The core idea behind pairs trading is simple. Find two assets that historically move together, wait for them to diverge, then go long the underperformer and short the outperformer, expecting them to converge back to their historical relationship. The profit comes from the convergence, regardless of whether both assets go up or both go down.
The statistical foundation is cointegration, which is different from correlation. Two assets can be highly correlated without being cointegrated, and vice versa. Cointegration means the spread between them is mean-reverting: it may wander away from its average, but it tends to return. The Engle-Granger test and the Johansen test are the standard statistical tests for cointegration.
In practice, finding cointegrated pairs requires screening a universe of related assets. Within crypto, pairs like BTC/ETH, different exchange tokens, or competing Layer 1 protocols can exhibit cointegration over certain periods. In equities, companies within the same sector (Coca-Cola and Pepsi, for example) are classic candidates. The key is that the assets share fundamental drivers that link their long-term performance.
Entry signals are typically generated when the spread (the difference or ratio between the two assets) moves a specified number of standard deviations from its mean. A common approach is to enter when the spread reaches 2 standard deviations and exit when it returns to the mean. Tighter thresholds (1.5 sigma) generate more trades but with lower average profit per trade. Wider thresholds (2.5 sigma) generate fewer but higher-conviction opportunities.
Risk management in pairs trading requires attention to the possibility that the cointegration relationship breaks down. This is called spread divergence, and it can happen when one company in the pair faces a company-specific event (earnings shock, regulatory action, acquisition) that permanently changes its relationship with the other. Stop losses on the spread level (exiting if the spread moves to 3-4 standard deviations) protect against this scenario.
Position sizing in pairs needs to account for the different volatilities of the two legs. If you go long $10,000 of a low-volatility stock and short $10,000 of a high-volatility stock, you are not market-neutral in risk terms. Beta-adjusting or volatility-adjusting the notional sizes of each leg creates a more balanced position where both sides contribute roughly equal risk.
Execution matters more in pairs trading than in directional trading because the edge is typically small on each trade. Transaction costs, slippage, and borrowing costs (for the short leg) can erode the spread's profit potential. High-frequency data and efficient execution are important, especially in markets with tighter spreads.
The advantage of pairs trading is that it can generate returns in any market environment: bull, bear, or sideways. The disadvantage is that the returns tend to be modest (low Sharpe per trade, compensated by high trade frequency) and the strategy is vulnerable to regime changes that break historical relationships. Diversifying across multiple pairs rather than concentrating in one or two helps smooth returns and reduces the impact of any single relationship breaking down.