Pairs Trading Explained: Hedge Ratios, Z-Scores and Market-Neutral Risk

Pairs trading: relative value without false neutrality

Pairs trading takes a long position in one asset and a short position in another, aiming to profit if their relative relationship converges. It can reduce broad market exposure, but “market-neutral” is a target, not a guarantee.

বাংলায় পড়ুন: Pairs Trading Guide: Hedge Ratio, Z-Score ও Market-Neutral Risk

Learning objective: select an economically defensible pair, estimate a hedge ratio, measure the spread, define convergence and failure, and include short-borrow and two-leg execution risks.

1. Begin with an economic relationship

Two prices moving together historically is not enough. Prefer assets exposed to similar business drivers, such as companies in the same industry, share classes representing similar claims or closely related funds.

Ask why the relationship should persist:

• Do the businesses sell similar products to similar customers?
• Are their costs affected by the same commodity, currency or interest rate?
• Do they have comparable leverage, geography and regulation?
• Could an acquisition, index change, accounting issue or product launch permanently separate them?

A high historical correlation can disappear when fundamentals change. Correlation measures co-movement; it does not prove that the price difference will return to a stable mean.

2. Correlation and cointegration are different

Two assets can have highly correlated daily returns while their price relationship drifts indefinitely. Pairs strategies usually need a spread that has shown reasonably stable mean-reverting behaviour.

Cointegration tests whether a linear combination of two non-stationary price series is more stable over time. A statistical test can reject or fail to reject a hypothesis; it cannot guarantee future convergence. Results depend on the look-back, structural breaks and data quality.

The classic NBER pairs-trading study by Gatev, Goetzmann and Rouwenhorst matched stocks using historical normalised-price distance and tested convergence rules. Its historical findings illustrate a method, not a current return promise; market structure, costs and crowding can change.

3. Estimate the hedge ratio

A common model is:

Spread = Price A − β × Price B

The coefficient β is the estimated hedge ratio. It can come from regression, factor exposure, dollar neutrality or another pre-specified method. These methods answer different questions:

• Dollar-neutral: long and short notionals are equal.
• Beta-neutral: estimated broad-market beta is offset.
• Regression hedge: historical price relationship defines the units.
• Factor-neutral: sector, size, value or other exposures are explicitly balanced.

Equal cash on each side does not automatically create beta neutrality. Write down which exposure the hedge is designed to reduce.

4. Standardise the spread with a z-score

A z-score describes how far the current spread is from its historical mean:

Z-score = (current spread − mean spread) ÷ spread standard deviation

A reading of +2 means the spread is two historical standard deviations above its estimated mean. It does not mean convergence is 95% certain. Financial spreads can be skewed, fat-tailed and unstable.

Entry, exit and stop rules must be chosen before testing. For example, a research rule might enter beyond ±2, exit near zero and stop if the spread exceeds ±3 or if a fundamental break occurs. Those are hypothetical research thresholds, not universal trading rules.

Worked example

Assume a historical model estimates β = 0.5 for two hypothetical shares:

• Share A trades at £52.
• Share B trades at £96.
• Current spread = £52 − 0.5 × £96 = £4.
• Historical mean spread = £1.
• Historical spread standard deviation = £1.50.

The z-score is:

(£4 − £1) ÷ £1.50 = +2.0

Under a mean-reversion hypothesis, A appears expensive relative to B. To short one unit of the modelled spread, the hypothetical position shorts 100 A shares and buys 50 B shares:

• Short A notional: 100 × £52 = £5,200.
• Long B notional: 50 × £96 = £4,800.
• Gross exposure: £10,000.
• Net cash exposure: £400 short before financing.

If A falls to £50 and B rises to £98, the spread becomes £1. The short A leg gains £200 and the long B leg gains £100, for £300 gross profit before commission, bid-ask spread, slippage, borrow fees and dividends owed on the short.

If A instead rises to £55 and B falls to £94, the spread becomes £8. The short loses £300 and the long loses £100, for a £400 gross loss before costs. Both legs can lose at the same time.

The example shows why pairs trading is not risk-free arbitrage. The relationship may diverge further or never converge.

5. Define structural invalidation

A statistical stop is not enough. Define events that can break the economic relationship:

• one company changes guidance or capital structure;
• a merger, spin-off or takeover affects only one leg;
• one company gains or loses a major customer;
• regulation affects the firms differently;
• borrow becomes unavailable or extremely expensive;
• the hedge ratio shifts materially in rolling estimates.

When the reason for convergence disappears, a wider z-score is not automatically a better entry. It may be evidence that the model is obsolete.

6. Account for the short leg

The SEC's Regulation SHO investor guidance explains that short sellers generally borrow shares, pay loan interest and owe equivalent dividends to the lender. A short position can theoretically lose without limit as the share price rises.

Borrow can be recalled, the fee can rise, and a hard-to-borrow share may become impossible to short at the planned size. Corporate actions can create further obligations. Test the strategy with actual historical borrow and dividend assumptions where possible.

Read Short-Selling Risk before treating the short leg as a simple negative number.

7. Two legs double execution work

One leg may fill while the other does not. During a fast move, that leaves unintended directional exposure. Market orders increase fill probability but can create slippage; limit orders control price but may leave the pair incomplete.

Include commission, two bid-ask spreads, market impact, borrow costs, dividends, financing and exit costs. A small statistical edge can disappear after friction. The Spread and Slippage Guide explains how to model these costs.

8. Size the pair by spread risk

Gross exposure is more informative than net cash alone. A £50,000 long and £49,000 short can appear to have only £1,000 net exposure while carrying £99,000 gross exposure.

Estimate loss under:

• a wider spread;
• a broad market gap with unequal betas;
• a borrow recall;
• one-leg execution failure;
• a company-specific earnings shock;
• a historical correlation breakdown.

Review Correlation and Concentration Risk before combining several pairs that share the same sector or factor.

Common mistakes

1. Selecting pairs only because their charts look similar.
2. Treating high correlation as proof of mean reversion.
3. Optimising the hedge ratio and z-score thresholds on the full sample.
4. Ignoring structural breaks and changing fundamentals.
5. Assuming equal notional means market-neutral.
6. Excluding borrow fees, dividends and two-leg slippage.
7. Sizing from net exposure while ignoring gross exposure.
8. Adding repeatedly as the spread widens without a hard invalidation rule.

Practical checklist

• What economic relationship supports the pair?
• Which prices, currencies and corporate-action adjustments are used?
• Is the spread stable across multiple subperiods?
• How was the hedge ratio estimated?
• Is the target dollar, beta, regression or factor neutrality?
• What defines entry, convergence, time exit and maximum divergence?
• Which fundamental events invalidate the relationship?
• Is the short borrow available, and at what fee?
• Are both legs' spreads, slippage, dividends and financing included?
• Has the complete rule survived out-of-sample and stress testing?

The Money Man View

Pairs trading is a conditional convergence trade. The strongest process combines an economic reason, a stable tested spread, realistic short-borrow data and explicit structural invalidation. Z-scores organise the evidence; they do not force two assets to converge.

Educational-risk disclosure

This tutorial uses a hypothetical example and was prepared with AI-assisted research and writing under The Money Man editorial workflow. It is educational information, not personalised investment advice or a recommendation to trade. Short losses can be unlimited, both legs can lose, relationships can break and execution costs can exceed the expected edge.