What it is
A pairs trade buys one instrument and sells a related one, betting that the difference (or ratio) between them will revert to its usual level. Classic pairs are two stocks in the same industry; classic spreads are two contract months of the same futures (contract-month spreads) or two related commodities. The direction of the market is hedged out, more or less; what remains is a bet on the relationship. The trade is attractive because it can work in any market direction, and dangerous because relationships change and spreads can widen for far longer than a stretched price can.
The logic
Two instruments driven by the same economic factors move together most of the time. When one moves and the other does not, it is often a temporary supply-demand imbalance in one leg (a large order, a rebalance, an index event) rather than a change in the relationship, and it corrects as the market notices. In futures calendar spreads, storage costs, contango and backwardation set an economic range for the spread, and deviations from it attract physical arbitrageurs.
The other side is whoever is pushing one leg: a large fund exiting one stock, a producer hedging one contract month. If their reason is temporary, the spread reverts and the pairs trader profits. If their reason is a change in the relationship (a merger, a product failure, a supply shock), the spread does not revert and the pairs trader is on the wrong side of new information with a hedge that no longer hedges.
Setup rules
- Market: for beginners, index or sector ETF pairs and liquid futures calendar spreads, where margins are lower and the relationships are well-understood. Single-stock pairs need borrow availability and carry idiosyncratic risk.
- Timeframe: daily data; holds of days to weeks.
- Pair selection: a clear economic link (same sector, same commodity complex, same index family); a 2-year rolling correlation of daily returns above 0.7; a spread that has crossed its 60-day mean at least 6 times in the last year (evidence of reversion rather than a trend).
- Stretch condition: the spread's z-score (current value minus 60-day mean, divided by 60-day standard deviation) beyond plus or minus 2.
- Hedge ratio: dollar-neutral for ETF pairs (equal dollar amounts each side); beta-neutral if the two have different volatilities; one-to-one contracts for calendar spreads.
- Disqualifiers: a scheduled event on one leg (earnings, an index change, a delivery period) inside the holding window; a spread whose stretch is explained by news that changes the relationship.
Entry, stop, target
Enter both legs at once (or within minutes) when the z-score crosses the threshold. Target: z-score returns to zero (the spread's mean). Stop: z-score reaches plus or minus 3.5, or the trade has been open for 20 days without reverting, or the correlation breaks down. The stop is essential because a relationship that has changed will keep going.
| Item | Value | Notes |
|---|---|---|
| Pair | Sector ETF A vs sector ETF B | Correlation 0.82 |
| Spread | Log ratio A/B | 60-day mean 0.150, SD 0.020 |
| Entry | Ratio 0.192, z equals 2.1 | Short A $20,000, long B $20,000 |
| Target | Ratio 0.150, z equals 0 | Spread move 4.2 percent, about $840 |
| Stop | Ratio 0.220, z equals 3.5 | Spread move 2.8 percent against, about $560 |
| Approximate R:R | 1.5R | Plus borrow cost on the short leg |
The dollar figures are small relative to the $40,000 of gross exposure, which is the point: pairs trades are low-volatility positions that need size to matter, and that is where the risk creeps in.
Position sizing and risk
Size by the stop on the spread, not by the notional of each leg; the loss at the stop in the example is $560 on $40,000 gross. If your risk budget is 1 percent of a $50,000 account, that is $500, so the size is roughly right at $18,000 per leg. Use /tools/position-size with the stop expressed in spread percentage. The temptation is to leverage the gross exposure because the spread is "hedged"; the risk framework in /learn/risk-management treats the gross exposure as real, because in a correlation breakdown it is.
What breaks it
- Relationship breaks. The dominant failure. A pair that co-moved for years diverges permanently on new information, and the z-score goes to 5 and stays there. The stop at 3.5 and the time stop are the only protection.
- Short-leg risk. In stock pairs the short leg can be squeezed, recalled or become expensive to borrow.
- Costs. Two legs, two spreads, two commissions, borrow fees, and small targets; pairs trading is among the most cost-sensitive strategies retail traders attempt.
- Margin. Futures spreads receive margin relief, which encourages over-sizing; a calendar spread in a physical commodity can move violently around delivery (see first-notice-day).
- Edge decay. Statistical arbitrage in liquid pairs is dominated by firms with better data and lower costs; the classic stock-pairs edge documented in the 1990s has largely been arbitraged away in large caps. What remains is in less liquid pairs, where costs are higher, and in relationship-driven discretionary trades that require domain knowledge.
How to test it
Start with a universe of related ETFs or a single futures complex. For each candidate pair, compute the rolling correlation and the spread's z-score over 5 or more years, apply the entry, target and stop rules, and record every trade with both legs' costs and borrow. Report the win rate, expectancy per trade in spread percentage, the worst trade and the number of relationship breaks. Then run the same rules on pairs chosen at random (with no economic link); if random pairs test nearly as well, the "edge" is just mean reversion of noise and will not survive live. Use walk-forward-testing to choose the z-score threshold and lookback out of sample. Minimum 200 trades across pairs.
Variations
- Futures calendar spreads in a single commodity, driven by storage and seasonality; lower correlation risk, higher delivery-period risk.
- Sector-neutral single-stock pairs with beta hedging; the classic form, harder to execute.
- Ratio trades between related indices; see relative-strength-rotation for a trend-following version of the same relationship.
Further reading
correlation, mean-reversion, contango, backwardation, contract-month, first-notice-day, short-selling, margin, hedge, expectancy.
Related playbooks: range-mean-reversion-20ma, relative-strength-rotation, funding-rate-mean-reversion, walk-forward-testing