What you will learn
- Explain the pairs-trading idea and market neutrality
- Use a spread z-score to time entries
- Distinguish true arbitrage from statistical arbitrage
- Understand the risk that a relationship breaks down
Building on mean reversion, pairs trading is a market-neutral strategy that trades the relationship between two related assets rather than betting on the market's overall direction. It introduces the broader concept of arbitrage, the pursuit of profit from pricing discrepancies, and it illustrates both the appeal and the hidden dangers of strategies built on statistical relationships. It connects mean reversion, correlation, and the hedging ideas from across the curriculum.
The pairs-trading idea
Pairs trading identifies two assets whose prices have historically moved together, often two companies in the same industry, and trades the spread between them. When the prices diverge unusually far from their typical relationship, the strategy bets on convergence: it buys the relative underperformer and simultaneously sells short the relative outperformer, profiting when the spread narrows back toward its normal level. The trade is on the relationship between the two assets, not on whether either one rises or falls in absolute terms. This is mean reversion, from the previous lessons, applied to the spread between a pair.
Pairs trading and statistical arbitrage (Quantpedia)
Explains the pairs-trading idea and its statistical basis. Focus on trading the spread, not market direction.
Market neutrality
A key feature of pairs trading is that it is market-neutral. Because the strategy is long one asset and short another in roughly balanced amounts, the overall exposure to the broad market is largely cancelled out, connecting to the beta-neutral idea from Unit 6 and the hedging concepts of Unit 7. Whether the market as a whole rises or falls, the pairs trade aims to profit purely from the relative movement of the two assets converging. This insulation from market direction is the strategy's main appeal, since it can in principle make money in rising, falling, or flat markets, as long as the spread behaves as expected.
The statistical foundation
For pairs trading to work, the spread between the two assets must be mean-reverting and stationary, with a stable average it tends to return to, drawing on the stationarity concepts of Unit 5. The deeper statistical idea is that the two price series should be cointegrated, meaning that although each price wanders on its own, a particular combination of them stays stable over time. When the spread is genuinely stationary, deviations are temporary and reversion is likely, and when it is not, the foundation of the strategy collapses. Verifying that the relationship is real and stable, rather than a coincidence of history, is the central analytical challenge.
Timing entries with the spread z-score
Just as a single-asset mean-reversion trader watches the z-score of a price, a pairs trader watches the z-score of the spread. You track the spread's rolling mean and standard deviation, then compute how many standard deviations the current spread sits from its average. When the z-score is extreme, the spread has diverged unusually far, and the trader bets on convergence: buy the relative underperformer and short the relative outperformer, closing as the spread's z-score returns toward zero.
- spread = current price gap between the pair
- mean spread = its rolling average
- spread standard deviation = its rolling spread
Key terms
- Spread
- The price relationship between the two assets in the pair.
- Market-neutral
- Long one asset and short another, so overall market exposure largely cancels out.
- Cointegration
- Two wandering price series whose particular combination stays stable over time.
- Statistical arbitrage
- Trading on statistical relationships that have held historically. Not risk-free, despite the name.
Compute a spread z-score
Two paired stocks have a spread that is currently 8 dollars. Its rolling mean is 5 dollars and its rolling standard deviation is 1.5. What is the spread's z-score?
The math behind pairs trading (Quantopian)
Goes into the statistics of the spread and cointegration. Reinforces the z-score and stationarity ideas.
True arbitrage versus statistical arbitrage
- True arbitrage is a risk-free profit from a genuine pricing discrepancy, such as the same asset trading at different prices in two markets, or a violation of the put-call parity relationship from Unit 7. Pure arbitrage opportunities are rare, tiny, and fleeting, seized almost instantly by the fastest participants.
- Statistical arbitrage, the category that includes pairs trading, is not risk-free. It relies on statistical relationships that have held in the past and are expected to continue, but these relationships can and do break down, which means statistical arbitrage carries real risk despite the reassuring word arbitrage in its name.
The word arbitrage suggests a sure thing, but statistical arbitrage is a bet that a relationship holds. Relationships break, and the market can stay irrational longer than you can stay solvent.
The critical risk
The main danger of pairs trading and statistical arbitrage is that the historical relationship between the assets breaks down permanently. Two companies that moved together for years may diverge for good because of a fundamental change in one of them, a merger, a technological shift, a deteriorating business, so that the spread never reverts and the position keeps losing. The assumed cointegration fails, and what looked like a temporary divergence proves permanent. This risk is magnified by leverage and by crowding, since when many traders hold the same statistical-arbitrage positions, a forced unwinding can drive the spread further apart precisely when participants need it to converge. History offers sobering examples of sophisticated funds that suffered enormous losses when relationships they had relied upon broke down under stress, a reminder that the market can stay irrational, or simply change, longer than a trader can remain solvent. The honest caveat is that statistical arbitrage is not the free money its name implies. It is a bet that a statistical relationship will persist, and such bets demand the same rigorous risk management and humility as everything else in quantitative trading.
True arbitrage or a bet?
A trader calls their pairs strategy 'arbitrage' and uses heavy leverage, believing it is essentially risk-free. Why is this dangerous?
The word arbitrage is misleading here. True arbitrage exploits a genuine, risk-free pricing discrepancy, but statistical arbitrage merely bets that a historical relationship will persist. That relationship can break down permanently, and with heavy leverage the loss can be enormous, especially if crowding forces an unwinding that drives the spread further apart. Treating it as risk-free is how sophisticated funds have blown up.The name versus the reality
In your own words, explain why statistical arbitrage carries real risk despite the reassuring word arbitrage in its name.
Write an answer before comparing it with the model response.
Model answer
True arbitrage means a risk-free profit from a genuine pricing discrepancy, like the same asset trading at two different prices at once, and such opportunities are rare, tiny, and grabbed instantly. Statistical arbitrage, including pairs trading, is different: it profits only if a statistical relationship that held in the past continues to hold, such as a spread between two stocks reverting to its average. That relationship is not guaranteed. A fundamental change in one company can break the cointegration permanently, so the spread never reverts and the position keeps losing. The risk is worsened by leverage, which magnifies the loss, and by crowding, since if many traders hold the same position a forced unwinding can push the spread further apart just when they need it to converge. So despite the comforting name, statistical arbitrage is a bet that a relationship persists, and it demands the same rigorous risk management and humility as any other strategy.