/
4. Thorp, E.O. (2017). A Man for All Markets. Random House. https://www.penguinrandomhouse.com/books/314051/a-man-for-all-markets-by-edward-o-thorp/

Disclaimer

This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.