Third-Party Research & Methodology Only

This section shares summaries of third-party academic research and descriptions of quantitative models. The content represents the findings of the original researchers, not the opinions or recommendations of Foxholm Financial. Foxholm Financial does not publish hypothetical or backtested performance metrics on its quantitative research pages. All content is restricted to methodology, signal construction, factor logic, and risk architecture. SEC rules require that investment advisers not present misleading performance data, and our methodology-only approach reflects that standard and the firm's fiduciary obligations.

Pairs Trading

Trading Strategy Market Neutral Statistical Method

Pairs trading is a market-neutral strategy (one designed with the goal of profiting regardless of whether the overall market rises or falls, though this is not guaranteed; see Risk Management) that involves simultaneously buying one asset and short-selling a related asset when their price relationship diverges from its historical norm. The strategy profits if the two prices converge back toward their typical spread, regardless of the direction the broader market moves.

The core idea is straightforward: find two assets that historically move together, wait for their prices to diverge beyond a normal range, then bet that the gap will close. The trader buys the asset that has become relatively cheap and sells short the one that has become relatively expensive. When the spread narrows back to its historical average, both positions are closed for a profit.

Definition

Pairs trading belongs to a family of strategies known as statistical arbitrage, where traders use quantitative methods to identify and exploit temporary mispricings between related securities. Unlike directional strategies that bet on whether a single stock will rise or fall, pairs trading bets on the relative performance of two assets.

Core Concept

A pairs trade has two simultaneous positions:

Long position: Buy the underperforming asset (the one that has fallen relative to its pair).
Short position: Sell short the outperforming asset (the one that has risen relative to its pair).

The profit comes from convergence: if the spread (the price gap between the two assets) narrows back toward its historical average, the long position gains more than the short position loses, or vice versa. Because the strategy holds both a long and a short position, net exposure to the overall market is approximately zero.

The strategy originated at Morgan Stanley in the mid-1980s, where a quantitative trading group led by Nunzio Tartaglia developed computer-driven models to identify and trade pairs of stocks whose prices had temporarily diverged. The approach was among the earliest systematic applications of mean reversion (the tendency for prices to return toward a long-run average) in equity markets.

How Pairs Trading Works

A pairs trading strategy follows a structured process: select a pair of related assets, construct a spread, define entry and exit signals, and manage position sizes. Each step relies on statistical tools to ensure the trading decisions are systematic rather than discretionary.

1. Pair Selection

The first step is identifying two assets whose prices share a stable long-run relationship. This is typically done using cointegration tests (described in the next section) or by screening for assets within the same industry or sector that share similar fundamental drivers.

2. Spread Construction

Once a pair is identified, the trader constructs a spread: a single time series that captures the price difference (or ratio) between the two assets. A common approach is to compute the spread as the price of Asset A minus a hedge ratio (a scaling factor derived from regression analysis) times the price of Asset B. The hedge ratio ensures that a one-dollar move in one asset is approximately offset by the corresponding position in the other.

3. Entry and Exit Signals

Entry and exit points are determined using the z-score of the spread. A z-score measures how many standard deviations (a measure of typical variation) the current spread is from its historical mean. Common thresholds include:

Entry: One common systematic trigger involves opening a position when the z-score exceeds +2 or falls below −2 (meaning the spread has widened to roughly twice its normal variability).
Exit: The position is typically closed when the z-score returns to zero or crosses back through a threshold near zero (meaning the spread has reverted to its average).
Stop-loss: The trade is generally exited if the z-score exceeds +3 or −3, indicating the spread may be diverging permanently rather than reverting.

4. Position Sizing

Position sizes are scaled so that the dollar value of the long position approximately equals the dollar value of the short position. This dollar-neutral construction minimizes exposure to broad market movements and ensures the trade's profit or loss depends primarily on the relative movement of the two assets. Some implementations use beta-adjusted sizing to account for differences in how sensitive each asset is to the overall market.

Pair Selection Methods

Several statistical and fundamental approaches are used to identify pairs with a stable, mean-reverting relationship. Each method has different strengths and trade-offs.

Method	How It Works	Strengths	Weaknesses
Distance Method	Measures the sum of squared differences between two normalized price series over a formation period; selects pairs with the smallest distance	Simple to implement; no distributional assumptions	Does not test for a true mean-reverting relationship; can select pairs that happen to move together temporarily
Cointegration (Engle-Granger)	Tests whether a linear combination of two non-stationary price series produces a stationary (mean-reverting) residual using a two-step regression and unit root test	Formally tests for a long-run equilibrium relationship; strong theoretical foundation	Sensitive to the lookback window chosen; relationship can break down over time
Correlation-Based	Screens for pairs with high historical return correlation (a measure of how closely two return series move together) and trades when the correlation temporarily weakens	Intuitive and easy to compute	High correlation does not imply mean reversion; correlated assets can drift apart permanently
Fundamental / Industry Matching	Selects pairs within the same industry or with similar business models, revenue drivers, and risk exposures, then applies a statistical filter	Economic logic supports the pairing; reduces risk of spurious statistical relationships	Limits the universe of potential pairs; industry membership alone does not guarantee price co-movement

In practice, many implementations combine fundamental screening (to ensure the pair makes economic sense) with cointegration testing (to confirm a statistically robust relationship). This two-stage approach reduces the risk of selecting pairs based on coincidental price patterns that are unlikely to persist.

Practical Example

Consider two large companies in the same industry, Company X and Company Y, whose stock prices have historically moved in close lockstep. Over the past two years, the spread between their prices (adjusted by a hedge ratio) has had a mean of $0 and a standard deviation of $2.

One day, Company X drops on a temporary earnings miss while Company Y holds steady. The spread widens to $4, or two standard deviations (2σ) from its mean. A pairs trader would interpret this as a divergence likely to revert:

Action: Buy Company X (the underperformer) and sell short Company Y (the relative outperformer).
Z-score at entry: +2.0 (the spread is two standard deviations above its average).
Expected outcome: Over the following days or weeks, the spread narrows back toward $0 as the market recognizes the earnings miss was temporary. The trader closes both positions when the z-score returns near zero.
Profit source: The gain on the long position in Company X exceeds the loss on the short position in Company Y (or vice versa), because the spread contracted.

This example illustrates the key assumption behind pairs trading: that the historical price relationship between the two assets reflects a genuine economic link, and that temporary deviations from that relationship will correct themselves. When that assumption holds, the strategy profits. When it does not, the spread can widen further, producing losses.

Risk Management

Effective risk management is essential for pairs trading because the core assumption (that the spread will revert) can fail. Several controls help limit downside exposure.

Stop-loss thresholds: If the spread widens beyond a predefined level (for example, 3 standard deviations), the trade is closed to prevent further losses. This protects against scenarios where the historical relationship has permanently broken down.
Maximum holding period: Trades that remain open beyond a set time limit (often 10 to 30 trading days) are closed regardless of whether the spread has reverted. Extended holding periods increase the risk of adverse events affecting one or both legs of the trade.
Spread divergence risk: The most dangerous scenario in pairs trading is a permanent structural break in the relationship between the two assets. A merger announcement, regulatory change, or fundamental shift in one company's business can cause the spread to widen indefinitely. No amount of patience will produce a reversion in these cases.
Liquidity monitoring: Both legs of the trade must be liquid enough to enter and exit without significant price impact. Illiquid securities can create situations where one leg is closed easily but the other cannot be, leaving the portfolio with unintended directional exposure.
Sector and factor concentration: Running multiple pairs within the same industry creates hidden correlation risk. If the entire sector experiences a shock, many pairs can move against the trader simultaneously.

Known Limitations

Limitations to Keep in Mind

Cointegration breakdown. The statistical relationship between two assets can weaken or disappear entirely due to structural changes such as mergers, acquisitions, regulatory shifts, or changes in a company's business model. A pair that was cointegrated for five years may suddenly stop reverting, turning a previously profitable trade into a persistent loss.
Crowding. As more traders adopt the same pairs and the same entry signals, the profit opportunity shrinks. Popular pairs can become crowded, meaning that many participants enter at the same z-score threshold, compressing the available return and increasing the risk of a sudden unwind if many traders exit simultaneously.
Transaction costs. Pairs trading involves frequent trading on both the long and short sides, each incurring commissions, bid-ask spreads, and market impact. For strategies with small expected profit per trade, these costs can consume most or all of the gross return.
Short-selling constraints. The short leg of a pairs trade requires borrowing shares, which incurs borrowing costs and may not always be available. In some markets or for certain securities, short-selling restrictions can prevent the strategy from being implemented at all. Unexpected short squeezes (rapid price increases in a heavily shorted stock) can produce large losses on the short side.
Mean reversion assumption may fail. Pairs trading assumes that price spreads are stationary (fluctuating around a stable average) and will revert to the mean. In reality, some divergences reflect permanent changes in relative value. Distinguishing a temporary dislocation from a permanent shift is the central challenge of the strategy.
Regime dependence. The profitability of pairs trading can vary significantly across different market regimes. Research suggests that the strategy tends to perform better during periods of high market volatility and less well during calm, trending markets when spreads remain narrow.

Academic Origin

The first rigorous academic study of pairs trading was published by Evan Gatev, William Goetzmann, and K. Geert Rouwenhorst in 2006, though a working paper version circulated from the late 1990s. Their paper, "Pairs Trading: Performance of a Relative-Value Arbitrage Rule," analyzed a simple distance-based pairs strategy applied to U.S. equities from 1962 to 2002.

The researchers found that top pairs generated statistically significant excess returns, even after accounting for transaction costs and risk. The study also documented that profits were concentrated in the first few days after a trade was opened, suggesting that the market corrected mispricings relatively quickly. Their work provided the academic foundation for a strategy that had already been practiced on Wall Street for nearly two decades.

Subsequent research has explored more sophisticated pair selection methods. Cointegration-based approaches, drawing on the work of Robert Engle and Clive Granger, offer a stronger statistical framework than the original distance method. Ganapathy Vidyamurthy's 2004 book, Pairs Trading: Quantitative Methods and Analysis, provided practitioners with a comprehensive guide to implementing cointegration-based pairs strategies. More recent work by Elliott, van der Hoek, and Malcolm (2005) introduced a Gaussian mean-reverting model for the spread, enabling more precise signal generation and risk management.